ProgrammableMarketing/speko.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 7
  • This quarter · 10
  • Big bets · 3
  • Wildcards · 2
Generated Nov 10, 2026 2 views
  1. Programmable
  2. Playbooks
  3. Speko
Speko logo

growth playbook · Nov 10, 2026

speko.ai (opens in a new tab)

Speko growth playbook

Diagnosis

Speko is a four-person YC S26 company with an unusually large content engine for its age: about 150 public benchmark pages per model, 48 per-language pages, 47 blog posts and 27 product updates in 90 days, plus listings in the main AI assistant and coding-tool directories.

Its only open role says enterprise demand is already arriving with no one to sell to it. The gaps are at the hand-offs: the main site's sign-up buttons open a login form, the Enterprise plan is an email address, the SOC 2 report is mentioned only on the About page, and the benchmark pages that attract builders don't show how to use a model through Speko.

This playbook fixes those hand-offs first, then turns the benchmarks into a distribution and sales asset (code snippets on model pages, vendor badges, per-language stacks, private evals as the pilot) and gives the new Head of Growth a front door, a pilot template and proof to sell with.

Growth Score

7/10

Biggest Opportunity

The benchmarks already bring builders and vendors to Speko every day; connect each benchmark page to a working route, and each enterprise visitor to a pilot, and the content engine becomes the sales engine.

Biggest Risk

Vapi, Retell and Bland sell to the same teams with compliance badges and named customer results, and a router that relies on others' models can be copied by larger platforms, so Speko has to win on trusted measurement and fast proof before that happens.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Give enterprise inbound a front door: an Enterprise page with the SOC 2 report request, the pilot format and four qualifying questions

    Draft the /enterprise page from the pricing page's Enterprise column and the job post's pilot description.

    High impact First result in 2 weeks
  2. 2

    Point every "Get API key" button at the sign-up form, not the login form, and show the $10 credit there

    Replace the homepage hero "Get API key" link with /sign-up?utm_source=site&utm_campaign=hero.

    High impact First result in 1 week
  3. 3

    Add a sign-up button and a migration guide to each comparison page

    Add a tagged sign-up button to the bottom of the Speko vs Vapi page.

    Medium impact First result in 2 weeks

02

Your idea portfolio

22 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 8 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 3Big bets · 7Fill-ins · 7Reconsider · 5
Effort →

Showing 22 of 22 ideas

This week 7

Quick wins with a clear first action.

  • 01

    On speko.ai, all four homepage buttons that should start an account ("Get API key" twice, "Build your own", "Start building") go to platform.speko.ai/sign-in, a login form with email, password and "Forgot password?", where a new developer has to find "Don't have an account? Sign up" at the bottom. The pricing, integration and solution pages do the same. The benchmark pages already do it right: their "Start building →" and "Try →" buttons go to /sign-up with tracking tags. Neither the homepage nor the sign-up form mentions the $10 credit that pricing promises, so the strongest reason to sign up today is missing at the moment of decision.

    High impactLow effort First result in 1 week Founding engineer

    First action

    Replace the homepage hero "Get API key" link with /sign-up?utm_source=site&utm_campaign=hero.

    How to measure it

    Main-site sign-ups per 100 button clicks

    Now
    –
    Target
    +25%
    By
    30 days

    Steps

    1. 1

      List the buttons

      Every link to platform.speko.ai/sign-in on the homepage, pricing, integrations, solutions, startups and comparison pages.

      Founding engineer (web) 1 hour

    2. 2

      Repoint to sign-up

      Change account-starting buttons to /sign-up; keep one "Sign in" link in the header.

      Founding engineer (web) 2 hours

    3. 3

      Tag the links

      Copy the benchmark UTM scheme: source, campaign by page, content by position.

      Founding engineer (web) 1 hour

    4. 4

      Show the credit

      One line on the homepage hero and above the sign-up form: $10 in credit on signup.

      Founder/CEO 1 hour

    5. 5

      Carry the source

      Store the UTM on the new workspace so first payment can be traced to the page.

      Founding engineer (platform) 1 day

    6. 6

      Compare

      Click-to-account and account-to-first-call rates, main site versus benchmark pages.

      Founding Head of Growth weekly

    7. 7

      Check monthly

      Add a link check so new pages can't ship pointing at /sign-in.

      Founding engineer (web) monthly

    First result
    1 week
    Cost
    $0, a few hours of engineering
    Owner
    Founding engineer (web)

    Smallest test

    Repoint the homepage hero button today and compare its sign-ups for two weeks against the old rate.

    Why it works

    Change every main-site account button to platform.speko.ai/sign-up, add the same UTM scheme the benchmarks use (utm_source=site, campaign by page and position), and put one line above the form: "$10 in credit on signup, no card needed" (or whatever the true terms are). Keep "Sign in" as a small link in the header for returning users. Then compare sign-up completion from the main site against the benchmark pages, which already land on the right form.

    A visitor who clicks "Get API key" has decided to try the product; landing on a login form reads as "you need an account first" and costs a step at the point of highest intent. Showing the credit on the form removes the last doubt about cost. The fix is a link change, and the benchmark pages give a ready-made control group.

    Ready to use

    ChecklistSign-up button fix list

    [ ] Homepage: hero "Get API key" -> /sign-up?utm_source=site&utm_campaign=hero
    [ ] Homepage: "Build your own", "Start building", bottom "Get API key" -> /sign-up (tagged)
    [ ] /pricing: "Get API key" on Router and Speko infra -> /sign-up
    [ ] /integrations/*: "Get started free" -> /sign-up
    [ ] /solutions/*: "Build your own", "Get started free" -> /sign-up
    [ ] Header keeps a small "Sign in" link
    [ ] "$10 in credit on signup" on the hero and above the form
    [ ] UTM saved on the new workspace
    [ ] Monthly link check for /sign-in on marketing pages

    Risks and how to handle them

    • Returning users click "Get API key" to log in

      Keep "Already have an account? Sign in" on the sign-up form, which it already has.

    • Credit terms change

      Pull the credit amount from the same source as the pricing page.

    Evidence

    • Your websiteHomepage (checked November 9, 2026): H1 "The Router for Voice AI" and subhead "Every speech model, benchmarked language by language, wired into one API."; all four sign-up buttons ("Get API key" twice, "Build your own", "Start building") link to platform.speko.ai/sign-in with no tracking tags; the $10 signup credit is not mentioned; there are no customer logos or testimonials.Source
    • Your websiteConsole pages: /sign-in is a login form (email, password, "Forgot password?", "Sign in") with "Don't have an account? Sign up" at the bottom; /sign-up says "Create your account". Neither mentions the $10 credit. Both show a testimonial from an insurance agency founder, the only customer quote found on any Speko page checked.Source
    • Your websiteThe benchmark pages' "Start building →" and per-model "Try →" buttons go to platform.speko.ai/sign-up, tagged utm_source=benchmarks with campaign (nav, footer, row_inline) and model tags.Source
    • Your websitePricing (updated November 3, 2026): Router at the provider's rate "plus 5%"; Speko infra at "$0.09 per minute" across speech-to-text, LLM and text-to-speech; both in public preview with no SLA during preview; Enterprise is custom pricing with a committed monthly minimum, a contract SLA and deployment reviews, reached by emailing [email protected]; a $10 signup credit; billing per connected second, with failed calls refunded automatically; numbers at "$1 setup and $2/month"; carrier charge plus 20%. A table lists Vapi at $0.05/min hosting, Retell AI at $0.07 to $0.31/min and Bland AI at $0.14/min. Speko's own security or compliance status is not mentioned.Source
    • Your websiteSolutions: four pages (HVAC and plumbing dispatch, law firm intake, moving company quotes, towing and roadside dispatch). The HVAC page, "AI dispatcher for HVAC and plumbing companies", has a live demo ("Press the sphere to call"), a sample transcript marked "Written for this page. It is not a recording.", a price "from $0.30" for a three-minute inbound call, and "Build your own" and "Get started free" buttons to sign-in; there is no customer proof and no ready-made template.Source
  • 02

    Speko's only open role says "Enterprise demand already comes to us, and we have no seller yet", naming a contact-center platform, a white-label reseller and a BPO consultant. On the site, the Enterprise plan's only action is an email to the same [email protected] inbox that handles support and billing, the SOC 2 Type 2 report is mentioned only on the About page, there is no security or trust page, and the demo booking asks nothing about company or volume. Vapi, Retell and Bland all put compliance on their homepages.

    High impactLow effort First result in 2 weeks Founding Head of Growth

    First action

    Draft the /enterprise page from the pricing page's Enterprise column and the job post's pilot description.

    How to measure it

    Enterprise form submissions with more than 50,000 monthly minutes

    Now
    –
    Target
    8 a month
    By
    90 days

    Steps

    1. 1

      Write the page

      Enterprise terms, pilot format, security summary, one form.

      Founding Head of Growth 2 days

    2. 2

      Add the report request

      Gated SOC 2 Type 2 request with an NDA step.

      Founder/CEO 1 day

    3. 3

      Build the form

      Company type, minutes, languages, current stack.

