Resources
10 new growth playbooks, each with its number. The best ones started by distrusting a number that looked fine.
Groove had 92% net revenue retention and still stalled: seat downgrades were 59% of the revenue it lost. Ro cut about $50M of revenue on its way to a $3B+ run rate. Flo gets 40% of its subscriptions after six months. 10 new growth playbooks. The best began by distrusting a number.

In September we published 32 growth playbooks. The lesson then was that the page that wins is the one a machine can quote.
On 10 November we added 13 more. The lesson was that the biggest moves changed the offer, not the funnel.
Nine days later the library has 10 more. One of them, an operator session, is split into two entries, so they are numbered 1 to 11. Each names a company or a person, gives a number, and links to a source you can open.
Vanity metrics in real startups: what the 10 new playbooks show
Five of the 10 new playbooks start the same way. A number on the dashboard said things were fine. The team stopped trusting it and found another number that explained what was happening. That was the one they acted on.
The misleading number differed each time: retention, revenue, a reply rate, last-touch attribution, a pile of experiment wins. Each team ended up counting something closer to the money.
Take a support company whose healthy retention hid stalled growth, and a health company that grew after throwing revenue away.
Groove had 92% net revenue retention and 85% logo retention. It was also stuck. By 2024 it made about $5M a year. It was profitable every year and had never raised. "On paper, things looked fine," Alex Turnbull writes.
But new customers stayed at about 22 a month, every month. "No matter what we tried, that number wouldn't move."
Then he broke down the lost revenue. In 2025, seat downgrades made up 59% of the revenue Groove lost, and they happened three times as often as cancellations. Groove charged per seat. Its customers' support teams were shrinking because of AI. "Our customers weren't leaving. They were shrinking, and we charged for the exact thing that was shrinking."
He decided to start over in September 2025. In March 2026 he closed signups on Groove at $4.7M ARR and 1,400+ customers. Groove's cash flow paid for the new product, Helply, priced in credits for the work the AI does. No bridge round (Alex Turnbull, Helply).
Ro cut about $50M of revenue and went from a $270M run rate to $3B+. In December 2021 Ro was growing about 9% a year at a $270M run rate. It was burning cash, with most of ~$800M raised still in the bank. Zach Reitano's word for it: "zombie company."
Some of that revenue was costing Ro money. Ro shut 10+ products and cut "~$50M+ of DTC revenue that was 'vanity revenue' (upside-down unit economics)." It cut all B2B work and all external API integrations, including over 10,000 in-home visits a month.
Today Ro is at a $3B+ run rate, counted as 12 times November revenue. It has 406 people, against 463 in December 2021 and 596 at the May 2022 peak (Zach Reitano, on X). We told the full story in a separate post.
Neither founder bought more traffic first. Turnbull looked past retention to new logos and downgrades. Reitano examined the unit economics of the revenue he was counting.
A number that is going up is not proof the business is working. Find the one beneath it that would move first if it were not.
Your customer's agents can expose the same problem. Jason Lemkin says one of SaaStr's core vendors saw SaaStr's bill rise 40% this year, almost all from AI usage, at about 35,000 calls a day. Then SaaStr's agents cut those calls by about 90%. "The vendor priced itself out of 90% of its own usage" (Jason Lemkin, SaaStr). For a while, that vendor's revenue line looked like growth.
The check takes ten minutes. Pick the number you report most often. Write down what would have to be true underneath it for the business to be healthy. Then pull that second number for the last six months.
Two samples, so you can see what the entries look like
Flo gets 40% of its subscriptions more than six months after install. "40% of all subscriptions take place after six months," says Mike Charniatsou, who works on growth and web onboarding at Flo. A team judging its paywall by day-zero conversion would see a leak. At Flo, that delay is part of the model.
Cycle tracking, the core use, stays free. Flo watches for a change in the user's life, such as trying to get pregnant, a pregnancy or perimenopause, and offers the paid feature then. Conversions land at three, six and 12 months.
