A&A INSIGHTS
Before scaling acquisition: how to count a customer as returned
Before spending more on acquisition, check whether your first-month customers came back on their own. With a handful of customers, use a named list, not a rate.
日本語で読む
THE STARTING POINT
The first metric to look at before increasing acquisition is not the number of new closes but whether the customers you took in your first month issued a request of the same kind again, on their own, in the next cycle. While you still have only a few customers, do not express this as a rate: observe it as a named list of those first-month customers, and write one line of reason for every customer who did not come back.
Before you scale acquisition, look at whether your first-month customers came back
A&A perspective
The first metric to put in front of acquisition spend is whether the customers you took in your first month issued a request of the same kind again, on their own, in the next cycle. The reader assumed here is a founder who delivers an AI-based service alone and has acquired the first few paying customers. You are almost certainly already counting inquiries and new closes. Those numbers describe how much is coming in; they say nothing about whether what came in stays. If you increase acquisition spend before you know why customers do not stay, you pay repeatedly for the same defect. This ordering is our own recommendation, not a procedure the sources prescribed for small businesses.
From the sources
Anu Hariharan, a partner at Y Combinator, said in the firm's discussion published on December 1, 2017 that the first step to check before you set up a growth team is whether you have strong retention. Her words were: "before you set up a growth team, the number one step you need to check is whether you have strong retention." She went on to say that too often companies form a growth team and then wonder why they are not growing fast, that this is a leaky bucket of water, and that the most important step is retention.
From the sources
Stripe reported on May 28, 2026, writing about companies incorporated through Stripe Atlas without any cofounders, that nearly 30% of the customers acquired in the first month returned the following month at top-decile solo startups, compared with 8% at middle-decile startups. The piece was written by Jesse Carey of Stripe Atlas. Stripe frames the gap as suggesting that the stronger founders reach product-market fit earlier; it does not assert that retention produced the revenue.
A&A perspective
For a founder delivering the work alone, this ordering is a question about where money goes. There are broadly only two destinations for the next unit of time and cash: acquire more customers, or repair what you deliver for the customers you already have. Without an observation of whether they came back, you are choosing between those two by instinct.
At four customers, a retention rate is not a usable form
From the sources
Stripe states the population behind those figures. The method was that they analyzed thousands of solo-founded Atlas startups incorporated in 2022 and 2023, each with at least two years of revenue data, and compared middle-decile solo founders with those in the top decile by total revenue in their first two years. The roughly 30% and 8% are therefore values computed for two groups out of a distribution of thousands of companies.
A&A perspective
When you have four customers, a retention rate is not a usable expression. One customer returning or not moves the figure by 25 points, so the number reflects who happened to be an early customer more than it reflects the quality of what you deliver. You can write that you improved from 25% to 50% month over month while the only thing that actually happened is that one customer came back. So in place of a rate, we propose a different unit of observation: a named list of the first-month customers. In a list, a small denominator stops being a weakness and becomes the advantage that you can write a reason against each line. This substitution is ours and is not in the sources.
Hypothetical example
Here is a hypothetical. A founder starts a business delivering AI-produced meeting minutes and summaries, and takes four customers in the first month. The list reads: Company A, a request of the same kind arrived the following month. Company B, no request; the reason is that the contact moved departments and no user was left inside the company. Company C, no request; the reason is that the delivered format could not be pasted into their internal report, which added work. Company D, no request; the reason is unconfirmed because they cannot be reached. As a rate this is 25%. As a list it shows that a candidate repair sits in the delivery format, and that one line is still unconfirmed. This is an invented example for explanation; there is no actual customer or result behind it.
| Decision | Counts | Does not count |
|---|---|---|
| One return | A paid next request of the same kind, initiated by the customer | A reply to your own message, a quote request alone, free rework |
| Observation window | One cycle in which your own work naturally repeats | Borrowing the sources' "following month" as it stands |
| Form of the denominator | A named list of the first-month customers | A percentage while you have only a few customers |
| Recurring billing | A record of requests or usage that actually occurred in the period | A record that the charge succeeded, on its own |
| Reason for no return | One written line per customer, with unconfirmed marked as such | Closing it out as chemistry or budget |
Count a return only when the customer initiates the next request of the same kind
From the sources
In the same discussion, Gustaf Alstromer, who worked on growth at Airbnb, said that for most companies you just want to figure out repeat purchase rate, repeat booking or repeat use of some kind, and added "that repeat use has to meaningful." He went on to say that you cannot count repeat use simply because you sent a notification and someone came back, and that "Just the act of coming back isn't meaningful unless you do something that gives you value from the product."