      Founding engineer (web) 1 day

    4. 4

      Route it

      Form answers to a pipeline tool, not [email protected].

      Founding Head of Growth 1 day

    5. 5

      Link it

      From pricing, the comparison pages, /book and the header.

      Founding engineer (web) 2 hours

    6. 6

      Review weekly

      Which answers turn into pilots.

      Founding Head of Growth weekly

    First result
    2 weeks
    Cost
    Staff time; a pipeline tool if none exists
    Owner
    Founding Head of Growth

    Smallest test

    Add the four questions to the enterprise email reply and see how many answer.

    Why it works

    Publish /enterprise: what Enterprise includes (committed pricing, contract SLA, deployment reviews), how a pilot works (written acceptance tests, an engineer on every pilot, Speko's benchmark data on the buyer's languages), a "Request the SOC 2 Type 2 report" button, and a short form with four questions: company type, monthly call minutes, languages, and current stack. Route answers to the Head of Growth's pipeline, not the shared inbox. Link it from pricing, the comparison pages and the header.

    Enterprise buyers screen vendors on security and process before they book a call; showing the report and the pilot method up front moves them from "email and wait" to a qualified meeting, and gives the new seller a pipeline on day one.

    Risks and how to handle them

    • Seller not yet hired

      The founder owns the queue until the Head of Growth starts.

    • Overstating compliance

      State only what the SOC 2 report covers; have the auditor's scope checked.

    Evidence

    • HiringFounding Head of Growth (San Francisco, onsite; $170K to $190K base, 1.25% to 1.75% equity): "Enterprise demand already comes to us, and we have no seller yet"; recent inbound includes "a contact-center platform, a white-label reseller, and a BPO consultant"; "There is no sales playbook yet, and we want you to write it"; the hire runs pilots with written pass/fail acceptance tests through security reviews, procurement and contracts, owns the pipeline in month 1 and has signed pilots by month 3, with "an engineer on every pilot and our benchmark data for every deal".Source
    • Your websiteAbout page: "Compliance | SOC 2 Type 2 report available"; one address, [email protected], for "Product questions, support requests, security reports, billing questions, and partnership inquiries"; "89 speech and language models across 18 languages" (board snapshot 2026-11-09). Neither sitemap lists a security or trust page, and speko.ai/security returns 404.Source
    • Your websiteDemo page: "Book a 25 minute demo with the founder" through a Google appointment schedule; the agenda is the current voice stack, "A live routed call on Speko using your language and provider mix" and pricing; it asks nothing about company, use case or call volume and does not mention enterprise, security reviews or pilots.Source
    • Competitor researchRivals' own homepages: Vapi, "Voice agents for builders", "1M+ developers", "SOC 2, HIPAA, and PCI compliant"; Retell AI, "#1 AI Voice Agent Platform for Automating PHONE Calls", with HIPAA, SOC 2 Type II, GDPR and ISO 27001 claims and named customer results; Bland, "Voice AI for the calls that can't go wrong", over 230 regulated companies, SOC 2 Type I and II and HIPAA BAAs; Artificial Analysis benchmarks speech-to-text, text-to-speech and speech-to-speech models with a public voice-voting feature.Source
  • 03

    Speko already has comparison pages against Vapi, Retell and Bland, a Vapi alternatives page and a voice agent platforms page, reviewed on November 3. The Vapi page is careful and sourced, but it ends with links to more reading: no sign-up button, no customer proof and no help moving an existing agent. Readers of a comparison page are usually mid-switch.

    Medium impactLow effort First result in 2 weeks Founding Head of Growth

    First action

    Add a tagged sign-up button to the bottom of the Speko vs Vapi page.

    How to measure it

    Comparison-page sign-ups

    Now
    –
    Target
    40 a month
    By
    60 days

    Steps

    1. 1

      Add the button

      Tagged sign-up with the credit line on all five pages.

      Founding engineer (web) 2 hours

    2. 2

      Map the fields

      Rival agent settings to Speko agent settings.

      Founding engineer (platform) 2 days

    3. 3

      Write the guides

      One migration guide per rival, with a sample agent.

      Founding Head of Growth 1 week

    4. 4

      Link enterprise

      "Moving more than 50,000 minutes?" to the Enterprise page.

      Founding engineer (web) 1 hour

    5. 5

      Fact check

      Re-check rival pricing and claims before each review date.

      Founding Head of Growth quarterly

    6. 6

      Measure

      Sign-ups and first calls per page.

      Founding Head of Growth monthly

    First result
    2 weeks
    Cost
    Staff time only
    Owner
    Founding Head of Growth

    Smallest test

    Add the button to the Vapi page only and count sign-ups for two weeks.

    Why it works

    At the end of each comparison page add a tagged sign-up button with the $10 credit, a "Move a Vapi agent to Speko" (or Retell, Bland) guide with the field mapping and a working example, and a route to the Enterprise page for teams above a minute threshold. Track sign-ups per comparison page.

    Comparison readers have high intent and a concrete task; a guide that removes the switching work turns the comparison into an action.

    Risks and how to handle them

    • Rival features change

      Keep the review date on each page and recheck sources quarterly.

    Builds on

    Evidence

    • Your websiteSpeko vs Vapi (published August 7, last reviewed November 3, 2026): "Choose the layer first. Compare products inside that boundary second."; a table on scope, telephony, routing, observability and benchmarks, and pricing side by side; no customer proof and no sign-up button.Source
    • Your websiteSitemaps (checked November 9, 2026): the main sitemap has 38 URLs, including comparison pages against Bland, Retell and Vapi, "vapi-alternatives", four solution pages, seven integration pages and /startups. The benchmarks sitemap has 280 URLs: about 150 model pages (STT, TTS, LLM, speech-to-speech, turn-taking), 10 stack pages, 48 per-language pages (17 STT, 16 TTS, 15 LLM), five "best ... 2026" guides, /compare/stt, /compare/tts, a glossary and 47 blog posts. Neither has case studies or customer stories.Source
  • 04

    Speko sells itself as "the first proxyless, zero-latency model router", and its buyers read latency charts for a living. Google's field data shows real visitors load the homepage fine (LCP 2.0s) but wait 215ms for the page to respond to a tap or click, so the homepage fails Core Web Vitals on INP. In the lab, mobile Total Blocking Time is 9.4 seconds, with 434 KB of JavaScript and 3,419 page elements from the interactive demos and benchmark tables.

    Medium impactMedium effort First result in 1 month Founding engineer

    First action

    Find which homepage component causes the longest task in a Chrome performance trace.

    How to measure it

    INP, 75th percentile

    Now
    –
    Target
    under 200ms
    By
    60 days

    Steps

    1. 1

      Trace

      Find the longest main-thread tasks on mobile.

      Founding engineer (web) half a day

    2. 2

      Defer demos

      Load the demo, screenshots and charts on scroll.

      Founding engineer (web) 2 days

    3. 3

      Slim the table

      Static top rows; full table on /benchmarks.

      Founding engineer (web) 1 day

    4. 4

      Cut JavaScript

      Unused and legacy bundles Google lists.

      Founding engineer (web) 2 days

    5. 5

      Re-test

      PageSpeed lab and field data.

      Founding engineer (web) weekly

    6. 6

      Guard it

      A performance budget in the build.

      Founding engineer (web) 1 day

    First result
    1 month (field data updates over 28 days)
    Cost
    About a week of engineering
    Owner
    Founding engineer (web)

    Smallest test

    Defer the benchmark chart alone and compare Total Blocking Time.

    Why it works

    Load the console screenshots, the live demo sphere and the benchmark chart only when they scroll into view; render the long benchmark table as static HTML with a "see all" link to /benchmarks; remove the 78 KB of unused and 33 KB of legacy JavaScript Google flags. Re-test until real-visitor INP is under 200ms.

    Technical buyers notice a sluggish page, and for a latency company it undercuts the main claim. A lighter homepage also helps mobile visitors coming from social and the directories.

    Risks and how to handle them

    • Losing the live demo's impact

      Keep it, but start it on tap rather than on load.

    Evidence

    • PageSpeedHomepage, tested November 2026: mobile performance score 46 (desktop 68); on mobile, lab LCP 3.8s and Total Blocking Time 9.4s, with 434 KB of JavaScript and 3,419 page elements. Real visitors: LCP 2.0s (good), but interaction delay (INP) is 215ms, so the homepage fails Core Web Vitals on INP. Largest fixes: unused JavaScript (78 KB) and legacy JavaScript (33 KB).Source
    • Your websiteHomepage (checked November 9, 2026): H1 "The Router for Voice AI" and subhead "Every speech model, benchmarked language by language, wired into one API."; all four sign-up buttons ("Get API key" twice, "Build your own", "Start building") link to platform.speko.ai/sign-in with no tracking tags; the $10 signup credit is not mentioned; there are no customer logos or testimonials.Source
  • 05

    Speko's credibility rests on measurement, yet its own pages disagree on the basic number: the About page says 89 models, the Models page 115, the Startup Program 31 providers, and the homepage "15 of 58" measured only in English, while the benchmark sitemap has about 150 model pages. The Models page front matter says it was last updated in August though its entries run into November.

