Calltree growth playbook

Give each page its own HTML file at its own URL, prerendered at build time with real text, a title, a meta description and a canonical tag: /, /terms, /privacy, /events and one URL per event. As a stopgap on day one, set CloudFront to answer missing paths with the app shell and a 200 status, so the links stop erroring. Then add a sitemap.xml and robots.txt and submit them in Search Console. Write the titles around what buyers search for, for example 'Calltree | AI support reps for technical troubleshooting calls'. Put the page language on <html>. Finally, check what a non-JavaScript reader sees on each page, because that is what link previews, many AI crawlers and security-review tools see.

Enterprise deals pass a vendor check before anyone books time: security and legal teams open the terms and privacy links, and buyers search the name or ask an AI assistant. A 404 or a blank title at that point reads as an unfinished company. Real, titled pages cost a few days of work, and every later idea in this plan depends on pages that can be found.

[ ] /terms, /privacy, /events return 200 (with and without www)
[ ] Each page is its own prerendered HTML file with real text
[ ] No {{ }} placeholders in the HTML served
[ ] Unique <title> and meta description per page
[ ] lang="en" on <html>, self-canonical on each page
[ ] sitemap.xml and robots.txt live and submitted
[ ] Unknown paths return a real 404 page with links home
[ ] Link previews checked in LinkedIn, Slack and email

• The stopgap 200 rule also hides real missing pages

  Keep it only until prerendered files exist, then return a true 404 for unknown paths.

• Prerendering breaks the interactive demo

  Prerender text and metadata only and let the app take over after load.

• Your websiteLinks (checked twice, November 9, 2026): calltree.ai/terms, /privacy and /events, and their www versions, return HTTP 404 when opened directly, as do /sitemap.xml and /robots.txt. The Terms ("Last updated July 11, 2025"), Privacy Policy and Events views exist only inside the homepage file.Source
• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source
• PageSpeedMobile test (November 9, 2026): performance 67 (desktop 95); the main content, the heading "Meet your AI CX agent.", appears at 3.9s (target 2.5s); 11,315 KB in 56 requests, 3,061 KB of it images; 440 ms blocking time, PostHog (253 KB) and Google Tag Manager (177 KB) the heaviest scripts; SEO 82 (no title, no description); accessibility 85 (no page language, low contrast, no main landmark). Too little real-visitor traffic for Core Web Vitals field data.Source
• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source

Pick one sentence built on the hardest thing Calltree does, for example: 'AI support reps that resolve long technical troubleshooting calls inside your own tools, and show you the cost of every second.' Use it as the homepage H1 and description, the YC one-liner, the LinkedIn tagline and the founders' bios. Then send corrections to the directories that list Calltree. Keep 'analytics' as the second line, so it reads as part of the product rather than a separate pivot.

In a crowded field of AI support agents, buyers remember the vendor that owns one specific job. Repeating the same sentence everywhere also teaches search engines and AI assistants a single answer to 'what is Calltree'.

• Narrowing scares off broader deals

  Lead with troubleshooting and list other call types underneath.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree

• Site historyIn December 2025 the homepage headline was "Augment your staffing with Agentic AI"; the headline readers now receive is "{{ activeEvent.name }}". Domain registered January 16, 2025.Source
• Y CombinatorYC profile: "Enterprise-grade AI support reps for call centers" (W2025); YC's customer-service list calls it "AI support reps that take voice calls and chats and can use a company's internal tools and dashboards", next to 18 other YC support companies including Parahelp, Open, Observe.AI and Netomi.See the research Source
• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source

Fill the trust center with what security reviewers ask for, in text they and their tools can read. Cover current audit status, with a date if an audit is in progress. Add the subprocessor list and data retention for call and screen recordings, and say whether customer data trains shared models. Add how agents get access to internal tools, plus a contact and an NDA request for the full report. Link it from the homepage footer and the demo follow-up email.

Security review is often the slowest step in an enterprise AI deal. Answering the standard questions up front shortens that step and shows a 14-person company is ready for large accounts.

• Overclaiming an audit in progress

  State exact status and dates only; have counsel check the wording.

• Your websiteTrust center: to an automated reader the page shows only the heading "Trust center", with no frameworks, audit status or subprocessors in its text.Source
• Competitor researchSierra: "The leading conversational AI platform for businesses"; outcome-based pricing ("Pay for a job well done"); dozens of enterprise logos and named-executive quotes; lists SOC 2, ISO 27001, ISO 42001, HIPAA, GDPR, PCI DSS and FedRAMP.See the research Source
• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source

Compress and resize the demo images and any video, and lazy-load everything below the first screen. Load analytics after the page renders. Decide whether both Google Analytics and PostHog are needed and who owns each. Remove duplicated JavaScript and set long cache lifetimes on static files. Retest on mobile after each change.

