What is an ICP (ideal customer profile) and how to build it from your own sales

30 September 2026

The ICP, ideal customer profile, is the portrait of the customer who brings you the most and costs you the least. In theory you write it in a workshop with the sales team: "installers with 10 to 50 staff, around Paris". In practice that workshop produces an opinion. Your orders contain the measurement, and it often contradicts the opinion.

The definition that holds

An ICP is the set of traits over-represented among your best customers compared with all your contacts. Not "the traits of your best customers": if 80% of your base is electricians, your best customers will mostly be electricians, and that teaches you nothing. What matters is the ratio between a trait's share among the best and its share in the base. That is called lift.

A real example

Security equipment distributor, 3,624 companies whose NAF code (the French industry code) is known from the registry, 164 of them in the top 20% by lifetime value. So 4.5% "best customers" in the base, the reference rate.

• 43.21A, electrical installation: 1,234 companies, a third of the base. 4% of them are best customers. Lift 0.9: the norm, not a signal.
• 80.20Z, security systems: 344 companies, 5% best customers. Lift 1.2. The "obvious" trade for a security distributor is barely above average.
• 42.22Z, construction of electrical and telecom networks: 83 companies, 12% best customers. Lift 2.6.
• 46.52Z, wholesale of electronic and telecom equipment: 73 companies, 11%. Lift 2.4.
• On size, companies with 1,000 to 1,999 staff: 62 in the base, 11% best customers, lift 2.4. The 10 to 19 staff band, the supposed core target: lift 1.2.

This distributor had done the workshop, a few months earlier. A grid in a spreadsheet: nine customers rated 1 to 3 on eight criteria (revenue, order frequency, premium brands, loyalty, potential...), and a profile sheet written from the ratings. The sheet's verdict: "low-voltage installer, 10 to 20 staff, 2 to 3 million revenue, regional". The orders say "network builders, wholesalers, and large accounts". And the grid itself rated the company's 7th and 11th largest customers "off target", below 60%. A method that puts your seventh customer off target is measuring something other than value.

How to build it, in five steps

1. Define "best customer". The top 20% by lifetime value, contacts rolled up to their company, on confirmed orders. An arbitrary threshold but an honest one, and it recomputes itself. The LTV calculation is here.
2. Describe every company with comparable traits. Odoo only has the name, the city and sometimes a hand-typed industry. The company registry, free in France, gives for each SIREN the NAF code, the headcount band, the creation date, the number of sites, the legal form and the published revenue. You first have to find the SIREN from the name and city, which works in 85 to 90% of cases.
3. Compute the lift of each value. Share among the best divided by share in the base. With one correction: on 3 companies, a lift of 3 means nothing. Shrink toward 1 when the count is small, so a rare trait does not dominate the profile by accident.
4. Group traits into axes. NAF code and collective agreement say the same thing (the trade). Headcount, INSEE category, revenue and number of sites say the same thing (size). Adding all six counts size four times. Blend inside each axis, then weight the axes: trade, size, maturity, growth, geography.
5. Score every lead, and treat unknown as neutral. A lead whose headcount is unknown must be neither punished nor favoured: its score on that axis is 50, the average. Then bring every axis onto the same scale, or the axis with the most variance decides on its own.

What an ICP must not contain

Anything a prospect does not have yet. Revenue made with you, order count, number of brands bought: those traits describe your customers, not the companies that could become one. Putting them in the profile means scoring prospects on things they cannot have, and pushing existing customers to the top of every list. We tested brand breadth as a criterion: over two years of conversions it added nothing. It is useful elsewhere, for growing existing customers.

How to know whether the profile is any good

One test only: does the profile computed on data before a date rank at the top the leads that actually converted after that date? At this distributor, eight 90-day periods, 160 converted leads: the top 10% of leads by score convert 2.9 times more than average. A workshop profile has never been tested this way, because it cannot be computed.

How ERP-BI does it

ERP-BI enriches every company from the registry, recomputes the profile every night on the top 20% of customers, scores every lead and customer (Fit: who they are; Intent: what they are doing right now), and searches the registry every week for companies that match the profile without being in your Odoo yet. The axis weights are visible and editable in the ICP configuration, and the page shows each trait's lift, so the profile stays something the team can argue about.