Plexe growth playbook

Make the self-serve tools a visible front door, then connect them to sales. 1) Add an /autosuggest product page on plexe.ai (demo box, the three plans, the Google Places comparison, the travel stories) and add it to the sitemap and llms.txt. 2) Add a “Self-serve tools” band to /pricing under the engagement models: Trial, Growth and Scale, plus room resolution once its access is clear (idea 5). 3) Link it from the platform page and the homepage Travel card. 4) In the Autosuggest console, ask for company and role when a key is created (optional fields). 5) Trigger a short founder email when an account passes half the free daily cap, upgrades to Growth or asks for supplier IDs (the “approved accounts” step). The email offers a scoped pilot on supplier routing or recommendations and links the matching story. 6) Report keys, upgrades and pilot calls every month.

A free API key is a stronger buying signal than a cold demo request: the team has a travel search problem and engineers ready to ship. A two-person company cannot find all those teams by outbound, but it can answer the ones that identify themselves within a day. Public prices on the main site also give cautious enterprise buyers a low-risk way to start working with Plexe.

[ ] /autosuggest page on plexe.ai with demo, plans and comparison
[ ] “Self-serve tools” band on /pricing (Trial, Growth $99, Scale $499)
[ ] Linked from the homepage Travel card and the platform page
[ ] Added to the sitemap and llms.txt
[ ] Optional company and role fields at key creation
[ ] Alerts: half the free cap, upgrade, supplier-ID request
[ ] One pilot email per trigger, each linking the matching travel story
[ ] Monthly report: keys, upgrades, pilot calls, pilots signed

• Self-serve users read the pilot offer as a hard sell

  Send one note, only after a usage trigger, and make it about their use case.

• Small hobby sites dilute the signal

  Trigger outreach only on usage, upgrades or supplier-ID requests.

• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source
• Your websitePricing page: H1 “Custom-built agents. Owned by you.”; “Plexe does not sell seat-based software or publish fictional tiers.”; three engagement models (Scoped pilot, Full custom build, Embedded team) with no prices or durations; “Book a demo” goes to Cal.com; no mention of Autosuggest, room resolution or any self-serve product.Source
• Your websitePlatform page: “The backend for models and agents”: Builder (“Natural language to production model”), Agents and Runtime (A/B testing against incumbents, drift monitoring, retraining); 15 components; a “graduation right” to take any solution fully in-house; every route goes through “Talk to us”; Autosuggest and room resolution are not mentioned.Source
• Your websiteSitemap (30 URLs, checked November 10, 2026): home, careers, platform, pricing, studio, five industry pages, the stories index and 13 stories, “plexe-vs-in-house-ai-team”, “what-are-ai-agents” and four legal pages; no blog, no product pages for Autosuggest or room resolution, and no lastmod dates.Source
• Your websiteIndustry pages: Travel has a “Plug and play” section, “Tools you can use today.”, linking Room Mapping to sandbox.plexe.ai and Autosuggest to suggest.plexe.ai, plus seven travel stories; the E-commerce page lists use cases (segmentation and churn flags, cross-sell, semantic catalogue search, campaign agents) but no plug-and-play tools, and two stories, one a proof of concept.Source
• Your websiteTravel aggregator story: “Every search hit every supplier API” and “Long-tail suppliers consumed budget but rarely converted”; real-time models now score and rank suppliers per search; −60% infrastructure cost; faster responses (no figure); the aggregator’s own team maintains and retrains the models.Source

Change the header to something exact (for example “13 builds, 8 in production”), add a status filter (In production / Pilot / Proof of concept) next to the industry filter, show production stories first, label the jewellery card E-commerce, and add one line to each pilot story on what would take it to production.

Precise labels make the production claims credible, and enterprise buyers judge vendors on exactly this kind of detail.

• Showing pilots weakens the page

  Lead with the eight production results and frame pilots as work in progress.