From the sources
In the same passage, Hariharan said the most important thing is a sign of the action of using your product, and gave examples. For Airbnb it was a booking, not a visit to the site. For Uber it was a trip booked, not cancelled and actually completed. For Stitch Fix, she explained that the company's north star metric for a long time was the number of second fixes in the first four months, because data had shown that customers who ordered a second shipment within four months of their first purchase retained markedly better than others.
A&A perspective
Carried over to a contract or service business, that definition becomes a rule that you count one return only when three conditions hold at once. First, the request originated with the customer; a reply to your own "how is it going" message does not count. Second, it is work of the same kind as before; an entirely different request is not a return but a different engagement. Third, it is agreed as paid work; a request for a quote alone does not count. This is our substitution of the sources' definition into contract work, not a standard Y Combinator set for contract businesses. We keep intact the distinction that repeat value and contract renewal are not the same thing.
Hypothetical example
Some examples of what does not count: a reply to an update you sent, a free additional revision, rework to correct a defect in the previous delivery, and a first request from a different company that arrived through a referral. The last of those is a good event for the business, but it is not an observation that a first-month customer came back, so it belongs on a separate line.
Set the observation window from your own cycle, not from "the following month"
From the sources
In the same discussion Alstromer said that Airbnb was unlike most products at the time because "travel is a very rare occurrence." Because people travel once or maybe twice a year, looking at retention on the guest side means either waiting a long time or holding several years of historical metrics in good shape, and he noted that when you start something you often do not even have a year's worth of metrics.
From the sources
Stripe's analysis, by contrast, used the following month as its window, while the window described for Stitch Fix in the same discussion was the first four months. Under the same word, retention, the observation window is chosen separately to fit the cycle of the business.
A&A perspective
So choose the window to match one cycle in which your own work naturally repeats. For monthly production or operations work, one cycle is a month; for quarterly research or planning, three months; for annual filings or audits, a year. What matters is that you fix the window in writing before you look at the number. Stretch the window afterwards and most numbers start to look better. Do not set the window and read the result on the same day. This is our operating recommendation.
Recurring billing makes "returned" happen by itself
From the sources
Stripe also reports that top-decile founders used recurring billing at a higher rate than middle-decile ones: 26 percentage points higher for B2B and 20 percentage points higher for B2C. Stripe does not state this as a cause, writing instead that part of the reason top solo founders retained more customers might be that they were much more likely to use recurring billing. The piece also reports that among solo-founded B2B startups, top-decile founders retained first-month customers at six times the rate of median founders.
From the sources
Returning to Alstromer's point, the act of coming back is not in itself meaningful unless you do something that gives you value from the product.
A&A perspective
There is a practical trap here. Switch to a monthly or recurring charge and the charge succeeds whether or not the customer does anything, which means a record that looks like a return appears automatically. A successful charge proves that the payment method was valid; it does not prove that the work happened again. If you introduce recurring billing, keep a separate field, apart from the billing record, for whether a request or actual usage occurred in that period. Note also that when you read reasons for departure out of billing records, you first have to separate customers who cancelled themselves from payments that failed. That is the subject of a separate article, "Before you read AI SaaS retention: separate churn from failed payments."
Being AI-native is not a reason customers come back
From the sources
Stripe lists four traits of the top decile, and retention is one of them. Another is building AI-native products, which the piece defines as meaning that the product's core functionality depends on AI models; top-decile founders were reported to be about twice as likely as median founders to be building AI-native companies. The remaining two traits are selling globally from launch and building for businesses.