    Medium impactLow effort First result in 1 week Benchmarks engineer

    First action

    Decide the four counts and where each comes from in data.json.

    How to measure it

    Mismatched headline counts

    Now
    –
    Target
    0
    By
    30 days

    Steps

    1. 1

      Define counts

      Measured, routable, providers, languages.

      Benchmarks engineer half a day

    2. 2

      Generate them

      From data.json at build time.

      Benchmarks engineer 1 day

    3. 3

      Replace text

      Homepage, About, Models, Startups, meta descriptions.

      Founding engineer (web) half a day

    4. 4

      Date them

      "As of" date next to each count.

      Founding engineer (web) 1 hour

    5. 5

      Fix front matter

      Last-updated dates generated, not typed.

      Founding engineer (web) 1 hour

    6. 6

      Check

      Search the site for old numbers monthly.

      Benchmarks engineer monthly

    First result
    1 week
    Cost
    $0
    Owner
    Benchmarks engineer

    Smallest test

    Fix the About and Models pages to match today.

    Why it works

    Generate the headline counts (models measured, models routable, providers, languages) from the benchmark data file at build time, show them with a date on the homepage, About, Models and Startup pages, and explain the difference between "measured" and "routable" in one line.

    A benchmark company is judged on consistency; skeptical developers (the Launch HN thread had plenty) notice mismatched numbers, and assistants that summarize the site repeat whichever one they find.

    Risks and how to handle them

    • Counts move often

      Generated numbers update themselves.

    Evidence

    • Your websiteModels page: 115 models (36 STT, 28 LLM, 40 TTS, 11 speech-to-speech) with front matter last updated 2026-08-24, while the About page says 89 models, the Startup Program page 31 providers and the homepage "15 of 58" models measured only in English.Source
    • Your websiteSitemaps (checked November 9, 2026): the main sitemap has 38 URLs, including comparison pages against Bland, Retell and Vapi, "vapi-alternatives", four solution pages, seven integration pages and /startups. The benchmarks sitemap has 280 URLs: about 150 model pages (STT, TTS, LLM, speech-to-speech, turn-taking), 10 stack pages, 48 per-language pages (17 STT, 16 TTS, 15 LLM), five "best ... 2026" guides, /compare/stt, /compare/tts, a glossary and 47 blog posts. Neither has case studies or customer stories.Source
    • Launch HN"Launch HN: Speko (YC S26) – OpenRouter for Voice AI" (118 points, August 2026). Commenters valued the per-language benchmarks; asked whether latency is measured on the transport teams actually ship (one saw about 0.9s streaming against 5s non-streaming for a TTS model), whether cost reflects caching and context length, and how naturalness ratings are gathered; one asked what happens "when OpenRouter just adds voice", another suggested an "LM Arena of voice models" framing. The founder said human blind listening rounds are still used for naturalness.Source
  • 06

    The only customer testimonial on any Speko page checked sits on the console sign-in and sign-up pages, from an insurance agency founder, where only people who already clicked through see it. The homepage, pricing, solution and comparison pages have no logos, quotes or case studies, while Retell and Bland lead with named customer results.

    High impactMedium effort First result in 2 weeks Founder/CEO

    First action

    Ask the quoted customer for permission to use the quote on the public site.

    How to measure it

    Case studies published

    Now
    –
    Target
    3
    By
    60 days

    Steps

    1. 1

      Get permission

      For the existing quote.

      Founder/CEO 2 days

    2. 2

      Place the quote

      Homepage and pricing.

      Founding engineer (web) 1 hour

    3. 3

      Pick three users

      Phone agent, Router/Gateway, startup.

      Founder/CEO 1 week

    4. 4

      Pull the numbers

      Minutes, cost per call, latency from the console.

      Founding engineer (platform) 2 days

    5. 5

      Write and approve

      One page each, customer-approved.

      Founding Head of Growth 3 weeks

    6. 6

      Distribute

      Solution, comparison and Enterprise pages.

      Founding engineer (web) 1 day

    First result
    2 weeks
    Cost
    Staff time; optional credits as a thank-you
    Owner
    Founder/CEO

    Smallest test

    Put the existing quote on the homepage this week.

    Why it works

    With the customer's permission, put that quote on the homepage and the pricing page now. Then write three short case studies (problem, stack chosen and why, minutes, cost per call, latency) from current users: one phone agent, one Router or Gateway team, and one from the startup batch. Add them to the solution and comparison pages.

    Developers trust benchmarks, but buyers also want to see someone like them in production; proof is the biggest gap against rivals that otherwise look similar.

    Risks and how to handle them

    • Customers won't be named

      Publish anonymized by industry with real numbers.

    Evidence

    • Your websiteConsole pages: /sign-in is a login form (email, password, "Forgot password?", "Sign in") with "Don't have an account? Sign up" at the bottom; /sign-up says "Create your account". Neither mentions the $10 credit. Both show a testimonial from an insurance agency founder, the only customer quote found on any Speko page checked.Source
    • Your websiteHomepage (checked November 9, 2026): H1 "The Router for Voice AI" and subhead "Every speech model, benchmarked language by language, wired into one API."; all four sign-up buttons ("Get API key" twice, "Build your own", "Start building") link to platform.speko.ai/sign-in with no tracking tags; the $10 signup credit is not mentioned; there are no customer logos or testimonials.Source
    • Competitor researchRivals' own homepages: Vapi, "Voice agents for builders", "1M+ developers", "SOC 2, HIPAA, and PCI compliant"; Retell AI, "#1 AI Voice Agent Platform for Automating PHONE Calls", with HIPAA, SOC 2 Type II, GDPR and ISO 27001 claims and named customer results; Bland, "Voice AI for the calls that can't go wrong", over 230 regulated companies, SOC 2 Type I and II and HIPAA BAAs; Artificial Analysis benchmarks speech-to-text, text-to-speech and speech-to-speech models with a public voice-voting feature.Source
  • 07

    The Head of Growth post says developers find Speko through the Claude connector directory (trending there in September 2026), the Cursor marketplace and the ChatGPT app directory, and the site has seven integration pages for these tools. But the integration pages send people to the untagged sign-in page, and connector installs arrive by OAuth, so the team can't yet compare which directory brings developers who make paid calls.

    Medium impactLow effort First result in 1 month Founding engineer

    First action

    Add a "source client" field to new workspaces.

    How to measure it

    Workspaces with source attribution

    Now
    –
    Target
    90%
    By
    30 days

    Steps

    1. 1

      Capture the client

      From OAuth client ID or MCP handshake.

      Founding engineer (platform) 1 day

    2. 2

      Tag the pages

      Integration-page buttons by tool.

      Founding engineer (web) 2 hours

    3. 3

      Define activation

      First call and first paid minute.

      Founder/CEO 1 hour

    4. 4

      Build the table

      Weekly cohort by source.

      Founding engineer (platform) 1 day

    5. 5

      Decide

      Double down on the best two sources.

      Founding Head of Growth monthly

    6. 6

      Improve listings

      Rewrite the top directory listing with what converts.

      Founding Head of Growth 1 week

    First result
    1 month
    Cost
    $0
    Owner
    Founding engineer (platform)

    Smallest test

    Tag the Claude integration page button and count sign-ups for two weeks.

    Why it works

    Record the connecting client (from the OAuth client or MCP handshake) on each new workspace, tag the integration-page buttons by tool, and build one weekly table: new workspaces, first call, first paid minute and 30-day spend by source. Put more effort into the directory that wins.

    With a team of four, choosing the right two channels matters more than adding a seventh integration.

    Risks and how to handle them

    • Users install in several tools

      Keep first-touch and last-touch both.

    Builds on

    Evidence

    • HiringThe same post says developers find Speko through the Claude connector directory (trending there in September 2026), the Cursor marketplace and the ChatGPT app directory; Speko is backed by Y Combinator (S26) and Heavybit, has a team of four, and its engineers work from Central Asia and Jakarta.Source
    • Your websiteClaude integration page: the connector places one outbound call per request, reads transcripts and searches the docs, but cannot create or deploy agents (those are built in Claude Code, the console or the API); a two-field business declaration is filed before the first outbound call, and every call starts with an AI disclosure that no prompt can turn off; the MCP server is free and new accounts get $10 in credit.Source
    • Your websiteHomepage (checked November 9, 2026): H1 "The Router for Voice AI" and subhead "Every speech model, benchmarked language by language, wired into one API."; all four sign-up buttons ("Get API key" twice, "Build your own", "Start building") link to platform.speko.ai/sign-in with no tracking tags; the $10 signup credit is not mentioned; there are no customer logos or testimonials.Source

This quarter 10

The core plays for the next three months.

  • 08

    Speko has about 150 model pages, each with measured WER, latency and the vendor's list price, and a "Run [model] on Speko →" button that goes to sign-up. None shows how to call that model through Speko (the model string, a three-line request, the Gateway option) or what it costs on Speko. A builder researching one model has to sign up before seeing whether using it is one line or a migration.

    High impactMedium effort First result in 1 month Benchmarks engineer

    First action

    Add the box to the ten most-visited model pages.