• Your websiteStories page: 13 stories, all “Details anonymised”, none with a customer quote; the header says “Shipped to production” but five are labelled Pilot or Proof of concept; the jewellery marketplace card has no industry label; stories carry engineering detail such as “~10,700 automated API tests”, “52 serverless functions” and “33 agent tools”.Source
• Competitor researchRivals’ own homepages: DataRobot, “the only agent workforce platform built for outcomes — not endless pilots”, with service-led “purpose-built agents” that run “On-prem. Hybrid. Cross-cloud.”; Faculty, “Frontier AI for the frontlines of the world”, AI services plus its Faculty Frontier product; Vervotech, AI hotel and room mapping with “1000+ travel portals using Vervotech”.See the research Source

For each engagement model, add a typical length (the pilot can reuse the homepage’s six-week build), what is delivered, who from Plexe works on it, and a “typically from” budget band. Add a short FAQ on ownership, the graduation right and where it runs (your VPC or hosted).

Price anchors qualify buyers before the call and save founder time on deals that cannot fit the budget.

• Bands put off larger buyers

  Publish starting points, not ceilings.

• Your websitePricing page: H1 “Custom-built agents. Owned by you.”; “Plexe does not sell seat-based software or publish fictional tiers.”; three engagement models (Scoped pilot, Full custom build, Embedded team) with no prices or durations; “Book a demo” goes to Cal.com; no mention of Autosuggest, room resolution or any self-serve product.Source
• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source
• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source

Write one positioning paragraph and use it everywhere. Add a “Plexe for enterprises” box to the docs welcome page linking to the platform page and stories. Add a “Running Plexe in production” section to the README. Update the YC one-liner and description. Keep the open-source library’s own pitch, but make clear how it relates to the platform.

Consistent descriptions shorten diligence, and AI assistants and search engines summarise the company from these exact pages.

• Open-source users feel abandoned

  Keep the library’s pitch and roadmap visible.

• DocsDocs welcome page: “Build ML models from natural language in minutes”, covering the open-source Python library (“pip install plexe”) and a “Managed Platform” at console.plexe.ai; no pricing and no mention of agents, industries or enterprise engagements.Source
• GitHubplexe-ai/plexe: “Build a machine learning model from a prompt”; 2.6k stars, 265 forks, Apache-2.0; the README describes Plexe as “a managed cloud service” and links to plexe.ai and docs.plexe.ai.Source
• Y CombinatorYC profile: “Open-source agents to build predictive ML models from a prompt”; founded 2025; team size 2; London; Spring 2025 batch.Source
• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source

Add a “Request a key” button to the sandbox and the Travel page. The short form asks for company, number of suppliers and monthly room volume, and goes to the founders. Publish the trial limits and say whether paid use is per call or per hotel. Every request is a qualified travel lead, and gets the same founder follow-up as idea 1.

Visitors who test a model on a real hotel are ready to act. Today the page gives them nowhere to go next.

• Requests from very small sites

  Ask for room volume and prioritise by it.

• #1 Put Autosuggest’s public prices on plexe.ai and turn every free key into a travel pilot conversation

• Your websiteRoom resolution sandbox (sandbox.plexe.ai): model plx-rm-1 “Groups every supplier’s offers for a hotel by room, and those rooms into room types”; visitors can try it in the browser, read docs and OpenAPI, or “sign in with your API key”, but the page does not say how to get a key or what it costs.Source
• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source
• Competitor researchRivals’ own homepages: DataRobot, “the only agent workforce platform built for outcomes — not endless pilots”, with service-led “purpose-built agents” that run “On-prem. Hybrid. Cross-cloud.”; Faculty, “Frontier AI for the frontlines of the world”, AI services plus its Faculty Frontier product; Vervotech, AI hotel and room mapping with “1000+ travel portals using Vervotech”.See the research Source

Add three booking questions: industry, the workflow they want to automate, and where their data lives (their cloud or hosted). Send an automatic confirmation with the closest story and a one-page security note (SOC 2 Type 2, runs in your VPC, you own the IP). Use the answers to open the call with the right example.