From the sources
The piece's wording on retention was that top solo founders retained a much larger share of their first-month customers than middle-decile founders, suggesting they reach product-market fit earlier.
A&A perspective
These are two separate traits that appeared together in the same top decile. The causal chain "customers come back because the product is AI-native" is written nowhere in that article. This is where a misreading is costly. That your product's core depends on AI models is only the fact that you share one trait with the stronger group; it is not a guarantee that a different trait, retention, comes attached. Do not take causation for your own business out of an observation about a distribution.
Decide what you will fix before you look at the number
A&A perspective
An observation that someone did not come back is not yet an action. Before you look, write down three candidate places to repair. First, the promise you sold: what the customer thought they were buying differs from what actually arrived. Second, the shape of the handover: the deliverable is correct, but it is not in a form their organization can use as it is. Third, the entrance to the next request: they are satisfied, but no procedure exists on their side for asking again. Which candidate the reason on each line points to is what decides your next move. A line whose reason is unconfirmed is a line to ask about, not to repair.
From the sources
That there is upside in repairing things for existing customers also appears as an observation in Stripe's analysis. The piece reports that by the start of the second year, customers acquired in the company's first month were spending 47% more at top-decile startups than they were initially, about twice the increase seen at middle-decile startups. It also reports that by the sixth month top-decile solo founders began winning back churned customers, roughly three months sooner than middle-decile founders. Both are correlational observations in a United States population, not forecasts for our business or the reader's.
Hypothetical example
The observation record needs six lines per customer. Customer name or identifier. The kind of work agreed in the first month. The window you chose. Whether a paid request of the same kind came from the customer's side within that window, and the date if so. One line giving the reason if it did not. The repair candidate that reason points to, or unconfirmed. Lay this out once per first-month customer and you can decide whether the next spend goes to acquisition or to what you deliver from a record rather than from instinct. This is a proposed form to fill in, not a record of any particular customer.
What this article does not decide, and the next step
From the sources
In the same discussion, Hariharan said it is more important to benchmark whether your stable retention is good versus benchmarks, or better than benchmarks, before you start focusing on growth.
A&A perspective
There is, however, no benchmark available here. Stripe's figures are an observation of a population of solo-founded companies incorporated as United States entities through Stripe Atlas in 2022 and 2023, comparing the top and middle of revenue outcomes after the fact: a correlation. Do not use roughly 30%, or 8%, as a target or a passing grade for a Japanese contract or small service business. The Y Combinator discussion is from 2017 and assumes product companies and SaaS with enough users to draw a cohort curve. Substituting "came back" with "the next request of the same kind" is our hypothesis. We present no retention rate of our own and no customer results in this article.
A&A perspective
There are also situations where this reading does not apply. Work that completes once, such as a single migration or a one-off piece of research, has no natural recurrence. Because repeat value and contract renewal are not the same thing, for one-off work you should look first at whether acceptance passed, whether the handover held, and whether a referral came out, rather than at whether anyone came back. If your business does have a cycle, you can start by deciding the list of first-month customers and the window on paper. If what remains is the judgement of which repair candidate the evidence points at, you are welcome to use our advisory on that single question. The overall map is set out in "AI-native GTM: a practical guide for solo founders and small teams."
The metric to look at before you scale acquisition is whether your first-month customers issued a request of the same kind, on their own, in the next cycle. While you have only a few customers, do not turn it into a rate: keep a named list and one line of reason for each customer who did not come back. Fix the window and the definition of a return in writing before you look at the number. If you use recurring billing, observe actual usage separately from a successful charge. The figures in the sources are correlational observations of a United States distribution, not targets.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- Solo founding is at an all-time high: Top performers have these traits in common
Stripe · 2026-05-28
Accessed 2026-10-01 - Growth Office Hours with Anu Hariharan and Gustaf Alstromer
Y Combinator · 2017-12-01
Accessed 2026-10-01
AI-assisted editorial production
A&A uses AI for research, writing, translation and editorial checks. Source facts, our analysis and hypothetical examples are labeled separately.
Editorial check: 2026-10-01