    How to measure it

    Model-page visitors who make a first API call within 7 days

    Now
    –
    Target
    2x
    By
    90 days

    Steps

    1. 1

      Rank pages

      Top model pages by visits.

      Benchmarks engineer half a day

    2. 2

      Design the box

      Model ID, code tabs, Speko price, button.

      Founding engineer (web) 1 day

    3. 3

      Generate snippets

      From the Router's OpenAPI and the model catalog.

      Benchmarks engineer 3 days

    4. 4

      Compute prices

      Provider rate plus 5%, from pricing data.

      Benchmarks engineer 1 day

    5. 5

      Preselect in console

      Pass the model to sign-up and open it in the playground.

      Founding engineer (platform) 3 days

    6. 6

      Roll out

      All model pages once the top ten work.

      Founding engineer (web) 1 week

    7. 7

      Measure

      First calls by model page.

      Founding Head of Growth weekly

    First result
    1 month
    Cost
    About two weeks of engineering
    Owner
    Benchmarks engineer

    Smallest test

    Add the box to one popular STT model page and compare first calls for a month.

    Why it works

    Add a "Use this model" box to each model page: the model ID for the Router, a short Python and TypeScript example, the Gateway/BYOK variant for LiveKit and Pipecat, and the price on Speko (provider rate plus 5%, or the $0.09 minute where it applies). Keep the sign-up button under it, passing the model so the console opens with it preselected.

    The model pages capture high-intent search traffic for specific model names; showing the exact code turns research into a first call and makes the router's value (one line to switch) visible.

    Risks and how to handle them

    • Benchmarks look less independent next to a buy button

      Keep the box below the results and show every model, winners and losers, the same way.

    • Model not routable today

      Show "measured, not yet routable" with a request button.

    Builds on

    Evidence

    • Your websiteModel page example, Nova-3: last measured 2026-10-01; WER, finalize latency, time to first token and the vendor's price per minute; a "Run Nova-3 on Speko →" button to sign-up; share options ("Download card (PNG)", "Post on X") and rank-badge embed code. There is no code snippet showing how to call the model through Speko and no Speko price for routing it.Source
    • Your websiteSitemaps (checked November 9, 2026): the main sitemap has 38 URLs, including comparison pages against Bland, Retell and Vapi, "vapi-alternatives", four solution pages, seven integration pages and /startups. The benchmarks sitemap has 280 URLs: about 150 model pages (STT, TTS, LLM, speech-to-speech, turn-taking), 10 stack pages, 48 per-language pages (17 STT, 16 TTS, 15 LLM), five "best ... 2026" guides, /compare/stt, /compare/tts, a glossary and 47 blog posts. Neither has case studies or customer stories.Source
    • Your websitePricing (updated November 3, 2026): Router at the provider's rate "plus 5%"; Speko infra at "$0.09 per minute" across speech-to-text, LLM and text-to-speech; both in public preview with no SLA during preview; Enterprise is custom pricing with a committed monthly minimum, a contract SLA and deployment reviews, reached by emailing [email protected]; a $10 signup credit; billing per connected second, with failed calls refunded automatically; numbers at "$1 setup and $2/month"; carrier charge plus 20%. A table lists Vapi at $0.05/min hosting, Retell AI at $0.07 to $0.31/min and Bland AI at $0.14/min. Speko's own security or compliance status is not mentioned.Source
  • 09

    Speko already makes "Embeddable badges for vendors ranked on our boards" and every model page has rank-badge embed code and a shareable PNG card. But the badges page lists only two vendors and gives no embed instructions or terms. Speech vendors love to cite wins; each badge on a vendor's site or launch post is a link and an audience pointing at Speko's board.

    High impactLow effort First result in 1 month Founding Head of Growth

    First action

    List this month's top-three vendors per board and language.

    How to measure it

    Vendor sites embedding a badge

    Now
    –
    Target
    15
    By
    1 quarter

    Steps

    1. 1

      Write terms

      Live link, date shown, removal on rank change.

      Founding Head of Growth 1 day

    2. 2

      Add instructions

      Embed code and terms on the badges page.

      Benchmarks engineer 1 day

    3. 3

      Auto-generate

      Badges per category and language from data.json.

      Benchmarks engineer 3 days

    4. 4

      Email vendors

      Top three per board after each update.

      Founding Head of Growth monthly

    5. 5

      List users

      Showcase vendors using badges.

      Founding engineer (web) ongoing

    6. 6

      Track links

      Referring domains and visits.

      Founding Head of Growth monthly

    First result
    1 month
    Cost
    $0
    Owner
    Founding Head of Growth

    Smallest test

    Email the five top STT vendors their badge this week.

    Why it works

    Each time the boards update, email every vendor whose model ranks top three in a category or language a ready-to-use badge, the share card and a one-paragraph quote they can use, with clear terms: link to the live board, show the date, remove if the rank changes. List every vendor using a badge on the badges page.

    It recruits the vendors themselves as distributors; their launch posts and docs reach exactly the developers Speko wants, and the link back strengthens the board in search.

    Risks and how to handle them

    • Looks like pay-to-rank

      Badges are free, automatic and tied to the published method.

    • Stale badges

      Badges render the current rank from the live board.

    Evidence

    • Your websiteBadges page: "Embeddable badges for vendors ranked on our boards"; two vendor badges are listed, with no embed instructions, terms or link-back on the page.Source
    • Your websiteModel page example, Nova-3: last measured 2026-10-01; WER, finalize latency, time to first token and the vendor's price per minute; a "Run Nova-3 on Speko →" button to sign-up; share options ("Download card (PNG)", "Post on X") and rank-badge embed code. There is no code snippet showing how to call the model through Speko and no Speko price for routing it.Source
  • 10

    The Head of Growth will run pilots "with written acceptance tests", priced and scoped before kickoff, and the job post calls pilot discipline "the part we cannot teach". Enterprise buyers rarely know how to write pass/fail criteria for a voice agent. Speko already measures the right things publicly: WER by language, end-of-turn latency and cost per solved call.

    Medium impactLow effort First result in 1 month Founding Head of Growth

    First action

    Write the template from the last two enterprise conversations.

    How to measure it

    Template downloads that become scoped pilots

    Now
    –
    Target
    25%
    By
    1 quarter

    Steps

    1. 1

      Draft

      Criteria, thresholds, test set, timeline.

      Founding Head of Growth 2 days

    2. 2

      Add reference values

      From the public boards per language.

      Benchmarks engineer 1 day

    3. 3

      Review with engineers

      What can be measured on every pilot.

      Founder/CEO 1 day

    4. 4

      Publish

      On the Enterprise page, editable copy.

      Founding engineer (web) half a day

    5. 5

      Use in calls

      Fill it in during discovery.

      Founding Head of Growth ongoing

    6. 6

      Refine

      Update after each pilot.

      Founding Head of Growth monthly

    First result
    1 month
    Cost
    $0
    Owner
    Founding Head of Growth

    Smallest test

    Send the draft to the next enterprise inbound before the call.

    Why it works

    Publish a downloadable pilot template on the Enterprise page: languages, call types, target WER, latency at p90, cost per solved call, transfer rate, test set size and duration, with Speko's public numbers as reference values. Offer to fill it in together on the first call.

    It qualifies serious buyers, shortens the scoping cycle, and makes Speko's benchmark method the yardstick the buyer uses for every vendor.

    Risks and how to handle them

    • Rivals copy it

      Speko's public reference numbers are the part they can't copy.

    Builds on

    Evidence

    • HiringFounding Head of Growth (San Francisco, onsite; $170K to $190K base, 1.25% to 1.75% equity): "Enterprise demand already comes to us, and we have no seller yet"; recent inbound includes "a contact-center platform, a white-label reseller, and a BPO consultant"; "There is no sales playbook yet, and we want you to write it"; the hire runs pilots with written pass/fail acceptance tests through security reviews, procurement and contracts, owns the pipeline in month 1 and has signed pilots by month 3, with "an engineer on every pilot and our benchmark data for every deal".Source
    • Your website"True cost per solve" (July 5, 2026): the cheapest stack that actually booked the appointment cost $0.058 per booking against $0.703 for an all-premium stack, about 12 times more for the same outcome; two low-cost stacks were flagged for claiming success without booking. The blog offers no newsletter or RSS feed.Source
  • 11

    The Startup Program offers $1,000 in credits (about 185 hours of voice), a showcase profile, features on X and LinkedIn and a badge, and welcomes idea-stage teams without an account. But it gives no deadline or selection date, so there is no reason to apply now, and the showcase it promises has nobody in it yet.

    Medium impactLow effort First result in 6 weeks Founding Head of Growth

    First action

    Pick the Batch 01 close date and put it on the page.

    How to measure it

    Batch 01 teams making paid calls after credits

    Now
    –
    Target
    10
    By
    1 quarter

    Steps

    1. 1

      Set dates

      Close, decision and showcase dates.

      Founder/CEO 1 hour

    2. 2

      Update the page

      Dates and selection criteria.

      Founding engineer (web) 1 hour

    3. 3

      Promote

      Discord, X, LinkedIn, accelerator communities.