Prepared buyers have better first calls, and the answers show which industries and workflows bring the most demand.

• Questions reduce bookings

  Keep them short and optional except industry.

• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source
• Your websitePricing page: H1 “Custom-built agents. Owned by you.”; “Plexe does not sell seat-based software or publish fictional tiers.”; three engagement models (Scoped pilot, Full custom build, Embedded team) with no prices or durations; “Book a demo” goes to Cal.com; no mention of Autosuggest, room resolution or any self-serve product.Source
• Y CombinatorYC profile: “Open-source agents to build predictive ML models from a prompt”; founded 2025; team size 2; London; Spring 2025 batch.Source
• HiringForward Deployed Engineer (India, remote; ₹1.5M–₹3M; 3+ years), Plexe’s only open role: “work directly with customers”, 30–50% travel to customer sites, integrate Plexe into customer infrastructure, run training sessions, write documentation and relay customer requirements to the roadmap.Source

Fix the generator so it includes every page in the sitemap. Add a “Self-serve tools” section linking suggest.plexe.ai and sandbox.plexe.ai with one-line descriptions and plans, add a short summary of the three engagement models, and add the 13th story. Each month, ask the main assistants three buyer questions and note whether Plexe appears.

The file is generated from source content, so the fix is small and it improves how every assistant summarises Plexe.

• Assistants ignore llms.txt

  Treat it as one input; the page fixes in ideas 1 and 4 matter too.

• Your websiteplexe.ai/llms.txt, “a curated index for AI assistants”: “We pioneered the production-AI platform. Our senior team forward-deploys it inside your product and ships AI agents you own - in weeks, not quarters.” It lists the home page, 12 of the 13 stories, the industry pages and legal pages, but not the platform, pricing, studio, comparison or guide pages, or the self-serve tools.Source
• Your websiteSitemap (30 URLs, checked November 10, 2026): home, careers, platform, pricing, studio, five industry pages, the stories index and 13 stories, “plexe-vs-in-house-ai-team”, “what-are-ai-agents” and four legal pages; no blog, no product pages for Autosuggest or room resolution, and no lastmod dates.Source
• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source

Serve the hero image as AVIF or WebP at responsive sizes through Astro’s image tools, preload it, and keep the decorative layers lazy. Fix the low-contrast text and the heading order, then re-run PageSpeed.

It stops the slide in field speed before it affects visitors, and it removes easy objections from a vendor accessibility check.

• Image quality drops

  Compare visually before shipping.

• PageSpeedHomepage test (November 10, 2026): mobile performance 72 (desktop 87); lab LCP 4.1 s, where the LCP element is the “galaxy” hero image; biggest fix “Improve image delivery” (about 1.5 s, 337 KB). Real visitors pass Core Web Vitals (p75 LCP 1.8 s, INP 81 ms, CLS 0), though field LCP is 13% slower than before; two accessibility issues (colour contrast, heading order).Source

Ask each logo customer for one of three things, from most to least public: a named story, an attributed quote on an anonymised story, or a reference call for serious prospects. Offer to write the draft and give them approval. Start with customers whose systems are in production. Add “Reference calls available” to the stories and pricing pages.

A recognisable reference shortens enterprise diligence more than any feature list, and a quote alone lifts trust in the published numbers.

• Large customers refuse to be named

  Offer the quote or reference-call options instead.