A faster first screen keeps the busy buyer who taps a link between meetings. It also removes the most visible lab warnings that buyers' own tools may flag.

• Losing analytics data

  Run both tools in parallel for two weeks before dropping one.

• PageSpeedMobile test (November 9, 2026): performance 67 (desktop 95); the main content, the heading "Meet your AI CX agent.", appears at 3.9s (target 2.5s); 11,315 KB in 56 requests, 3,061 KB of it images; 440 ms blocking time, PostHog (253 KB) and Google Tag Manager (177 KB) the heaviest scripts; SEO 82 (no title, no description); accessibility 85 (no page language, low contrast, no main landmark). Too little real-visitor traffic for Core Web Vitals field data.Source

Add Open Graph and Twitter tags to the homepage, events pages and any case study: a title, a one-line description and a 1200x630 image. The image should show the product handling a real troubleshooting call, with the positioning sentence. Make one share image per event, so invites look like invites. Fix the accessibility basics PageSpeed flags at the same time: page language, contrast and a main landmark.

A clear preview gets more clicks on shared links, and founder-led posts are Calltree's cheapest channel.

• Cached old previews on LinkedIn

  Refresh the URL in LinkedIn Post Inspector.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree

• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source
• PageSpeedMobile test (November 9, 2026): performance 67 (desktop 95); the main content, the heading "Meet your AI CX agent.", appears at 3.9s (target 2.5s); 11,315 KB in 56 requests, 3,061 KB of it images; 440 ms blocking time, PostHog (253 KB) and Google Tag Manager (177 KB) the heaviest scripts; SEO 82 (no title, no description); accessibility 85 (no page language, low contrast, no main landmark). Too little real-visitor traffic for Core Web Vitals field data.Source

Write three short case studies. Where a customer will not be named, describe it generically, for example 'a public consumer-electronics retailer'. Each should cover the call type, the starting handle time and cost, what the AI does inside which systems, escalation rate, resolution rate and the change in cost per call. Add one quote from an operations leader, even if anonymous. Put the headline numbers on the homepage, and use them in every proposal.

Operations leaders buy on measured results from a comparable queue. Anonymised numbers from a public brand still carry weight, and long technical calls are a category bigger rivals rarely show proof for.

• Customers refuse any disclosure

  Publish ranges and aggregate numbers across deployments instead.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree

• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source
• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source
• Competitor researchCresta: "The AI operating system for customer experience"; named case studies with numbers (Snap Finance "5.5x higher containment", Windstar Cruises "70% chat containment"); quality management and conversation intelligence on 100% of conversations.See the research Source

Build a simple calculator. A visitor enters calls per month, average handle time, loaded cost per agent hour and the share of calls that are technical troubleshooting. It shows the cost of one second, the cost of the current handle time, and savings at three levels of automation and time saved. Offer a one-page PDF of the result and a button to book a call that starts from their numbers.

Operations leaders need a business case for their CFO. Giving them the first draft makes Calltree the vendor whose numbers frame the decision.

Cost per second = loaded cost per agent hour / 3,600
Annual cost of one second per call = cost per second x calls per month x 12
Current annual handle cost = AHT seconds x cost per second x calls per month x 12
Savings = (seconds saved per call x eligible calls + automated calls x AHT) x cost per second x 12
Example: 1,000,000 calls a month, $40 an hour, 20 seconds saved = about $2.7M a year

• Inflated savings hurt credibility

  Show conservative, expected and best cases, with assumptions visible.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree

• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• Competitor researchLeaping AI: "AI voice and text agents" for home-improvement and home-services firms; named customers (Eurowings launched multilingual voice AI in its call centers in under two months); a pricing page and an ROI question in its FAQ.See the research Source
• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source

Package a fixed-scope audit: the prospect shares 1,000 recent call and screen recordings from one queue under NDA. Within two weeks Calltree returns three things: the top five time sinks with seconds and dollars attached, the share of calls an AI rep could resolve, and a proposed pilot. Make it the main call to action for cold outbound and the homepage, next to 'book a demo'.

It turns a hard sell (put AI on our customer calls) into an easy one (show us where our money goes). The audit's numbers become the pilot's success targets, and they come from the product Calltree already built.

• Audits eat team time without converting

  Offer only to qualified accounts with a named budget owner.

• Data-sharing approval is slow

  Accept redacted or on-premises processing options.