He also runs one check at the end of every year. Add up the uplifts the team reported from its experiments. Then ask: "Did your business grow by the same percentage?" If it did not, the gap is the share of wins that were never business wins (Mike Charniatsou, on RevenueCat's Sub Club).
Caveat: spoken figures, with no base numbers and no date for the 40%. He calls waiting six months "a luxury": an app that must pay back its ads from day zero cannot.
Needle booked 321 calls in two months. Its own warm network booked none. Needle messaged 2,253 people on LinkedIn over two weeks, 15 to 29 October 2026. People who had just done something public about the problem replied most. The reply rate was 45.8% for those who engaged with a creator's post on the topic, 34.2% for commenters in the category. Followers of Needle's own company page replied at 9.9%, and none booked.
The warm contacts were the clearest disappointment. Around its Product Hunt launch, Needle sent 558 messages to first-degree connections. It got 92 replies and zero bookings. Seventy people whose companies had just posted an SDR job gave zero positive replies. Its rule: "Judge a signal by positive replies. A reply can be a no" (Maurice Dücker, Needle).
Caveat: Needle sells the tool the play runs on, and the figures are its own. The top row is 59 people. Needle itself says to treat 45.8% "as a direction, not a benchmark."
Family 1: The number underneath (new entries)
Every entry gives the number in the source's own terms and explains the mechanism you can copy. It includes a caveat where a figure is self-reported or old.
1. Groove → Helply: high retention, flat new logos, shrinking seats. The details are above (Alex Turnbull). Bolting AI onto Groove in June 2024 (drafts, summaries, tagging) "didn't change the actual work," so it did not change the bill. A separate agent launched in January 2025. It reached 252 customers paying 1.3× what they paid before, mostly Groove's own, then stalled. On Helply's own support, median first response went from 18 minutes to under 6, and the team from six to two; the other four moved into sales and success. Caveat: the first issue of Helply's own content series. Every figure is self-reported. Check: split last year's lost revenue into cancellations and downgrades.
2. Ro: cut the revenue that loses money, then gate every hire. The details are above (Zach Reitano). From January to May 2022, about seven VPs left. In June, about 20% of the company was let go. In August, everyone was asked to opt in or opt out against newly written principles, and about 6–8% opted out within 90 days. The founders gave top performers more equity and asked for 90 days. After the layoffs Ro eliminated every open role. New headcount needs a written business case, approved by the CFO and the founders, with no automatic backfills. Caveat: Reitano tells readers to take the post "with a grain of salt." The run rate is not audited, and much of the growth rode GLP-1 drugs. He grants the wave, "The wave was real," while arguing Ro was the one that led it. Check: list your revenue lines by unit economics and find the ones that lose money on every sale.
3. Flo: judge the paywall by the year, not the first week. The details are above (Mike Charniatsou). One test shows why the team checks. A new onboarding step for perimenopause lost almost 10% of conversions. A redesign lost again. Only then did the team look at the audience: almost 75% of users were never assigned a phase. A rewrite with Flo's medical team won 23%. A fourth test that removed the step lost, which confirmed it. "Explanation usually tells you what you want to be true... the evidence tells you where to look." Check: sum last year's reported experiment wins and set the total next to real growth.
4. Needle: count positive replies, not replies. The details are above (Maurice Dücker). The sequence is deliberately plain: a connection invite, one line naming the signal, one follow-up after three days, then stop. A human approves every message. They tried to remove the human, over "hundreds of prompts, eval loops, different models." The AI got the heavy lifting right and "the judgement wrong." Check: sort last month's outbound by the signal that triggered it, and count meetings, not replies.