• #2 Split the stories page into “In production” and “Pilots”, so the header matches the labels

• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source
• Your websiteStories page: 13 stories, all “Details anonymised”, none with a customer quote; the header says “Shipped to production” but five are labelled Pilot or Proof of concept; the jewellery marketplace card has no industry label; stories carry engineering detail such as “~10,700 automated API tests”, “52 serverless functions” and “33 agent tools”.Source
• Your websiteFortune 500 travel story: a recommendations model A/B tested against the incumbent, “+7% conversion uplift”; a self-serve modelling workflow for PMs and analysts; “6 models live, run by PMs and analysts”; no quotes and no dates.Source

Add a “Taking this to production at your company” section to the README and a matching docs page: what changes in production (evals, drift monitoring, guardrails, VPC), and a link to the platform page and stories. Run monthly open office hours for library users. Each month, review the organisations on public issues and discussions to spot teams in Plexe’s five industries, and invite them to office hours, never cold-pitch.

Open-source users are already working on the problem. A clear path from library to production converts a share of them without outbound sales.

• The community sees it as a bait-and-switch

  Keep shipping the library and keep office hours useful.

• #4 Bring the docs, GitHub README and YC profile in line with “Production AI for the enterprise”

• GitHubplexe-ai/plexe: “Build a machine learning model from a prompt”; 2.6k stars, 265 forks, Apache-2.0; the README describes Plexe as “a managed cloud service” and links to plexe.ai and docs.plexe.ai.Source
• DocsDocs welcome page: “Build ML models from natural language in minutes”, covering the open-source Python library (“pip install plexe”) and a “Managed Platform” at console.plexe.ai; no pricing and no mention of agents, industries or enterprise engagements.Source
• Your websiteFortune 500 travel story: a recommendations model A/B tested against the incumbent, “+7% conversion uplift”; a self-serve modelling workflow for PMs and analysts; “6 models live, run by PMs and analysts”; no quotes and no dates.Source
• Your websitePlatform page: “The backend for models and agents”: Builder (“Natural language to production model”), Agents and Runtime (A/B testing against incumbents, drift monitoring, retraining); 15 components; a “graduation right” to take any solution fully in-house; every route goes through “Talk to us”; Autosuggest and room resolution are not mentioned.Source

Add a text box where the visitor describes the workflow. Return a draft diagram (trigger, agent steps, approvals, actions, systems touched) plus the five questions Plexe will ask in scoping. Saving it needs a work email, and it is attached to the booking so the first call starts from their draft. A founder reviews each draft before the call.

Visitors get something useful right away, Plexe gets a qualified, pre-scoped lead, and the page shows the product instead of describing it.

• Drafts look naive to experts

  Label them as a starting point and have a founder review them.

• Spam and junk input

  Rate-limit and require email to save.

• #6 Ask three questions on the booking page and send the matching story before the call

• Your websiteStudio page: “Describe your agent.” above a mock workflow canvas (an email trigger, a support agent, an approval check and a Slack message) and a “Build done” panel; the only call to action is “Want this in production?” with “Talk to us”; there is nothing to try.Source
• Your websitePlatform page: “The backend for models and agents”: Builder (“Natural language to production model”), Agents and Runtime (A/B testing against incumbents, drift monitoring, retraining); 15 components; a “graduation right” to take any solution fully in-house; every route goes through “Talk to us”; Autosuggest and room resolution are not mentioned.Source
• Your websitePricing page: H1 “Custom-built agents. Owned by you.”; “Plexe does not sell seat-based software or publish fictional tiers.”; three engagement models (Scoped pilot, Full custom build, Embedded team) with no prices or durations; “Book a demo” goes to Cal.com; no mention of Autosuggest, room resolution or any self-serve product.Source

Start with the supplier already on the logo wall. Propose a listing in its integration or partner pages, a joint how-to (“add hotel search that returns your supplier IDs in one script tag”), and a tracked signup link. Offer that supplier’s clients a longer free trial. Repeat with the next supplier if the first one works.

Suppliers want their clients to launch faster, and their integration pages reach exactly the developers who need Autosuggest.

• Suppliers see Autosuggest as competing with their own tools

  Position it as the search box in front of their inventory, not a mapping rival.