• #7 Turn the launch post's "20 seconds per call = $5M a year" math into a calculator on the site

• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• Third-party listingDirectory listing: agents "learn from top-performing reps", work inside CRMs and dashboards and hand off to humans with full context; claims costs cut "by up to 70%"; workflow analysis and "Opex savings reports"; no pricing, demo only.Source
• Site historyIn December 2025 the homepage headline was "Augment your staffing with Agentic AI"; the headline readers now receive is "{{ activeEvent.name }}". Domain registered January 16, 2025.Source

Set up a demo line with a scripted but realistic scenario: a customer whose router keeps dropping, a backorder check and a warranty claim. Show the number on the homepage with 'Call our agent now'. After the call, text a link to a transcript with handle time and the actions the agent took in each system, plus an offer to run the same scenario on the buyer's own data.

Hearing the agent deal with a messy, multi-step problem is more convincing than any screenshot. It also qualifies buyers who care about voice and long calls.

• Spam or abuse calls

  Rate limits per number and a short cap on call length.

• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source
• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source

Write one deep page for consumer hardware and electronics support. It should show the calls Calltree handles (setup, connectivity, diagnostics, warranty, backorders), the systems it works in, and typical handle times and escalation rates from anonymised deployments. Close with the audit offer. Add a second page only when a second vertical has proof.

Searchers and buyers with the same problem see themselves on the page, and the vertical focus differentiates Calltree from general AI support platforms.

• Too narrow for other buyers

  Keep the homepage general and use the page for targeted campaigns.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree
• #2 Say what Calltree is in one sentence, and use it on the site, YC and every directory

• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source
• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source

Write factual comparison pages for the three rivals that come up most in deals. Each should cover the call types handled (especially long technical calls), working inside legacy tools, human supervision, the analytics that price each second, time to go live and pricing model. Be generous where rivals are stronger, and be specific about where Calltree is different.

Comparison searches come from buyers late in the process. Honest pages also give AI assistants a clear description of how Calltree differs.

• Claims go out of date

  Date each page and cite rivals' public pages.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree
• #2 Say what Calltree is in one sentence, and use it on the site, YC and every directory

• Y CombinatorYC profile: "Enterprise-grade AI support reps for call centers" (W2025); YC's customer-service list calls it "AI support reps that take voice calls and chats and can use a company's internal tools and dashboards", next to 18 other YC support companies including Parahelp, Open, Observe.AI and Netomi.See the research Source
• Competitor researchSierra: "The leading conversational AI platform for businesses"; outcome-based pricing ("Pay for a job well done"); dozens of enterprise logos and named-executive quotes; lists SOC 2, ISO 27001, ISO 42001, HIPAA, GDPR, PCI DSS and FedRAMP.See the research Source
• Competitor researchCresta: "The AI operating system for customer experience"; named case studies with numbers (Snap Finance "5.5x higher containment", Windstar Cruises "70% chat containment"); quality management and conversation intelligence on 100% of conversations.See the research Source

Write a short 'How pricing works' page with no list prices. Describe a fixed-fee pilot on one queue with success targets from the audit. Then explain production pricing: per resolved call or per minute handled, with a floor and volume tiers. Say which costs are included (integration, supervision by AI Operations Analysts, reporting) and give an example deal shape. Use the same structure in every proposal.

A clear model removes a common reason not to take a meeting. Tying price to resolved calls matches what operations leaders are measured on and answers Sierra's outcome pitch.

• Locks in a model too early

  Describe the model only; keep the numbers in proposals.

• #8 Offer a free recording audit: analyse 1,000 of a prospect's calls and hand back the cost of every wasted second

• Competitor researchSierra: "The leading conversational AI platform for businesses"; outcome-based pricing ("Pay for a job well done"); dozens of enterprise logos and named-executive quotes; lists SOC 2, ISO 27001, ISO 42001, HIPAA, GDPR, PCI DSS and FedRAMP.See the research Source
• Competitor researchLeaping AI: "AI voice and text agents" for home-improvement and home-services firms; named customers (Eurowings launched multilingual voice AI in its call centers in under two months); a pricing page and an ROI question in its FAQ.See the research Source
• Third-party listingDirectory listing: agents "learn from top-performing reps", work inside CRMs and dashboards and hand off to humans with full context; claims costs cut "by up to 70%"; workflow analysis and "Opex savings reports"; no pricing, demo only.Source

Package it as a named offer. During the first 60 days, a trained analyst watches live calls, takes over when needed and logs every intervention. The customer gets a weekly report on interventions, fixes shipped and the drop in escalations. Publish the playbook outline and the intervention rate trend from anonymised deployments, and include the analyst in the kickoff call.