      Founding Head of Growth 3 weeks

    4. 4

      Select

      Score on readiness to ship calls.

      Founder/CEO 1 week

    5. 5

      Showcase

      Profiles of accepted teams.

      Founding engineer (web) 1 week

    6. 6

      Follow up

      30-day build logs; pick case studies.

      Founding Head of Growth monthly

    First result
    6 weeks
    Cost
    Credits given ($1,000 per team, mostly variable cost)
    Owner
    Founding Head of Growth

    Smallest test

    Add a close date to the page and compare weekly applications.

    Why it works

    Set a close date and a decision date for Batch 01, announce them on the page, in the Discord and to YC and accelerator communities, and publish accepted teams with what they are building. Require a short build log after 30 days to keep the credits, and turn the best into case studies.

    Deadlines create applications; a visible cohort shows momentum to developers and gives Speko a stream of early production users and stories.

    Risks and how to handle them

    • Credits used and abandoned

      Release credits in two halves, the second after a build log.

    Evidence

    • Your websiteStartup Program (Batch 01): $1,000 in credits (about 185 hours of voice at $0.09 a minute), a showcase profile, features on X and LinkedIn and a program badge; idea-stage teams are welcome and no Speko account is needed to apply; no deadline or selection date is given.Source
  • 12

    The four solution pages (HVAC dispatch, law firm intake, moving quotes, towing) have a live demo you can call, but the sample transcript is marked "Written for this page. It is not a recording.", and "Build your own" sends visitors to sign-in and a blank agent. Recent changelog entries (call lists, a library of about 800 voices, per-country rates) make a ready-made agent practical.

    Medium impactMedium effort First result in 1 month Founding engineer

    First action

    Export the HVAC demo agent as a template.

    How to measure it

    Agents from templates with a live number

    Now
    –
    Target
    50
    By
    1 quarter

    Steps

    1. 1

      Export agents

      The four demo agents as templates.

      Founding engineer (platform) 1 week

    2. 2

      Add the button

      "Use this agent" to sign-up with the template ID.

      Founding engineer (web) 1 day

    3. 3

      Record calls

      Real demo recordings with checks.

      Founder/CEO 2 days

    4. 4

      Swap transcripts

      Recordings replace written samples.

      Founding engineer (web) 1 day

    5. 5

      Track

      Template starts to live numbers.

      Founding Head of Growth weekly

    6. 6

      Add templates

      Two more from demo-call demand.

      Founding engineer (platform) 1 month

    First result
    1 month
    Cost
    About two weeks of engineering
    Owner
    Founding engineer (platform)

    Smallest test

    Ship the HVAC template only and count live numbers attached.

    Why it works

    Add "Use this agent" to each solution page: sign-up opens the console with that agent's workflow, prompts, voice and test callers preloaded, ready to attach a number. Replace the written transcript with a real demo-call recording and its scored checks. Add two more templates where the demo calls show demand.

    Small-business builders and agencies want to start from something that already works; a template takes the first call from hours to minutes, and real recordings prove the latency claims.

    Risks and how to handle them

    • Regulated uses like legal intake

      Ship templates with the AI disclosure and a review note for the builder.

    Builds on

    Evidence

    • Your websiteSolutions: four pages (HVAC and plumbing dispatch, law firm intake, moving company quotes, towing and roadside dispatch). The HVAC page, "AI dispatcher for HVAC and plumbing companies", has a live demo ("Press the sphere to call"), a sample transcript marked "Written for this page. It is not a recording.", a price "from $0.30" for a three-minute inbound call, and "Build your own" and "Get started free" buttons to sign-in; there is no customer proof and no ready-made template.Source
    • Your websiteChangelog: 27 updates in the last 90 days; recent entries include about 800 filterable voices and voice cloning from MCP clients (November 3), Vietnamese STT benchmarks (November 3), per-country call rates shown before a call (November 2), CSV call lists (October 31), automatic refunds for calls that fail on Speko's side (October 19) and Canadian numbers and SMS (October 15 and 25).Source
  • 13

    Speko has 48 per-language benchmark pages, but each covers one stage (STT, TTS or LLM), and the 10 stack pages are named by budget, not language. The homepage says 6 different models win across 18 languages, which is exactly why a team building a Hindi or Arabic phone agent can't simply reuse the English stack. Nobody publishes the whole answer per language.

    High impactMedium effort First result in 6 weeks Benchmarks engineer

    First action

    Build the stack page for Spanish from the existing es pages.

    How to measure it

    First API calls from language stack pages

    Now
    –
    Target
    100 a month
    By
    1 quarter

    Steps

    1. 1

      Define the pick

      Rule for combining the three stages per language.

      Benchmarks engineer 2 days

    2. 2

      Generate pages

      One per measured language.

      Benchmarks engineer 1 week

    3. 3

      Add the snippet

      Router request for the stack.

      Benchmarks engineer 1 day

    4. 4

      Interlink

      From STT, TTS and LLM language pages.

      Founding engineer (web) 1 day

    5. 5

      Promote

      Post each language in its developer communities.

      Founding Head of Growth ongoing

    6. 6

      Refresh

      Automatically on each board update.

      Benchmarks engineer ongoing

    First result
    6 weeks
    Cost
    About two weeks of engineering
    Owner
    Benchmarks engineer

    Smallest test

    Ship Spanish and Hindi and compare traffic with their single-stage pages.

    Why it works

    Create /benchmarks/stacks/<language> for each measured language: the recommended STT, LLM and TTS together, with WER, end-of-turn latency and cost per minute, the runner-up, and a one-line Router request using that stack. Update it automatically when the boards change, and link each language page to it.

    "Best voice AI for Spanish" style questions are frequent and poorly served; a complete, dated answer per language is both a search page and a ready configuration.

    Risks and how to handle them

    • Missing stages for some languages

      Mark "not measured" and invite requests.

    Builds on

    Evidence

    • Your websiteSitemaps (checked November 9, 2026): the main sitemap has 38 URLs, including comparison pages against Bland, Retell and Vapi, "vapi-alternatives", four solution pages, seven integration pages and /startups. The benchmarks sitemap has 280 URLs: about 150 model pages (STT, TTS, LLM, speech-to-speech, turn-taking), 10 stack pages, 48 per-language pages (17 STT, 16 TTS, 15 LLM), five "best ... 2026" guides, /compare/stt, /compare/tts, a glossary and 47 blog posts. Neither has case studies or customer stories.Source
    • Your websiteHomepage (checked November 9, 2026): H1 "The Router for Voice AI" and subhead "Every speech model, benchmarked language by language, wired into one API."; all four sign-up buttons ("Get API key" twice, "Build your own", "Start building") link to platform.speko.ai/sign-in with no tracking tags; the $10 signup credit is not mentioned; there are no customer logos or testimonials.Source
    • Your websiteModels page: 115 models (36 STT, 28 LLM, 40 TTS, 11 speech-to-speech) with front matter last updated 2026-08-24, while the About page says 89 models, the Startup Program page 31 providers and the homepage "15 of 58" models measured only in English.Source
  • 14

    Speko ships constantly (27 changelog updates in 90 days, new models most weeks, new languages such as Vietnamese) and writes sharp findings, such as the cheapest honest stack costing $0.058 per booking against $0.703 for a premium one. But the blog has no newsletter or RSS, so a developer who reads one post has no way to hear about the next ranking change except by coming back.

    Medium impactLow effort First result in 1 month Founding Head of Growth

    First action

    Add an email field to the blog and benchmarks footers.

    How to measure it

    Newsletter subscribers

    Now
    –
    Target
    2,000
    By
    1 quarter

    Steps

    1. 1

      Pick a tool

      Simple email tool and RSS from the blog.

      Founding engineer (web) 1 day

    2. 2

      Add sign-up

      Benchmarks, blog, model pages.

      Founding engineer (web) 1 day

    3. 3

      Generate the diff

      Rank changes from data.json week over week.

      Benchmarks engineer 3 days

    4. 4

      Write issue one

      Diff, one finding, one changelog item.

      Founding Head of Growth half a day

    5. 5

      Personal alerts

      Account holders whose model was overtaken.

      Founding engineer (platform) 1 week

    6. 6

      Measure

      Opens, clicks, re-routes.

      Founding Head of Growth weekly

    First result
    1 month
    Cost
    Email tool, under $50 a month
    Owner
    Founding Head of Growth

    Smallest test

    Send one issue to Discord members and measure clicks.

    Why it works

    Launch a short weekly email: models added, rank changes by category and language, one finding from the blog, and one changelog item. Offer it on every benchmark and blog page, and publish the same feed as RSS. Send account holders a personal line when a model they use is overtaken.

    Rank changes are news to the people who chose a model; a weekly email keeps Speko top of mind between evaluations and gives existing users a reason to re-route through Speko rather than switch elsewhere.

    Risks and how to handle them

    • Weekly is too much for a team of four

      Generate most of it from the data; keep the writing to one paragraph.