• #1 Put Autosuggest’s public prices on plexe.ai and turn every free key into a travel pilot conversation

• Your websiteAutosuggest vs Google Places page: claims “2.4x to 4.4x” lower cost from 100,000 search sessions a month (for 1,000,000 sessions, $3,550 at Google’s cheapest setup vs $799 on Scale); results “can be shown in your own design”; returns hotel IDs from Vervotech, ZentrumHub, Hotelbeds and Xeni “for approved accounts”; Google prices “checked 19 October 2026”.Source
• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source
• Your websiteHotel search story: Autosuggest covers 224 countries and ships as “@plexe/autosuggest, a web component with no dependencies plus a React wrapper”; “anyone can sign up at suggest.plexe.ai, create a website key and add the box to a page with one script tag”; each pick is reported back to tune ranking, and sites can turn this off; no accuracy or usage figures are published.Source

Build a calculator page: searches per day, suppliers called per search, cost per call, and the share of bookings from the top suppliers. Show the estimated spend on calls that rarely convert, and a savings range with clear caveats. Offer a free one-week analysis of a sample of their search logs, run in their own environment, which leads into a scoped pilot.

It turns a case study into the buyer’s own number, and the log analysis is a natural first step towards the paid pilot.

• Estimates set false expectations

  Show a range and the assumptions behind it.

• Your websiteTravel aggregator story: “Every search hit every supplier API” and “Long-tail suppliers consumed budget but rarely converted”; real-time models now score and rank suppliers per search; −60% infrastructure cost; faster responses (no figure); the aggregator’s own team maintains and retrains the models.Source
• Your websiteIndustry pages: Travel has a “Plug and play” section, “Tools you can use today.”, linking Room Mapping to sandbox.plexe.ai and Autosuggest to suggest.plexe.ai, plus seven travel stories; the E-commerce page lists use cases (segmentation and churn flags, cross-sell, semantic catalogue search, campaign agents) but no plug-and-play tools, and two stories, one a proof of concept.Source
• Your websiteSitemap (30 URLs, checked November 10, 2026): home, careers, platform, pricing, studio, five industry pages, the stories index and 13 stories, “plexe-vs-in-house-ai-team”, “what-are-ai-agents” and four legal pages; no blog, no product pages for Autosuggest or room resolution, and no lastmod dates.Source

Publish “Plexe vs AutoML and agent platforms” and “Plexe vs AI consultancies”. Compare on ownership of code and models, where it runs, time to production, who maintains it after launch, and how it is priced. Keep it fair and dated, cite each rival’s own pages, and link the matching stories. Have each page reviewed for claims before publishing.

Comparison searches come from buyers already shortlisting, and the pages answer the question sales hears most.

• Outdated rival claims

  Date every page and cite sources.

• Hacker NewsLaunch HN (November 2025, 85 points): “Build production-grade ML models from prompts”, a hosted product with usage pricing and a “$10 minimum top-up”; commenters asked how it differs from AutoML or an LLM wrapper, and raised training times (about 45 minutes to a few hours), tabular-only data and fuller code exports.Source
• Your websiteContent pages: “Plexe vs. an in-house AI team” cites McKinsey and BCG, gives no price figures and says there are “no public fictional tiers or per-seat packages”; “What are AI agents?” is a guide of about 1,000 words with no date or author. They are the only content pages outside the stories.Source
• Competitor researchRivals’ own homepages: DataRobot, “the only agent workforce platform built for outcomes — not endless pilots”, with service-led “purpose-built agents” that run “On-prem. Hybrid. Cross-cloud.”; Faculty, “Frontier AI for the frontlines of the world”, AI services plus its Faculty Frontier product; Vervotech, AI hotel and room mapping with “1000+ travel portals using Vervotech”.See the research Source

Offer every handed-over customer a 45-minute quarterly review: drift and retraining history, A/B results, incidents and costs, in a one-page report. End each review with one proposed next workflow from that industry’s use-case list, scoped as a pilot. Describe the review on the pricing page as part of hand-over.

