Your first ten customers: a founder-led sequence

The first ten customers are not a smaller version of the first thousand. They are a different job. At ten, you are not running a funnel; you are running a series of conversations, and every one of them should change something about the product, the pitch or the price. This article lays out a sequence the founder can run by hand, in order, and explains why doing it by hand is the point.

Why the founder does this, and not a channel

Paul Graham’s essay “Do Things that Don’t Scale” makes the case plainly: early users do not arrive by themselves, so founders go and get them. The essay is a useful reminder that the manual work is not a phase to rush through. It is where a founder learns which sentence makes a stranger lean forward, which objection comes up in every call and which feature nobody mentions.

A hired salesperson or an automated campaign removes that learning from the founder’s hands. Delegate later. For the first ten, the person who can change the product should be the person in the room.

The five steps

The sequence below is deliberately short. Each step has an output you can open and inspect the next day.

Step 1: name the buyer

If you cannot name the buyer in one sentence, you will not know where to look for them. “Small businesses” is not a buyer. “Bookkeepers at firms of under ten people who reconcile client accounts every month” is a buyer, because you can find such people, ask what they do today and tell within minutes whether they match.

Write the sentence on a page and keep it where you can see it. When a conversation shows that the real buyer is someone else, change the sentence and note the date. The history of that sentence is a record of what you learned.

Step 2: list forty names

Ten customers rarely come from ten attempts. Plan for a list several times larger than the target, because most people will not reply and many who do will not match. For each name, write down where you found them. That one habit has two benefits: you can see which source produces matches, and you can avoid contacting the same person twice from two directions.

Good sources are the places where your buyer already talks about the problem: professional communities, forums, newsletters, event attendee lists that are public and conference talk Q&As. Be honest in how you reach out. Do not pretend a relationship exists.

Step 3: ask for a conversation, not a sale

The message should be short, specific and easy to decline. Say who you are, what problem you are studying, why you are writing to this person and what you are asking for. Do not attach a deck. Do not ask for a “quick call to pick your brain” with no stated topic.

In the conversation itself, follow the discipline described in our article on customer discovery interviews: ask about what the person did the last time the problem occurred, not whether they would use your idea. The goal at this stage is to find out whether the problem is frequent and costly enough for someone to change their behavior.

Step 4: offer a working version

Graham’s essay describes founders doing the setup work for early users themselves. The principle holds for a SaaS product: if a prospect confirms the problem, offer to do the first task for them inside your product, by hand if necessary. You may import their data yourself, configure their account or write the first report manually behind a thin interface.

This does three things. It shows you where the product’s first-run experience breaks. It gives the prospect a result before they have learned your interface. And it makes the commitment question in step 5 concrete, because the person has seen their own work come out the other side.

Step 5: ask for the commitment

A kind word is not a commitment. Decide in advance what counts: a payment, a signed pilot agreement, a written purchase intent with a date or a named colleague joining the next call. Then ask directly. Record the answer in the commitment log, including the refusals and the reasons given.

The log needs only a few columns: name, source, date of first contact, problem confirmed (yes, no, partly), what you did for them, what they committed to, and the next dated step. Review it weekly. A pattern in the “no” column is usually more useful than a pattern in the “yes” column.

Keep a weekly number, and keep it honest

“Startup = Growth” argues that a startup is defined by its growth rate and that a founder should pick a measure and track it weekly. You do not need the essay’s specific targets to use the idea. Pick one measure that reflects paying customers or committed pilots, write it down every week and do not change its definition when the number disappoints.

At ten customers, the right measure is small and countable: paying customers, signed pilots or committed dated next steps. Percentages of a tiny base are noise. If you have three customers and add one, you grew by a third and learned almost nothing about the market. Keep the raw counts beside any rate, and say how few observations sit underneath.

Common ways this goes wrong

• Selling to friends. A supportive friend is a poor test. Count a friend as a customer only if they have the problem and would pay a stranger’s price. Note them separately in the log.
• Discounting to close. A heavy discount may win a customer while hiding whether the price is wrong. If you discount, record why, and treat the discounted customers as pilots rather than evidence about pricing.
• Skipping the log. Without a written record, the last three conversations dominate your memory. The log corrects that.
• Building for one loud customer. A request from one customer is a data point. A request from four of the ten is a pattern.
• Hiding the manual work. Doing tasks by hand is fine. Misdescribing it is not. Tell customers what is automated and what is not. That honesty is part of building trust.

What this article does not know

This sequence is a method, not a forecast. It cannot tell you how many conversations your market will need, how long a sale to your buyer takes or whether your buyer pays at all. Those depend on your product, your price and your category, and no general rule replaces your own log. We have not included conversion benchmarks because we have no public primary source that would apply to your buyer, and an invented benchmark would do more harm than good.

A short checklist for the week

1. The buyer sentence is written and dated.
2. The name list has at least forty entries, each with a source.
3. You have sent this week’s messages and logged the replies.
4. At least one confirmed prospect has a working version set up for them.
5. Every conversation that ended has a logged commitment or a logged refusal with a reason.

When this checklist is routine, you are ready to ask what should be automated, which is the question the next stage answers. For the work around it, see the earn your first customers playbook and the customer interview guide, linked from the related items on this page.

In the data

These figures describe U.S. businesses in general, not WE’s results. Each one names its source.

Show the data as a table

Share of U.S. employer jobs, by the age of the firm, 2023

Firm age	Share of jobs and of firms, 2023
Age 0 to 5	10.7% of jobs, 36.2% of firms
Age 6 to 10	7.8% of jobs, 16.4% of firms
Age 11 to 15	6.8% of jobs, 10.9% of firms
Age 16 to 25	12.4% of jobs, 15.9% of firms
Age 26 and older	24.2% of jobs, 15.4% of firms
Founded before 1977 (exact age not measured)	38.1% of jobs, 5.4% of firms

The Census Bureau cannot date firms born before 1977, so they are shown on their own; every one of them was at least 45 years old in 2023.

Source: U.S. Census Bureau, Business Dynamics Statistics, firm age , 1978 to 2023

Retrieved 2026-10-16. Each share is the sum of the published employment (or firm) counts of the age classes in the group, divided by the sum over all classes, which equals the economy-wide total in bds2023.csv.

Sources

1. Do Things that Don't Scale, Paul Graham . Accessed October 6, 2026.
2. Startup = Growth, Paul Graham . Accessed October 6, 2026.

Spotted an error? Read how we handle corrections in our editorial policy.