    Evidence

    • Your website"True cost per solve" (July 5, 2026): the cheapest stack that actually booked the appointment cost $0.058 per booking against $0.703 for an all-premium stack, about 12 times more for the same outcome; two low-cost stacks were flagged for claiming success without booking. The blog offers no newsletter or RSS feed.Source
    • Your websiteChangelog: 27 updates in the last 90 days; recent entries include about 800 filterable voices and voice cloning from MCP clients (November 3), Vietnamese STT benchmarks (November 3), per-country call rates shown before a call (November 2), CSV call lists (October 31), automatic refunds for calls that fail on Speko's side (October 19) and Canadian numbers and SMS (October 15 and 25).Source
  • 15

    Speko's Gateway is an MIT-licensed customer-side runtime for LiveKit and Pipecat, with docs pages for both, but the repository has 12 stars and 6 forks after 311 commits, and its README mentions the hosted Router and SPEKO_API_KEY without linking to sign-up or pricing. Open-source frameworks are where voice-agent builders start.

    Medium impactMedium effort First result in 1 quarter Founder/CEO

    First action

    Add a tagged sign-up link and a quickstart to the Gateway README.

    How to measure it

    Sign-ups tagged from GitHub

    Now
    –
    Target
    150 a quarter
    By
    2 quarters

    Steps

    1. 1

      Rewrite README

      Two quickstarts and a tagged sign-up link.

      Founding engineer (platform) 2 days

    2. 2

      Build examples

      One per framework, per-language routing.

      Founding engineer (platform) 1 week

    3. 3

      Contribute

      Pull requests to community example collections.

      Founding engineer (platform) 2 weeks

    4. 4

      Reach maintainers

      Ask about integration listings.

      Founder/CEO 1 month

    5. 5

      Show up

      Answer routing questions in their communities.

      Founder/CEO weekly

    6. 6

      Measure

      Stars, tagged sign-ups, Router usage from Gateway.

      Founding Head of Growth monthly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Founder/CEO

    Smallest test

    Tag the README link and count sign-ups for a month.

    Why it works

    Rewrite the README around a two-minute quickstart for each framework, with a tagged "get a key with $10 credit" link. Contribute working examples to the LiveKit and Pipecat community example collections, write one guide each on per-language routing, and ask both communities for a listing among their integrations.

    Builders adopt the open runtime first and buy the hosted Router when they want managed routing; every example in a framework's own ecosystem is a lasting acquisition channel.

    Risks and how to handle them

    • Frameworks ship their own routing

      Lead with the benchmark data, which they don't have.

    Evidence

    • GitHubOpen Gateway repository: MIT licence, 12 stars, 6 forks, 311 commits; the README describes "the open customer-side runtime for real-time voice AI" and mentions the hosted Router and SPEKO_API_KEY but links to no pricing or sign-up page.Source
    • Launch HN"Launch HN: Speko (YC S26) – OpenRouter for Voice AI" (118 points, August 2026). Commenters valued the per-language benchmarks; asked whether latency is measured on the transport teams actually ship (one saw about 0.9s streaming against 5s non-streaming for a TTS model), whether cost reflects caching and context length, and how naturalness ratings are gathered; one asked what happens "when OpenRouter just adds voice", another suggested an "LM Arena of voice models" framing. The founder said human blind listening rounds are still used for naturalness.Source
  • 16

    Recent enterprise inbound named in the job post includes a white-label reseller, a contact-center platform and a BPO consultant. Speko's pricing has Router, Speko infra and a custom Enterprise tier, but nothing for companies that resell minutes or embed Speko in their own product: sub-accounts, volume rates, white-label options or partner terms.

    High impactMedium effort First result in 1 quarter Founding Head of Growth

    First action

    Interview the reseller and contact-center leads on what they would need to sign.

    How to measure it

    Signed platform or reseller partners

    Now
    –
    Target
    2
    By
    2 quarters

    Steps

    1. 1

      Interview

      Reseller, contact-center and BPO leads.

      Founding Head of Growth 2 weeks

    2. 2

      Define the tier

      Sub-accounts, volume rates, white-label, partner fee.

      Founder/CEO 1 week

    3. 3

      Build sub-accounts

      Per-customer usage and billing export.

      Founding engineer (platform) 1 month

    4. 4

      Publish

      Tier structure on pricing and Enterprise pages.

      Founding engineer (web) 1 day

    5. 5

      Pilot

      With the first reseller, using the pilot template.

      Founding Head of Growth 1 month

    6. 6

      Recruit partners

      Consultants and agencies with a referral fee.

      Founding Head of Growth ongoing

    First result
    1 quarter
    Cost
    About a month of engineering for sub-accounts
    Owner
    Founding Head of Growth

    Smallest test

    Offer the reseller lead a written tier proposal before building anything.

    Why it works

    Design a Platform tier: sub-workspaces per end customer, usage reporting and billing export per sub-account, volume-based per-minute rates, optional white-label, and partner terms for consultants who bring deals. Publish the structure (not necessarily every price) on the pricing and Enterprise pages and pilot it with the reseller already in the pipeline.

    One platform partner can bring the minutes of hundreds of end customers; a clear tier turns inbound curiosity into a contract the new seller can close.

    Risks and how to handle them

    • Margin squeeze on volume rates

      Set rates from measured cost per minute with a committed minimum.

    Builds on

    Evidence

    • HiringFounding Head of Growth (San Francisco, onsite; $170K to $190K base, 1.25% to 1.75% equity): "Enterprise demand already comes to us, and we have no seller yet"; recent inbound includes "a contact-center platform, a white-label reseller, and a BPO consultant"; "There is no sales playbook yet, and we want you to write it"; the hire runs pilots with written pass/fail acceptance tests through security reviews, procurement and contracts, owns the pipeline in month 1 and has signed pilots by month 3, with "an engineer on every pilot and our benchmark data for every deal".Source
    • Your websitePricing (updated November 3, 2026): Router at the provider's rate "plus 5%"; Speko infra at "$0.09 per minute" across speech-to-text, LLM and text-to-speech; both in public preview with no SLA during preview; Enterprise is custom pricing with a committed monthly minimum, a contract SLA and deployment reviews, reached by emailing [email protected]; a $10 signup credit; billing per connected second, with failed calls refunded automatically; numbers at "$1 setup and $2/month"; carrier charge plus 20%. A table lists Vapi at $0.05/min hosting, Retell AI at $0.07 to $0.31/min and Bland AI at $0.14/min. Speko's own security or compliance status is not mentioned.Source
  • 17

    The Launch HN thread (118 points) valued the per-language boards but asked pointed questions: is latency measured on the transport teams actually ship (one commenter saw about 0.9s streaming against 5s non-streaming for a TTS model), does cost reflect caching and context length, and how are naturalness ratings gathered. Model pages show results and a methodology anchor, but these specific answers aren't visible next to the numbers.

    Medium impactMedium effort First result in 1 month Benchmarks engineer

    First action

    Write the panel for the STT board.

    How to measure it

    External pages citing Speko benchmarks

    Now
    –
    Target
    +50%
    By
    2 quarters

    Steps

    1. 1

      List questions

      From Launch HN, Discord and sales calls.

      Founder/CEO 1 day

    2. 2

      Write the panel

      Transport, set, n, date, cost basis, rating method.

      Benchmarks engineer 1 week

    3. 3

      Add to pages

      Boards and model pages from data.json.

      Founding engineer (web) 1 week

    4. 4

      Publish gaps

      A public measurement roadmap.

      Benchmarks engineer 1 day

    5. 5

      Reply in public

      Return to the thread and communities with the answers.

      Founder/CEO 1 day

    6. 6

      Track

      Citations and links.

      Founding Head of Growth monthly

    First result
    1 month
    Cost
    Engineering time
    Owner
    Benchmarks engineer

    Smallest test

    Add the panel to the STT board and share it in Discord.

    Why it works

    Add a "How this was measured" panel to every board and model page: transport (streaming or not, region), audio set, sample size, date, how cost is computed (list price or effective, with caching assumptions), and how naturalness is rated. Where a measurement doesn't exist yet, say so and add it to a public measurement roadmap.

    Independence and method are the moat against generic benchmark sites; answering the sharpest critics in public turns skeptics into citers.

    Risks and how to handle them

    • Exposes weak spots

      That honesty is the point; pair each gap with a date.

    Evidence

    • Launch HN"Launch HN: Speko (YC S26) – OpenRouter for Voice AI" (118 points, August 2026). Commenters valued the per-language benchmarks; asked whether latency is measured on the transport teams actually ship (one saw about 0.9s streaming against 5s non-streaming for a TTS model), whether cost reflects caching and context length, and how naturalness ratings are gathered; one asked what happens "when OpenRouter just adds voice", another suggested an "LM Arena of voice models" framing. The founder said human blind listening rounds are still used for naturalness.Source
    • Your websiteModel page example, Nova-3: last measured 2026-10-01; WER, finalize latency, time to first token and the vendor's price per minute; a "Run Nova-3 on Speko →" button to sign-up; share options ("Download card (PNG)", "Post on X") and rank-badge embed code. There is no code snippet showing how to call the model through Speko and no Speko price for routing it.Source

Big bets 3

Larger plays that can change the trajectory.

