ICP for Startups: Define One With Few Customers

A startup with few customers should write its ICP as a labelled hypothesis: describe the customers you have, add evidence borrowed from site visitors and competitor case studies, pick one narrow starting segment, and revisit it after every ten deals. Precision comes from iteration, not from waiting for more data.

The early-stage problem

The standard advice, analyse your last 20 wins and 20 losses, assumes you have 40 deals. A seed-stage company may have three customers and a pipeline of conversations. The temptation is either to skip the exercise or to describe a dream customer. Both create the same result: a vague profile that never guides a decision.

BusinessMCP’s ICP guide tackles this directly. It says to derive the ICP from real wins and losses, borrow evidence from site visitors and competitor case studies when customers are few, and revisit the profile after every ten deals for early-stage teams. It is a vendor guide, but the advice is sensible and cheap to follow.

Step 1: start from the customers you do have

Write a short note on each customer, even if you have only two or three.

• Why did they buy, in their words?
• Who inside the company drove it?
• What were they doing before?
• Did they pay without heavy discounting or personal favours?

Rank them by how repeatable they look. A customer who found you through search and signed within a week is better evidence than one who is a former colleague of the founder.

Step 2: borrow evidence responsibly

With little first-party data, outside evidence fills the gaps, as long as you label it.

Source	What it can tell you	Caveat
Your site visitors	Which companies show interest	Interest is not purchase
Competitor case studies	Who competitors win	Vendors publish selectively
Job postings	Who is hiring for the problem you solve	Postings can mislead
Community threads	The language and pain used	Small, self-selected sample
Sales calls that did not close	What blocks deals	Biased toward what people say

If you want to see which companies visit your site, BusinessMCP publishes a page on website visitor identification; check its documentation for what it resolves and its limits before relying on it. Its sales-triggers guide claims 20 to 35 percent of B2B traffic resolves to named companies, which is the vendor’s figure, so expect your own rate to differ.

Key takeaway: in the early days, mark every ICP field as observed (from a customer) or assumed (from outside). Convert assumptions to observations as deals arrive.

Step 3: choose one narrow segment

Resist the urge to describe a large market. Pick the smallest segment where you can name 100 real companies and reach their buyers directly. Messaging is easier, learning is faster and mistakes are cheaper. When the first segment works, add a second profile; the guide suggests one to three in total.

A narrow segment still needs hard edges. Write three disqualifiers even now, based on the conversations that went nowhere; see ICP disqualifiers.

Step 4: draft quickly, with a tool if useful

Spend an hour, not a week. Fill in the ICP template, or answer the six questions in the free BusinessMCP ICP generator, which the page says needs no signup and keeps answers in your browser. The output is a draft with a buying committee table, qualification criteria, disqualifiers and a scoring skeleton. For an early-stage team, that is exactly the right level of ceremony.

Step 5: set a fast review rhythm

Early profiles go out of date quickly. Put a reminder in the calendar for every ten closed deals (won or lost). At each review:

1. Add the new deals to the spreadsheet.
2. Re-rank traits shared by wins and losses; see how to create an ICP.
3. Promote assumed fields to observed ones where evidence arrived.
4. Remove criteria that no longer separate wins from losses.
5. Note the date and what changed.

Founder-led sales: a special case

In founder-led sales, the founder’s charm can win deals the product would not. Be honest about it. Record which wins relied on personal relationships and treat the profile for those as separate from the repeatable one. If a future sales hire cannot win in the segment without the founder, the segment is not yet a real ICP.

Pitfalls specific to startups

• Chasing logos. One impressive customer does not make a segment.
• Pivoting without updating the document. Product changes should trigger a profile review.
• Over-fitting to three data points. Hold conclusions loosely.
• Ignoring pricing. If every customer needed a heavy discount, your profile may be aimed above your real market.
• Skipping the loss side. Even five lost conversations reveal patterns.

When your pipeline grows, move from hypothesis toward evidence by following the full process, and build a lightweight scoring model so rep time goes to the accounts that look most like your best customers.

A lightweight ICP for the first quarter

If you want something you can write this week, use this five-line format and expand later.

1. Who: the narrowest company type you can name 100 examples of.
2. Why now: the event that makes them look for a solution.
3. Who buys: the title that raised the problem and the title that signs.
4. Proof: the customers or conversations behind each claim, marked observed or assumed.
5. Skip: three types of company you will not pursue this quarter.

Share it with anyone who talks to prospects, whether that is a salesperson, a founder or a customer success lead, and ask them to challenge line 1 and line 5. Those two lines drive the most behaviour. Then run a four-week test: send outreach or run a campaign aimed at that segment only, count replies and booked calls, and compare with last month’s broader effort. The result will not be statistically perfect, but it will tell you whether the hypothesis deserves another round.

When to graduate from the lightweight version

Move to the full process once you have around twenty closed deals, a repeatable acquisition channel or your first sales hire. At that point a new person is relying on the document, and vague wording that worked inside a founder’s head will fail. Build a scoring model, document the buying committee and schedule quarterly reviews. Keep the early notes as history; they show how the profile evolved and why.