Customers who see measured results each quarter renew and expand, and the report gives the internal champion something to share upward.

• Reviews feel like upselling

  Lead with the health report; keep the proposal to one slide.

• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source
• Your websitePlatform page: “The backend for models and agents”: Builder (“Natural language to production model”), Agents and Runtime (A/B testing against incumbents, drift monitoring, retraining); 15 components; a “graduation right” to take any solution fully in-house; every route goes through “Talk to us”; Autosuggest and room resolution are not mentioned.Source
• Your websiteFortune 500 travel story: a recommendations model A/B tested against the incumbent, “+7% conversion uplift”; a self-serve modelling workflow for PMs and analysts; “6 models live, run by PMs and analysts”; no quotes and no dates.Source

Add three things to every trip: a one-hour “what we built and how” session for other teams at the customer, one meeting with a prospect in the same city from the target industries, and a short field note on new problems heard. Track pipeline that comes from visits.

Inside a customer, the best lead is the team next door, and a working system on site is the strongest demo.

• Selling distracts from delivery

  Cap it at one session and one meeting per trip.

• HiringForward Deployed Engineer (India, remote; ₹1.5M–₹3M; 3+ years), Plexe’s only open role: “work directly with customers”, 30–50% travel to customer sites, integrate Plexe into customer infrastructure, run training sessions, write documentation and relay customer requirements to the roadmap.Source
• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source

Publish one build note a month under /notes: the architecture, how it was evaluated, what failed and what changed, anonymised like the stories, dated and signed by the engineer. Share each one on LinkedIn and X, and submit the best to Hacker News. Add a date and author to the agents guide.

Technical buyers trust engineering detail over claims, and dated, signed pages build the credibility search engines and AI assistants reward.

• Revealing customer details

  Same anonymisation as the stories, plus customer sign-off.

• Your websiteSitemap (30 URLs, checked November 10, 2026): home, careers, platform, pricing, studio, five industry pages, the stories index and 13 stories, “plexe-vs-in-house-ai-team”, “what-are-ai-agents” and four legal pages; no blog, no product pages for Autosuggest or room resolution, and no lastmod dates.Source
• Your websiteContent pages: “Plexe vs. an in-house AI team” cites McKinsey and BCG, gives no price figures and says there are “no public fictional tiers or per-seat packages”; “What are AI agents?” is a guide of about 1,000 words with no date or author. They are the only content pages outside the stories.Source
• Your websiteStories page: 13 stories, all “Details anonymised”, none with a customer quote; the header says “Shipped to production” but five are labelled Pilot or Proof of concept; the jewellery marketplace card has no industry label; stories carry engineering detail such as “~10,700 automated API tests”, “52 serverless functions” and “33 agent tools”.Source
• HiringForward Deployed Engineer (India, remote; ₹1.5M–₹3M; 3+ years), Plexe’s only open role: “work directly with customers”, 30–50% travel to customer sites, integrate Plexe into customer infrastructure, run training sessions, write documentation and relay customer requirements to the roadmap.Source

Run a small, capped search-ad test on queries about Places autocomplete pricing and hotel search alternatives, landing on the comparison page. Link the page from the new /autosuggest page and the Travel page. Refresh the price table monthly and keep the dated note. Measure keys created and upgrades per dollar spent.

Developers comparing autocomplete costs are close to switching, and a dated, specific comparison converts better than a product page.

• Comparison prices go out of date

  Keep the checked date and refresh monthly.

• #1 Put Autosuggest’s public prices on plexe.ai and turn every free key into a travel pilot conversation

• Your websiteAutosuggest vs Google Places page: claims “2.4x to 4.4x” lower cost from 100,000 search sessions a month (for 1,000,000 sessions, $3,550 at Google’s cheapest setup vs $799 on Scale); results “can be shown in your own design”; returns hotel IDs from Vervotech, ZentrumHub, Hotelbeds and Xeni “for approved accounts”; Google prices “checked 19 October 2026”.Source
• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source

Package semantic catalogue search as a sandbox: upload a product file or point at a public catalogue feed, try natural-language search against it, and compare it with keyword results. Give it a free trial and published plans like Autosuggest’s, and route active users to churn and cross-sell pilots.