  • 18

    Speko's public boards already measure what enterprise buyers care about, including cost per solved call (the cheapest honest stack at $0.058 per booking against $0.703 for a premium stack). Launch HN commenters argued the real value is in automatic evals rather than routing, and the Head of Growth will run pilots with written acceptance tests. A private run of the same harness on the buyer's own recordings and languages is a product in itself.

    High impactHigh effort First result in 1 quarter Founding Head of Growth

    First action

    Offer the first eval to the contact-center lead at a fixed price credited to the pilot.

    How to measure it

    Eval-to-contract conversion

    Now
    –
    Target
    50%
    By
    2 quarters

    Steps

    1. 1

      Define the offer

      Scope, sample size, price, credit terms.

      Founding Head of Growth 1 week

    2. 2

      Secure uploads

      Data handling and deletion terms.

      Founder/CEO 2 weeks

    3. 3

      Run by hand

      Three evals with the existing harness.

      Benchmarks engineer 1 month

    4. 4

      Standard report

      Ranked stacks, failures, recommended route.

      Benchmarks engineer 2 weeks

    5. 5

      Productize

      Self-serve upload and report in the console.

      Founding engineer (platform) 1 quarter

    6. 6

      Track

      Eval-to-pilot-to-contract.

      Founding Head of Growth monthly

    First result
    1 quarter
    Cost
    Engineering time and compute per eval
    Owner
    Founding Head of Growth

    Smallest test

    Run one eval by hand for a current inbound lead.

    Why it works

    Offer a fixed-price "Speko Eval": the buyer uploads a sample of calls, Speko runs the full benchmark harness on it across candidate stacks and returns a ranked report with WER, latency, cost per solved call and failure examples, plus the recommended route. The fee is credited against a pilot or Enterprise contract. Productize it in the console once three have been run by hand.

    It answers the HN critique by making evals the entry product, charges for the work that pilots otherwise give away, and ends every engagement with Speko's router as the obvious next step.

    Risks and how to handle them

    • Sensitive call recordings

      Deletion after the report, written terms, and the SOC 2 scope checked first.

    • Buyer takes the report and leaves

      The fee is paid either way, and routing changes as models change.

    Builds on

    Evidence

    • Your website"True cost per solve" (July 5, 2026): the cheapest stack that actually booked the appointment cost $0.058 per booking against $0.703 for an all-premium stack, about 12 times more for the same outcome; two low-cost stacks were flagged for claiming success without booking. The blog offers no newsletter or RSS feed.Source
    • HiringFounding Head of Growth (San Francisco, onsite; $170K to $190K base, 1.25% to 1.75% equity): "Enterprise demand already comes to us, and we have no seller yet"; recent inbound includes "a contact-center platform, a white-label reseller, and a BPO consultant"; "There is no sales playbook yet, and we want you to write it"; the hire runs pilots with written pass/fail acceptance tests through security reviews, procurement and contracts, owns the pipeline in month 1 and has signed pilots by month 3, with "an engineer on every pilot and our benchmark data for every deal".Source
    • Launch HN"Launch HN: Speko (YC S26) – OpenRouter for Voice AI" (118 points, August 2026). Commenters valued the per-language benchmarks; asked whether latency is measured on the transport teams actually ship (one saw about 0.9s streaming against 5s non-streaming for a TTS model), whether cost reflects caching and context length, and how naturalness ratings are gathered; one asked what happens "when OpenRouter just adds voice", another suggested an "LM Arena of voice models" framing. The founder said human blind listening rounds are still used for naturalness.Source
  • 19

    Speko's founder said on Launch HN that human blind listening rounds are still used to rate naturalness, and a commenter suggested an "LM Arena of voice models" framing. Artificial Analysis already runs public voice voting, but only on generic samples. Speko can run the same idea on phone-call audio, by language, where its boards are already the reference.

    High impactHigh effort First result in 1 quarter Benchmarks engineer

    First action

    Build a voting page for one language with ten voices.

    How to measure it

    Monthly votes

    Now
    –
    Target
    20,000
    By
    2 quarters

    Steps

    1. 1

      Design the round

      Same line, two voices, phone-quality audio.

      Benchmarks engineer 1 week

    2. 2

      Build the page

      Vote, reveal, share.

      Founding engineer (web) 2 weeks

    3. 3

      Protect it

      Rate limits and vote screening.

      Founding engineer (web) 1 week

    4. 4

      Publish votes

      Next to measured TTS scores.

      Benchmarks engineer 1 week

    5. 5

      Recruit panels

      Native speakers per measured language.

      Founding Head of Growth 1 quarter

    6. 6

      Report

      Monthly results post.

      Founding Head of Growth monthly

    First result
    1 quarter
    Cost
    Engineering time and TTS generation costs
    Owner
    Benchmarks engineer

    Smallest test

    Run one Spanish round in Discord for a week.

    Why it works

    Launch a public "which sounds better on a call?" page: two anonymous clips of the same phone-agent line, per language, one vote, then the reveal. Publish the vote counts on the TTS boards next to the measured scores, rate-limit and screen for vote stuffing, and invite native speakers in each measured language to join rated panels.

    Voting is shareable, brings repeat visits and grows a panel of raters per language that nobody else has; it also strengthens the naturalness numbers critics asked about.

    Risks and how to handle them

    • Vendors game the votes

      Anonymous clips, screening and published method.

    Builds on

    Evidence

    • Launch HN"Launch HN: Speko (YC S26) – OpenRouter for Voice AI" (118 points, August 2026). Commenters valued the per-language benchmarks; asked whether latency is measured on the transport teams actually ship (one saw about 0.9s streaming against 5s non-streaming for a TTS model), whether cost reflects caching and context length, and how naturalness ratings are gathered; one asked what happens "when OpenRouter just adds voice", another suggested an "LM Arena of voice models" framing. The founder said human blind listening rounds are still used for naturalness.Source
    • Competitor researchRivals' own homepages: Vapi, "Voice agents for builders", "1M+ developers", "SOC 2, HIPAA, and PCI compliant"; Retell AI, "#1 AI Voice Agent Platform for Automating PHONE Calls", with HIPAA, SOC 2 Type II, GDPR and ISO 27001 claims and named customer results; Bland, "Voice AI for the calls that can't go wrong", over 230 regulated companies, SOC 2 Type I and II and HIPAA BAAs; Artificial Analysis benchmarks speech-to-text, text-to-speech and speech-to-speech models with a public voice-voting feature.Source
  • 20

    Speko has the raw material for the definitive industry report: about 150 measured models across STT, TTS, LLM, speech-to-speech and turn-taking, 48 per-language pages, cost-per-solve data, an open CC BY 4.0 data file, and 47 blog posts of findings. Today the findings are spread across posts; there is no single, citable report that journalists, analysts and enterprise buyers can quote.

    High impactMedium effort First result in 1 quarter Founder/CEO

    First action

    Choose the ten charts the first report will lead with.

    How to measure it

    Enterprise forms from the report appendix

    Now
    –
    Target
    40 per edition
    By
    2 quarters

    Steps

    1. 1

      Pick the charts

      Ten headline findings.

      Founder/CEO 1 week

    2. 2

      Pull the data

      Six-month trends from data.json history.

      Benchmarks engineer 2 weeks

    3. 3

      Write it

      Short, chart-led, dated.

      Founder/CEO 2 weeks

    4. 4

      Enterprise appendix

      Per-industry stacks behind the Enterprise form.

      Founding Head of Growth 1 week

    5. 5

      Brief press

      Journalists and analysts a week early.

      Founding Head of Growth 1 week

    6. 6

      Launch

      Site, HN, X, LinkedIn, newsletter.

      Founder/CEO 1 day

    First result
    1 quarter
    Cost
    Staff time; optional design help
    Owner
    Founder/CEO

    Smallest test

    Publish one "six months of STT" post and measure links.

    Why it works

    Publish a report every six months: how accuracy, latency and cost moved; which languages improved and which still lag English; cost per solved call trends; and the recommended stacks. Release the charts under the same CC BY licence, brief a handful of journalists and analysts ahead, and gate a deeper enterprise appendix behind the Enterprise form.

    A dated, data-backed report earns links and press that individual posts don't, and the gated appendix feeds the new seller's pipeline.

    Risks and how to handle them

    • Too much work for four people

      Reuse the boards and posts; keep it to ten charts.

    Builds on

    Evidence

    • Your websiteSitemaps (checked November 9, 2026): the main sitemap has 38 URLs, including comparison pages against Bland, Retell and Vapi, "vapi-alternatives", four solution pages, seven integration pages and /startups. The benchmarks sitemap has 280 URLs: about 150 model pages (STT, TTS, LLM, speech-to-speech, turn-taking), 10 stack pages, 48 per-language pages (17 STT, 16 TTS, 15 LLM), five "best ... 2026" guides, /compare/stt, /compare/tts, a glossary and 47 blog posts. Neither has case studies or customer stories.Source
    • Your websiteMachine-readable surfaces: robots.txt welcomes AI crawlers with "Content-Signal: ai-train=yes, search=yes, ai-input=yes"; the site publishes llms.txt and llms-full.txt, a public docs MCP at speko.ai/.well-known/mcp, and benchmark data at /benchmarks/data.json under CC BY 4.0.Source
    • Your website"True cost per solve" (July 5, 2026): the cheapest stack that actually booked the appointment cost $0.058 per booking against $0.703 for an all-premium stack, about 12 times more for the same outcome; two low-cost stacks were flagged for claiming success without booking. The blog offers no newsletter or RSS feed.Source

Wildcards 2

Unexpected experiments, each with a cheap way to test it.

