An ideal customer profile from your best customers

An ideal customer profile (ICP) describes the companies where you win, where clients stay and where the work pays. It is built from your own deal history, not from the size of the market. A list of everyone is a list of no one: five hundred companies with over fifty staff tells you how big the field is, not where Klasna for Teams wins.

Teams has a small but real history: forty companies since 2024, twelve won, six still clients. That's where the profile starts. The question is not "who is biggest?" but "which clients would we clone?": the ones who bought fast, use their lessons, stayed, and cost little to serve.

Two shortcuts mislead. Starting from industries the founders find interesting builds a profile around opinion. Starting from the biggest companies ever contacted assumes one large client is worth ten small ones, and ignores whether that client stayed and what serving it cost. Both feel ambitious; neither predicts where the next win will come from.

A profile is useful only if its criteria can be checked from the records of past deals, not from a hunch. For Klasna for Teams, six criteria cover what matters:

• Industry.
• Size, and how many people need English.
• Why they need it: part of the job, or a perk?
• Deal size: lessons a month.
• How fast they bought: days from the first call to a signed contract.
• Cost to serve: coordinator hours and paid no-shows each lesson carries.

Each criterion is a question about a real deal. That's what makes the profile testable: a new lead either matches a line or it doesn't, and after a quarter you can see whether the matching leads really won and stayed.

Industry and size alone are the usual first draft, and the weakest part of it. Two companies in the same industry and of the same size can be very different customers: one needs English for daily calls with foreign clients, the other offers it as a benefit nobody uses.

Of the six criteria, why they need it says the most about whether a client will stay. A company that needs English for the job, such as an IT firm whose engineers talk to foreign customers every day, keeps paying because the need doesn't go away. A company that buys English as a perk is buying a benefit, and benefits are the first thing cut when a budget tightens.

Teams' history shows it plainly. All three retail chains bought English as a perk for staff, and none of them is still a client. The IT services clients use English in calls with foreign customers; it is part of their work, and that need survives a bad quarter.

So a large contract doesn't make a perk buyer a good fit. Size makes the first invoice bigger, but if the reason to buy is weak, the client leaves sooner and the effort of winning it is lost. In the profile, the "why" line separates a work task from a nice-to-have, and it should be checkable on the first call: "What do your people use English for today?"

Delivering a lesson costs Klasna about the same everywhere: 430 UAH to the tutor and 31 UAH for video and AI. Serving a client does not. Some clients eat coordinator hours: rescheduling shift workers, preparing separate reports for each store, and paying tutors for lessons learners didn't show up to. That is the cost to serve, and it differs a lot from one client to another.

That's why it belongs in the profile. Two clients with the same price per lesson can leave very different amounts once service is counted. A client with scattered shift workers and frequent no-shows can look healthy on revenue while the work around each lesson quietly takes most of what's left.

A profile that ignores cost to serve tends to favor big, busy accounts, because their revenue is visible and their service cost isn't. Asking about it early, such as how schedules are set and who books lessons, shows whether a lead will be easy or expensive to keep.

A profile also says whom you won't chase. Disqualifiers are conditions that make a deal a bad bet however big it looks. Each one has to come from the reasons behind lost or unprofitable deals, not from a guess.

Teams' losses rhyme: English wanted only as a perk, with no work task behind it; nobody owning the training budget, so the deal faded; a deep discount demanded before any scope was agreed. Those three patterns are sound disqualifiers.

The trap is turning an industry into a disqualifier. Teams lost all four logistics deals it pursued, yet Veloria is a logistics company. Four losses say "we haven't cracked it yet", not "never". The reasons matter: shift schedules twice, no budget owner once, and last spring procurement renewed the incumbent. Screen for those reasons, not for the label. Size can mislead the same way: "fewer than five hundred staff" isn't a disqualifier when most clients who stayed are well under that. Veloria becomes a bet outside the profile: a paid pilot, a known approval path, a schedule built around shifts.

Strong opinions about the "right" customers are common, especially from founders. Treat each one like any other claim: check it against the rows. Leo is sure big retail is Teams' future: thousands of staff, one contract. The deal history says three retail clients, all three buying a perk, none still a client. A profile built on who sounds impressive repeats old mistakes.

This doesn't mean ignoring Leo. It means putting his idea next to the evidence, and if he still wants to try retail, treating it as a bet outside the profile with its own conditions, not as a profile line.

The opposite problem is thin data. Twelve wins isn't much, and some profile lines will rest on two or three deals. Say so: mark such a line "confidence: low", and review the profile every quarter against new wins and losses. A profile is a hypothesis you keep testing, not a monument. Honest confidence marks also tell sales which lines to trust when they decide where to spend a week.

Building a first ICP follows a simple order, and each step guards against a specific mistake:

1. List every won deal and mark who stayed and who left.
2. Find what the clients who stayed and pay off have in common.
3. Check those patterns against the lost deals and their reasons.
4. Write the profile lines and the disqualifiers.
5. Mark how confident you are in each line.
6. Review the profile against new wins and losses every quarter.

Step 3 is the one most often skipped, and the one that matters most. A trait shared by both wins and losses predicts nothing: if Teams' won and lost deals are both mostly companies with English-speaking staff, "has English-speaking staff" can't tell a good lead from a bad one. Only traits that separate the two belong in the profile.

Confidence comes after the lines exist, because you can only judge how much evidence supports a line once it's written. The quarterly review keeps the profile honest as the data grows.

A profile is for use, so its lines must let someone decide before the first call whether a company fits. A useful line names an industry, a size and a work reason you can check: "IT services firms, 60 to 300 people, where 15 to 60 of them talk to foreign clients in English", or "design and product studios of 15–60 people where staff present work to foreign clients in English".

Weak lines fail the check in different ways. "Forward-thinking companies that invest in their people" can't be verified: every company says that about itself. "Any company that has ever asked about English lessons" is not a profile at all. "Large companies, because they have big budgets" repeats the size bias that the deal history argues against.

The finished product is one page: the profile in five or six lines, the disqualifiers, the bets outside the profile, and a confidence mark on each line. That page answers Marta's request for something usable on Monday, not a strategy deck.

A profile earns its keep when a lead arrives from outside it. When a big retail chain calls anyway, don't answer from excitement or from the profile alone: check it against the disqualifiers. A real work need and a budget owner who can describe the approval path make it worth a careful look, perhaps as a bet with its own conditions. A perk and a discount request before any scope: say no politely.

Saying no to a big name feels wrong. But a month spent on a client who won't stay is a month not spent on one who will. That trade is what the profile protects.

How it fits together: an ICP starts from the clients worth cloning, uses six checkable criteria, takes its disqualifiers from the reasons behind losses, keeps opinions next to the data with honest confidence marks, and fits on one page sales can use. Next lesson turns these criteria into a weighted score, and shows why the biggest client isn't always the best one.