Travel shows the pattern: a tool people can try creates self-identified leads. E-commerce has many more mid-size buyers than travel.

• Distraction from paid delivery

  Build only after the interviews confirm demand, and reuse existing components.

• #1 Put Autosuggest’s public prices on plexe.ai and turn every free key into a travel pilot conversation

• Your websiteIndustry pages: Travel has a “Plug and play” section, “Tools you can use today.”, linking Room Mapping to sandbox.plexe.ai and Autosuggest to suggest.plexe.ai, plus seven travel stories; the E-commerce page lists use cases (segmentation and churn flags, cross-sell, semantic catalogue search, campaign agents) but no plug-and-play tools, and two stories, one a proof of concept.Source
• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source
• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source

Each quarter, publish aggregated, anonymised trends: fastest-rising destinations, landmark and area searches versus city searches, and the share of searches in non-Latin scripts. Never show site-level data, and leave out sites that opted out. Gate the full data pack by work email, and pitch the headline findings to travel trade press.

Original data earns coverage and links that a two-person team cannot buy, and it presents Plexe as a travel data company to its exact buyers.

• Customers object to data use

  Aggregate only, minimum thresholds, honour opt-outs and say so in the terms.

• Volumes too small to be meaningful

  Wait until volume supports it, or start with an annual edition.

• #1 Put Autosuggest’s public prices on plexe.ai and turn every free key into a travel pilot conversation

• Your websiteHotel search story: Autosuggest covers 224 countries and ships as “@plexe/autosuggest, a web component with no dependencies plus a React wrapper”; “anyone can sign up at suggest.plexe.ai, create a website key and add the box to a page with one script tag”; each pick is reported back to tune ranking, and sites can turn this off; no accuracy or usage figures are published.Source
• Your websiteAutosuggest (suggest.plexe.ai), “Hotel and destination autocomplete for travel sites”: Trial free (5,000 requests a day, “every endpoint, no card”), Growth $99 a month (500,000 requests included), Scale $499 a month (5,000,000 requests included), billed monthly through Stripe; its own console, API docs, playground, OpenAPI spec, llms.txt and a comparison page against Google Places.Source

For use cases with a clean live metric (recommendations, supplier routing, churn), offer a scoped pilot with a reduced fixed fee plus a success fee that applies only if the A/B test clears an agreed lift. Publish the method: metric, test length, minimum sample and how ties are handled. Start with two pilots before listing it on the pricing page.

It removes the buyer’s biggest fear, paying for a pilot that goes nowhere, and only a team confident in its A/B results can make the offer.

• Tests fail for reasons outside Plexe’s control

  Agree data and traffic requirements before the start.

• Cash flow strain

  Keep a fixed fee that covers delivery cost.

• #3 Give the three engagement models a typical length and a starting budget band

• Your websiteFortune 500 travel story: a recommendations model A/B tested against the incumbent, “+7% conversion uplift”; a self-serve modelling workflow for PMs and analysts; “6 models live, run by PMs and analysts”; no quotes and no dates.Source
• Your websitePlatform page: “The backend for models and agents”: Builder (“Natural language to production model”), Agents and Runtime (A/B testing against incumbents, drift monitoring, retraining); 15 components; a “graduation right” to take any solution fully in-house; every route goes through “Talk to us”; Autosuggest and room resolution are not mentioned.Source
• Your websitePricing page: H1 “Custom-built agents. Owned by you.”; “Plexe does not sell seat-based software or publish fictional tiers.”; three engagement models (Scoped pilot, Full custom build, Embedded team) with no prices or durations; “Book a demo” goes to Cal.com; no mention of Autosuggest, room resolution or any self-serve product.Source
• Competitor researchRivals’ own homepages: DataRobot, “the only agent workforce platform built for outcomes — not endless pilots”, with service-led “purpose-built agents” that run “On-prem. Hybrid. Cross-cloud.”; Faculty, “Frontier AI for the frontlines of the world”, AI services plus its Faculty Frontier product; Vervotech, AI hotel and room mapping with “1000+ travel portals using Vervotech”.See the research Source