5. Clay: an agent re-read a $200k deal credited to a Google ad. Davide Grieco, head of growth at Clay, rebuilt attribution in about two hours with an agent. It takes every new Stage 2 opportunity and pulls Salesforce, Snowflake and Gong. It separates the last-touch source from earlier influences, then posts to Slack and logs to a sheet. A recent $200k deal had been credited to a paid Google ad. The agent found the champion had started a free trial two months earlier. They had also answered "Network" to "how did you hear about us?", and registered for two webinars (Davide Grieco, at an operator session, written up here). The same session had four more operator plays, filed as entry 8. Caveat: one deal, told by Clay, which sells the platform it was built on. Check: open your three largest recent deals and list every touch before the one that got the credit.
6. Revolut: measure cohorts, and kill bets on payback. Nik Storonsky told John Collison that Revolut was "purely organic" for its first seven years. It started serious growth budgets in 2022, with payback and LTV models per customer. Business accounts went from about 10 sign-ups a day "without any marketing" to about 1,500 a day, of which about 30% is paid. Revolut dropped its funnel model, a "state machine" of steps, because "it just became too complex." It now tracks join cohorts by country at 12, 18 and 24 months. New bets get a team of about 10 people and 12–18 months. Then they are judged on growth. Revolut Lite, an international money-transfer product in 100+ countries, did not grow organically, and its performance marketing failed on payback (Nik Storonsky, on Cheeky Pint). His warning about any single number: "KPI can be met because of external factors and because of luck." Caveat: spoken figures, with no dates on the business-account ramp. We checked the "purely organic" claim against Revolut's own accounts in a separate post. The date holds, and marketing spend then grew about tenfold in four years. Check: replace one funnel chart with a cohort chart.
How to use this library
Entries 1 to 6 are above. Entries 7 to 11 are below. Entry 8 comes from the same operator session as entry 5, filed apart because it answers a different question. Entries 7 and 9 are later additions to entries that were already in the library.
Every entry links to its source. Where a figure is self-reported by the company that benefits from it, the entry says so. The earlier entries are in the first edition and the second, and are not repeated here.
Family 2: When software buys, and who does the selling
These three look at what happens when software is the buyer, and at who, agent or person, now does the go-to-market work.
7. Lemkin's two price tests: does the agent double ACV, and is your unit price cheap enough for agents? Two SaaStr posts, four days apart (12 and 16 November). In the first, a portfolio company at about $30M ARR launched its first real agent at double the ACV, with no free version. In the first week, 3% of the customer base bought it. His rule: a real agent "does a job a human used to do" and gets paid from the labour budget. Improving a workflow lets a vendor charge a premium, not double. He now won't back an agentic B2B company unless its deal sizes are at least 2× those of pre-AI competitors (Jason Lemkin, SaaStr). In the second, he argues the unit price itself now gets tested every day. That is the 40%, 35,000-calls, 90% story above. His arithmetic: cut the price per action 80%, get 20× the agent usage, and revenue from that customer goes up 4× (Jason Lemkin, SaaStr). Caveat: the $30M company and the 40% vendor are both unnamed. The 4× is arithmetic, not a result. Check: "If you raised prices 8% last year, did you add 30% more value to your product?"
8. Four more operator plays: two where an agent took over a GTM job, two built on people. From the same operator session as entry 5.
incident.io: CMO Tom Wentworth replaced agency fees with a Claude skill that runs paid campaigns on LinkedIn, Google, Meta and X. An ABM campaign that took an agency 2–3 days now takes anyone in marketing 3 minutes, and BDR conversion is double on accounts that get one. Once a week the skill reviews its own runs and opens a pull request for him to approve.
Vercel: Cheryl Wolf built an event-operations agent, iris, herself with Claude Code. Her team went from 1–2 field events a month to 3–6, with no added headcount. Exec dinners of 20–30 seats yield 5+ sales-qualified opportunities each.
Granola: Rob Denton runs 200+ operator ambassadors: CMOs, founders and engineers, each with an agreement to post about Granola regularly. Granola boosts their posts as thought-leader ads. Those ads get 10× the engagement of its other LinkedIn ads.