Risk of a bad AI call reaching a customer is the main blocker in enterprise deals. A visible human net with measured interventions makes the first yes easier, and the open role becomes part of what customers buy.

• Implies the AI needs a babysitter

  Frame it as a launch phase and show the trend falling.

• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source
• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source

For each target account, estimate monthly call volume and handle time from public sources: annual reports, support pages, job posts for agents. Run those through the calculator, and open with one line: 'At your volume, every second on a support call is worth about $X a year.' Offer the free recording audit to confirm it. Have the founders send it from their own addresses, in small batches, with follow-ups.

A number tied to the reader's own operation earns replies far more often than a feature list, and it leads straight to the audit.

• Wrong estimates look careless

  Show the assumptions and invite correction.

• #7 Turn the launch post's "20 seconds per call = $5M a year" math into a calculator on the site
• #8 Offer a free recording audit: analyse 1,000 of a prospect's calls and hand back the cost of every wasted second

• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• Y CombinatorYC profile: "Enterprise-grade AI support reps for call centers" (W2025); YC's customer-service list calls it "AI support reps that take voice calls and chats and can use a company's internal tools and dashboards", next to 18 other YC support companies including Parahelp, Open, Observe.AI and Netomi.See the research Source

Each month, send a one-page report to the customer's budget owner. It should cover calls handled, minutes saved, cost saved against the agreed baseline, escalation and resolution rates, the top new time sink found, and the next fix proposed. Keep a running total since go-live. Review it in a quarterly call that also plans the next queue.

A visible running total of dollars saved is the strongest defence at renewal, and it builds the case for expansion.

• Disputed baselines

  Agree them in writing at kickoff.

• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• Third-party listingDirectory listing: agents "learn from top-performing reps", work inside CRMs and dashboards and hand off to humans with full context; claims costs cut "by up to 70%"; workflow analysis and "Opex savings reports"; no pricing, demo only.Source

Every quarter, run the analytics across all of a customer's recorded queues, not just the automated one. Rank the others by volume, handle time, repetitiveness and fit for an AI rep. Present the top candidate with projected savings and a 30-day add-on pilot. Price the expansion with the same per-resolution model.

The customer sees expansion as the next step in a proven saving, not a new purchase, and Calltree uses data no rival on the account has.

• Customer will not share other queues

  Start with aggregated call-reason data instead of recordings.

• #15 Send each customer's executive a monthly savings report in dollars, from the before-and-after measurement Calltree already does

• Your websiteHomepage (November 9, 2026): the HTML readers receive has no <title> or meta description and shows unfilled placeholders such as "{{ heroHeading }}" and "{{ p.label }}"; tagline "Conversational AI for smarter customer support."; a demo chat with "Check backorder" and "Check warranty" and an analyst asking "What have our WiFi calls been about lately?"; no pricing, named customers, case studies or testimonials; contact is [email protected].Source
• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source

Ask each satisfied operations leader to join a small reference group. In return, they get early access to new features, a seat on a twice-yearly product council, and recognition they can use internally. Agree up front what each will do: a reference call, an anonymous quote, a named quote or a logo. Track reference calls per deal and thank people promptly.

Peer references from comparable operations are often the deciding factor in enterprise purchases, and a council keeps the product close to buyers.

• Over-asking the same references

  Cap calls per reference per quarter.

• #6 Publish three anonymised case studies of long troubleshooting calls, with handle time and cost per call

• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source
• Competitor researchSierra: "The leading conversational AI platform for businesses"; outcome-based pricing ("Pay for a job well done"); dozens of enterprise logos and named-executive quotes; lists SOC 2, ISO 27001, ISO 42001, HIPAA, GDPR, PCI DSS and FedRAMP.See the research Source

Run one small dinner or roundtable a month for 10-12 operations and support leaders in San Francisco, then other cities where customers are. Pick a topic buyers care about, for example 'What one second per call costs you', with a customer co-host where possible. Give each event its own prerendered page with a title, date, location, Event structured data and a share image. Follow up within two days with the session notes and the audit offer.

Small rooms of peers build trust faster than ads, and the invite form already exists. Indexable event pages also give the site fresh, specific content.

• Low attendance

  Co-host with a customer and confirm by phone the week before.