Create a platform plan: one integration, volume pricing across all the platform’s client sites, the platform’s own branding, and reporting per client. Offer the platform a revenue share or a discount tied to volume. Start with one travel-tech platform Plexe already knows from its customer base.

Distribution through platforms multiplies reach without sales headcount, which suits a two-person team.

• Dependence on one partner

  Keep direct self-serve growing alongside.

• #1 Put Autosuggest’s public prices on plexe.ai and turn every free key into a travel pilot conversation

• Your websiteAutosuggest vs Google Places page: claims “2.4x to 4.4x” lower cost from 100,000 search sessions a month (for 1,000,000 sessions, $3,550 at Google’s cheapest setup vs $799 on Scale); results “can be shown in your own design”; returns hotel IDs from Vervotech, ZentrumHub, Hotelbeds and Xeni “for approved accounts”; Google prices “checked 19 October 2026”.Source
• Your websiteHotel search story: Autosuggest covers 224 countries and ships as “@plexe/autosuggest, a web component with no dependencies plus a React wrapper”; “anyone can sign up at suggest.plexe.ai, create a website key and add the box to a page with one script tag”; each pick is reported back to tune ranking, and sites can turn this off; no accuracy or usage figures are published.Source
• Your websiteHomepage (checked November 10, 2026): title “Plexe - Production AI for the enterprise”; the only call to action is “Speak to us”, a Cal.com booking link; “Backed by Y Combinator” and “SOC 2 Type 2”; 15 customer logos including Expedia Group and lastminute.com; four anonymised results (+7% conversion, −60% infrastructure cost, +20% enrollment, +28% repeat purchase); “How we work”: Embed (week 1), Build (weeks 2–6), Hand over, with Plexe staying on call.Source

Open an invite-only tier for tabular prediction (the case commenters said was supported): fixed monthly plans with compute caps, a clear limit on training time, and a one-click “take this to production with Plexe” upgrade into an engagement. Invite open-source users and Autosuggest customers first.

Analysts who get a model working on their own become internal champions for a production engagement.

• Support load for a two-person team

  Keep it invite-only with strict caps.

• Blurs the enterprise positioning

  Present it as a trial path into the platform, not a separate product.

• #10 Make the 2.6k-star open-source library a front door to enterprise builds

• Hacker NewsLaunch HN (November 2025, 85 points): “Build production-grade ML models from prompts”, a hosted product with usage pricing and a “$10 minimum top-up”; commenters asked how it differs from AutoML or an LLM wrapper, and raised training times (about 45 minutes to a few hours), tabular-only data and fuller code exports.Source
• DocsDocs welcome page: “Build ML models from natural language in minutes”, covering the open-source Python library (“pip install plexe”) and a “Managed Platform” at console.plexe.ai; no pricing and no mention of agents, industries or enterprise engagements.Source
• Your websiteFortune 500 travel story: a recommendations model A/B tested against the incumbent, “+7% conversion uplift”; a self-serve modelling workflow for PMs and analysts; “6 models live, run by PMs and analysts”; no quotes and no dates.Source
• Your websitePlatform page: “The backend for models and agents”: Builder (“Natural language to production model”), Agents and Runtime (A/B testing against incumbents, drift monitoring, retraining); 15 components; a “graduation right” to take any solution fully in-house; every route goes through “Talk to us”; Autosuggest and room resolution are not mentioned.Source