Lovable: Mallory Contois's SheBuilds virtual hackathon draws about 2,500 applicants a season. It accepted 1,000 this season, up from 200 (the operators, written up here).
Caveat: self-reported, from a recorded panel. Lovable gives no conversion figure. Check: map one GTM job into its manual loops before you automate any of it.
9. Resend: get into the repo before the agent picks a provider. Zeno Rocha says Resend had 4.3 million users and 120,000 paying customers when he recorded with swyx. Its first distribution was a template library, React Email, released with docs showing how to send it through SendGrid. By the time a developer switched providers, Resend's code was already in the repo. Add .md to Resend's pricing URL and an agent gets a table in one request, where a person drags a slider. Its MCP server grew from about 20 tools to more than 80 (Zeno Rocha, on Latent Space). Caveat: the page also cites a study in which Claude Code picked Resend first in 64 of 102 email recommendations. The same page says a later, broader study was more competitive, with Postmark on top for most models. We do not lean on either. Check: ask an agent what your product costs, using only your pricing page.
Family 3: Getting found, and getting sold (new entries)
10. Webflow: ChatGPT traffic converted 24% against 4% for non-brand Google. Josh Grant, then VP of Growth at Webflow, put ChatGPT traffic at ~6× the conversion rate of non-brand Google search (24% vs 4%). He reported LLM-attributed customers up ~8× year on year in Q3 2025 (Josh Grant, StackedGTM). His playbook for winning citations on other people's pages has five steps. Mine real buying questions from sales calls and tickets. Find which sources each AI engine cites for competitors. Go after niche, data-rich publications, not famous ones. Fight for the top three spots of a cited list. Rank targets on five factors, including commercial value and cost (Josh Grant, interviewed by Maja Voje). Caveat: the figures are Q3 2025 and are not tied to this exact playbook. The comparison is generous: someone clicking from ChatGPT has often compared options already, so branded search would be a fairer match. Grant now sells this work. Check: split your LLM referrals out of "direct" before you judge the channel.
11. ConvertKit: from $1,300 to $725,000 MRR by asking about frustrations. This one is from 2017. It is new to the library, not new. Nathan Barry sent one email "over a thousand times": "Is anything frustrating you with [current tool]?" Don't pitch, just ask, then get on a call with anyone who replies. He won a tiny niche where the top customers knew each other, then stepped sideways into sibling niches. Switching cost killed deals, so ConvertKit migrated accounts over $79 a month for free. Those accounts churned at about 1.5%, against a typical 5.5% (Nathan Barry). Caveat: the post is dated July 2017 and does not say when ConvertKit was at $1,300. The tools it names have changed. Check: find the one thing that stops a buyer from switching, and do it for them.
The checks, collected
- Split lost revenue into cancellations and downgrades.
- Put new-logo count next to retention, every month.
- Rank revenue lines by unit economics and cut the ones that lose money.
- Make every new hire a written business case.
- Judge a paywall by twelve months of conversions, not the first week.
- Sum last year's experiment wins and compare them with real growth.
- Count meetings by signal, not replies.
- List every touch before the one that got credit on your biggest deals.
- Swap one funnel chart for a cohort chart.
- Test your unit price against an agent that calls it 35,000 times a day.
- Map one GTM job into its manual loops before you automate any of it.
- Ask an agent what your product costs.
- Split LLM referrals out of "direct" before you judge the channel.
- Remove the biggest switching cost for the buyer yourself.
The one to take
Read entries 1 and 2 together. Before you change the play, check that the number telling you to change it is the right one.
Keep reading
The rest of this article: 10 new growth playbooks, each with its number. The best ones started by distrusting a number that looked fine.
The remaining sections are behind this point.
It costs an email address. The same address goes on Makersfuel, the MakersClaw briefing that runs Tuesday to Saturday, and every issue carries an unsubscribe link.
The address is used for the newsletter and nothing else. Privacy.