• #1 Fix the 404s on calltree.ai/terms, /privacy and /events and give every page a real title, so buyers, search engines and AI assistants can read Calltree

• Your websiteEvents section: search, filters and a "Request an invite" form ("Space is limited") asking for name, email and phone, but the page readers receive lists no event and says "No events match your search or filters."Source
• Your websiteLinks (checked twice, November 9, 2026): calltree.ai/terms, /privacy and /events, and their www versions, return HTTP 404 when opened directly, as do /sitemap.xml and /robots.txt. The Terms ("Last updated July 11, 2025"), Privacy Policy and Events views exist only inside the homepage file.Source

Offer a standard first deployment: Calltree's AI, supervised by an analyst, takes troubleshooting calls outside business hours and at weekends, with a warm handoff to the morning team. Measure answer rate, resolution rate and handoffs against the old voicemail or callback baseline. After 60 days, propose extending into daytime overflow.

It replaces something customers currently get poorly (a callback or a long wait) rather than a human they like. Approval comes faster, and success is easy to see.

• Thin supervision at night

  Start with fewer hours and widen as intervention rates fall.

• #13 Sell the supervised launch: name it, measure it, and put the AI Operations Analysts who watch live calls in front of buyers

• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source
• Third-party listingDirectory listing: agents "learn from top-performing reps", work inside CRMs and dashboards and hand off to humans with full context; claims costs cut "by up to 70%"; workflow analysis and "Opex savings reports"; no pricing, demo only.Source

Aggregate anonymised data across customers and audits, with permission. Report average handle time by call type, seconds lost to system switching and manual entry, the cost of a second at typical volumes, and how AI resolution changes these for long technical calls. Publish the headline findings openly, gate the full benchmark, and present it at the events and in outreach.

Original benchmarks get cited by analysts, press and AI assistants. They make Calltree the reference for call-time economics, not one more AI agent vendor.

• Sample too small to generalise

  State the sample size and present it as a first edition.

• #8 Offer a free recording audit: analyse 1,000 of a prospect's calls and hand back the cost of every wasted second
• #15 Send each customer's executive a monthly savings report in dollars, from the before-and-after measurement Calltree already does

• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• Third-party listingDirectory listing: agents "learn from top-performing reps", work inside CRMs and dashboards and hand off to humans with full context; claims costs cut "by up to 70%"; workflow analysis and "Opex savings reports"; no pricing, demo only.Source

Choose the one or two contact-center platforms most common among current customers and prospects. Build a certified integration for each, and apply to its partner programme and marketplace. Publish an integrations page listing every system the agents already work in. Then co-sell: brief the platform's account teams on the long-technical-call use case and the free audit.

Partner marketplaces shorten procurement and bring warm introductions from account teams who want their customers to automate more. For a small vendor, that borrowed trust is hard to get any other way.

• Platform builds a competing feature

  Focus on long technical calls and analytics the platform does not offer.

• Y CombinatorYC profile: "Enterprise-grade AI support reps for call centers" (W2025); YC's customer-service list calls it "AI support reps that take voice calls and chats and can use a company's internal tools and dashboards", next to 18 other YC support companies including Parahelp, Open, Observe.AI and Netomi.See the research Source
• YC launchLaunch post: an AI agent that finds workflow inefficiencies in call-center recordings and quantifies "the time and cost of each action"; under 1% of screen recordings are reviewed; switching between five legacy apps costs "20 seconds per call", or "$5M/year"; impact is measured before and after fixes. It asks readers to share it with VPs of Operations; one founder led product for AWS's contact-center solution.Source
• Third-party listingDirectory listing: agents "learn from top-performing reps", work inside CRMs and dashboards and hand off to humans with full context; claims costs cut "by up to 70%"; workflow analysis and "Opex savings reports"; no pricing, demo only.Source

Each month, publish three anonymised cases where the AI stumbled on a long technical call: a misread symptom, a loop, a wrong system action. Explain how the analyst caught it and what changed in the product, and show the trend in interventions. Keep customer details out and have each entry approved.

Openness about failure is rare in AI sales and builds trust with careful enterprise buyers. It also shows the supervision process in action and gives the founders a steady source of posts.

• Rivals use it against Calltree

  Always pair each failure with the fix and the falling trend.

• #13 Sell the supervised launch: name it, measure it, and put the AI Operations Analysts who watch live calls in front of buyers

• HiringAI Operations Analyst (San Francisco, $70K-$100K, new grads welcome, open on YC's job board): supervises "AI agents during live customer conversations", is the escalation point for frustrated customers, documents patterns in AI performance and tests new AI experiences; one weekend day a week. Calltree (YC W25, team of 14) has "multiple enterprise customers, including public retail brands" and its AI handles "30+ minute technical troubleshooting calls". Applicants record a 2-3 minute support roleplay video.Source
• Competitor researchCresta: "The AI operating system for customer experience"; named case studies with numbers (Snap Finance "5.5x higher containment", Windstar Cruises "70% chat containment"); quality management and conversation intelligence on 100% of conversations.See the research Source