  • 21

    Speko already reaches developers inside AI assistants and coding tools, publishes a public, no-auth docs MCP at speko.ai/.well-known/mcp, and releases its benchmark data as CC BY 4.0 JSON. But the public MCP serves docs; the question a builder actually asks their assistant mid-build is "what's the best STT for Hindi under 300ms?", and the answer lives on the boards.

    Medium impactMedium effort First result in 1 month Founding engineer

    First action

    Define the tool's inputs and outputs from data.json fields.

    How to measure it

    recommend_stack calls per week

    Now
    –
    Target
    1,000
    By
    1 quarter

    Steps

    1. 1

      Define the tool

      Inputs, outputs, ranking rule.

      Benchmarks engineer 2 days

    2. 2

      Build it

      Read-only, from data.json.

      Founding engineer (platform) 1 week

    3. 3

      Add config output

      Router and Gateway configuration in the answer.

      Founding engineer (platform) 2 days

    4. 4

      Announce

      Integration pages, listings, changelog.

      Founding Head of Growth 1 day

    5. 5

      Log questions

      Anonymous counts by language and use case.

      Founding engineer (platform) ongoing

    6. 6

      Feed the boards

      Measure what people ask about most.

      Benchmarks engineer monthly

    First result
    1 month
    Cost
    About two weeks of engineering
    Owner
    Founding engineer (platform)

    Smallest test

    Ship the tool for STT only and count calls.

    Why it works

    Add a read-only "recommend_stack" tool to the public docs MCP: input language, use case, latency and budget limits; output the ranked models with WER, latency, price and the exact Router configuration, plus a link to the board. Mention it on each integration page and in the directory listings.

    It puts Speko's measurements at the exact moment a stack is chosen, inside tools where Speko is already listed, and every answer ends in a Speko configuration.

    Risks and how to handle them

    • Answers go stale

      Serve from the live data file with its date.

    Evidence

    • Your websiteMachine-readable surfaces: robots.txt welcomes AI crawlers with "Content-Signal: ai-train=yes, search=yes, ai-input=yes"; the site publishes llms.txt and llms-full.txt, a public docs MCP at speko.ai/.well-known/mcp, and benchmark data at /benchmarks/data.json under CC BY 4.0.Source
    • HiringThe same post says developers find Speko through the Claude connector directory (trending there in September 2026), the Cursor marketplace and the ChatGPT app directory; Speko is backed by Y Combinator (S26) and Heavybit, has a team of four, and its engineers work from Central Asia and Jakarta.Source
  • 22

    Speko has built calling safeguards that many buyers would want to see: every call starts with an AI disclosure that no prompt can turn off, outbound callers file a business declaration first, SMS goes through carrier registration, and failed calls aren't charged. These sit scattered on the Claude integration, pricing and docs pages, while rivals sell trust to regulated buyers with compliance badges on their homepages.

    Medium impactLow effort First result in 1 month Founder/CEO

    First action

    List every safeguard already live, with its page.

    How to measure it

    Enterprise forms that visited the page

    Now
    –
    Target
    50%
    By
    1 quarter

    Steps

    1. 1

      List safeguards

      Disclosure, declaration, SMS registration, retention.

      Founder/CEO 1 day

    2. 2

      Check wording

      Describe only what the product does; no legal promises.

      Founder/CEO 1 day

    3. 3

      Write the page

      One page, plain language.

      Founding Head of Growth 2 days

    4. 4

      Add security contact

      A dedicated address for security reports.

      Founder/CEO 1 hour

    5. 5

      Link it

      Footer, Enterprise, solution and comparison pages.

      Founding engineer (web) 2 hours

    6. 6

      Keep current

      Update with each safety-related release.

      Founding Head of Growth monthly

    First result
    1 month
    Cost
    $0
    Owner
    Founder/CEO

    Smallest test

    Add a "Safeguards" section to the Enterprise page first.

    Why it works

    Write one page, "How Speko keeps calls responsible": the disclosure, the declaration, SMS registration, recording and retention options, how security reports are handled, and the SOC 2 report request. Link it from the Enterprise, solution and comparison pages and from the footer.

    Buyers in legal, home services and contact centers worry about outbound calling risks; showing the safeguards in one place answers it before security review and sets Speko apart from platforms that leave it to the builder.

    Risks and how to handle them

    • Reads as legal advice

      State product behaviour only and tell builders to check their own obligations.

    Builds on

    Evidence

    • Your websiteClaude integration page: the connector places one outbound call per request, reads transcripts and searches the docs, but cannot create or deploy agents (those are built in Claude Code, the console or the API); a two-field business declaration is filed before the first outbound call, and every call starts with an AI disclosure that no prompt can turn off; the MCP server is free and new accounts get $10 in credit.Source
    • Your websiteAbout page: "Compliance | SOC 2 Type 2 report available"; one address, [email protected], for "Product questions, support requests, security reports, billing questions, and partnership inquiries"; "89 speech and language models across 18 languages" (board snapshot 2026-11-09). Neither sitemap lists a security or trust page, and speko.ai/security returns 404.Source
    • Competitor researchRivals' own homepages: Vapi, "Voice agents for builders", "1M+ developers", "SOC 2, HIPAA, and PCI compliant"; Retell AI, "#1 AI Voice Agent Platform for Automating PHONE Calls", with HIPAA, SOC 2 Type II, GDPR and ISO 27001 claims and named customer results; Bland, "Voice AI for the calls that can't go wrong", over 230 regulated companies, SOC 2 Type I and II and HIPAA BAAs; Artificial Analysis benchmarks speech-to-text, text-to-speech and speech-to-speech models with a public voice-voting feature.Source
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • LunaPrompts growth strategyAn AI-native technical screening platform for hiring AI engineers: prompt engineering, agent code, RAG, SQL, algorithms and system design…23 growth ideas
  • Alder growth strategyAn AI employee for accounts receivable. It finds out why each invoice is late and clears the blocker: stalled approvals, AP portal…24 growth ideas
  • PST.AG growth strategyGlobal trade data for 160+ countries: customs tariffs and duty rates, free trade agreements, sanctions party lists and export control…22 growth ideas
  • Crowd Cow growth strategyAn online store for craft meat and seafood from named farms and fisheries, and a cold-chain fulfilment business that ships frozen goods in…20 growth ideas
  • Numeric growth strategyClose management, cash matching, billing and revenue recognition software for finance teams, now extended into the Financial Data…21 growth ideas
  • Kijimea growth strategyA German probiotics brand (founded 2011) selling K53, PRO and Women capsules; kijimea.com is its US direct-to-consumer store, run by…22 growth ideas
  • CreatorIQ growth strategyAn enterprise creator marketing platform and data layer (the Creator Graph) for discovery, campaign management, payments, measurement…22 growth ideas
  • Mantra Health growth strategyVirtual mental health and wellness services for colleges and universities (therapy, psychiatry, 24/7 crisis care, coaching, self-guided…21 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on November 10, 2026, and cites them throughout. Speko did not commission, review or endorse it, and Programmable is not affiliated with Speko. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how Speko runs its business.

Something wrong, or do you represent Speko? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 307 ideas

    Fix the hand-offs: sign-up buttons, the enterprise front door, comparison-page actions, homepage responsiveness, one model count, visible proof and directory attribution.

  2. Days 31 to 6010 ideas

    Turn the benchmarks into routes and referrals: model-page snippets, vendor badges, the pilot template, Startup Batch 01, solution templates, per-language stacks, the weekly changes email, Gateway partners, reseller pricing and methodology answers.

  3. Days 61 to 905 ideas

    Commit to the bigger bets: private evals as the pilot, public blind listening, a twice-yearly voice AI report, benchmark answers over the docs MCP and a responsible-calling page.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value public, per-language benchmarks that include losing results. The most cited problem: re-running evaluations every time a new model ships.

  • One API key across many speech and language models
  • Models that look good in English and fail in other languages
  • In their words: “voice stack”

Market

Competitive landscape

4 competitors compared: Vapi, Retell AI, Bland and Artificial Analysis.

  • Gap to exploit: Benchmarks that turn straight into a working route
  • Vapi: Provider costs and telephony billed on top
  • Retell AI: Wide per-minute price range

Hiring signals

Speko is a team of four hiring its first go-to-market person: a Founding Head of Growth in San Francisco who will own enterprise sales "from first meeting to collected cash", write the first sales playbook and run…

  • Founding Head of Growth: Enterprise inbound with no seller; pilots with written acceptance tests; developers found via assistant directories. Ideas…

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks firstNew accounts get 100,000 free enrichment credits with code BETA100K (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammableMediaAn AI note taker, interpreter and webinar host for your calls (opens programmable.media in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence