Sales and go-to-market · Answered by Golden Section from more than 400 B2B software companies observed
Build it from your best existing customers rather than from everyone who could buy. Score every current and past customer on a written definition of quality, including ACV, expansion, churn, acquisition cost and cycle length, support load, custom work and how clean the ROI story is. Test at least ten segmentation hypotheses against those scores, keep the two or three traits that cluster with quality, and check that the resulting segments add up to at least 25% of your serviceable market. Then write the buyer persona and the negative persona you will stop pursuing. Start by exporting every customer with its ACV, churn and expansion history.
An ICP is right when it predicts customer quality in your own data and still leaves room to grow. Narrow until the traits cluster with your best customers; stop narrowing when the segments fall below roughly a quarter of your serviceable market.
| Metric | Value | What it means | Source |
|---|---|---|---|
| Segmentation hypotheses | at least 10 | candidate traits tested against customer qualityCustomer count should significantly outsize the number of hypotheses | Golden Section playbookCustomer Segmentation play |
| Target segments | 2–3 | segments with the strongest relationship to customer qualityOutput of the segmentation play | Golden Section playbookCustomer Segmentation play |
| Minimum market coverage | at least 25% of SAM | sum of chosen segments as a share of serviceable addressable marketBelow 25%, reconsider the chosen segments | Golden Section playbookCustomer Segmentation play |
| Verticals to pursue | 1–2 | verticals chosen from 4–5 scored product/market combinationsScored 1–5 per factor; favor high-probability, low-complexity verticals | Golden Section playbookVertical Specific play |
An ICP is a bet about where your sales, product and service effort earns the most durable revenue, and your own customer base is the best evidence you have. The customer segmentation play turns that evidence into a quality score and tests which traits predict it. Watch the relationship between customization and ACV: segments that need heavy customization at low contract value are what the play calls death valley, revenue that costs more to serve than it returns.
An ICP that is too broad shows up in operations before it shows up in the numbers. A crushing product backlog with deadlines from customers who need different things is the sign the vertical specific play names, and sales efficiency falls because messaging has to fit everyone. Too narrow is easy to test with the 25% rule.
The ICP then has to reach the people doing the selling. Buyer personas describe the person who buys inside each segment, and core admission tests keep accounts that fail the profile out of the revenue line your valuation rests on.
A company selling scheduling software to clinics has 140 customers across dental, veterinary, physical therapy and small hospitals. Scoring each on ACV, expansion, churn, support load and sales cycle, it finds the top quartile is almost entirely multi-location veterinary and physical therapy groups with three or more sites, while small hospitals sit in the bottom quartile on support and custom work. Those two segments cover about 30% of its serviceable market. It narrows marketing and qualification to them and routes new hospital prospects into a paid pilot. All figures are invented for illustration.
Very early companies without enough customers to find patterns should use customer interviews and the projected scorecard in the vertical play, then revisit once the data exists. A strategic account outside the ICP can still be worth taking if it is kept out of the core revenue line.
From the Golden Section mistakes list, each paired with the play that prevents it.
A product built for everyone is harder to sell to anyone, which is what an over-broad ICP produces.
Switching target customers before messaging and channels have had time to work resets the evidence to zero.
Adjacent markets behave like a new startup, so expanding the ICP sideways is rarely the cheap growth it looks like.
Spreading sales and product effort across many segments dilutes the force needed to win any one of them.
In the order we would run them. Each is on its own page, most with a free Excel template.
Chooses the one or two verticals where your product solves the most acute pain.
Scores customer quality and finds the segments that predict it.
Describes the buyer inside each segment, plus the negative persona to avoid.
Turns the ICP into admission tests that keep off-profile accounts out of core revenue.
Sales & marketing plays The go-to-market plays run in order from vertical to segment to persona to value proposition, and the ICP is their shared foundation.
Probably, if your product backlog carries conflicting deadlines from different customer types, win rates vary widely by segment, or your sales team cannot state who not to sell to. If a disqualification rule does not exist, the ICP is too broad to guide anyone.
Narrow enough that two or three traits clearly separate your best customers from the rest, and wide enough that those segments still cover at least 25% of your serviceable market. As our mistakes list puts it, we have never seen a product that was too small.
Score four or five product and market combinations on qualitative and quantitative factors from 1 to 5, weigh how many things must go right in each, and choose one or two with high probability and low complexity. Then ask what would make that vertical fail in hindsight before committing.
A defensible position in one industry is one of the first things we look for as investors, and we pass on horizontal platforms without it. Minority equity fits when the ICP is proven and the next step, such as a second vertical, is a real change to fund.
Growth equity →Reviewed by Dougal Cameron, CEO & Co-Founder on 2026-09-23. Golden Section observations are labeled separately from external benchmarks and illustrative arithmetic.