Ideal customer profile

An ideal customer profile should come from the companies that already bought, not from a room full of people guessing who ought to buy. The quickest way to make an ideal customer profile useful is to compare your closed-won accounts, find the patterns they share, and use those patterns to choose which companies deserve attention.

That sounds obvious. Most teams still get it backwards.

They start with a market they like, add a few traits that sound sensible, and call the result an ICP. "US-based B2B companies with 100 to 1,000 employees" is not much of a profile. It's a filter. It doesn't tell a rep why one account is worth contacting this week and another nearly identical account is not.

Build an ideal customer profile from buying evidence

Start with your won deals. Not your biggest logo. Not the accounts your founder wants. Pull a reasonable sample, such as the last 20 to 40 closed-won customers, then look at what those companies looked like before the sales cycle began.

You're looking for repeated conditions. Maybe the wins are mostly regional logistics firms with 150 to 600 employees. Maybe they've recently opened a second warehouse, hired a VP of operations, or replaced an outdated system. Maybe they sell through channel partners and have a small finance team that can't keep up with manual reporting.

The trigger often matters more than the company label.

For example, suppose a 12-person sales team sells workflow software to US manufacturers. Its first ICP says "manufacturing companies with 200 or more employees." After reviewing 28 wins, the team finds something more useful: 19 customers had opened a new facility in the previous nine months, and most had hired an operations systems manager during that period. The winning accounts weren't simply large manufacturers. They were manufacturers dealing with a specific change.

That changes the outbound work. Instead of building a giant list of manufacturers, the team can look for companies opening facilities, hiring for operations systems, or announcing a new production line. That's a real selection rule. It gives a rep something to investigate and a reason to write now.

My view is blunt: teams get too attached to attractive segments and not attached enough to disconfirming evidence. If a segment looks perfect on paper but rarely closes, it shouldn't stay in the ICP because everyone likes the story.

What belongs in the profile

An ICP usually combines three kinds of information.

First, there are company facts: industry, employee count, revenue range, geography, ownership, business model, and technology in use. These help narrow the market, but they rarely explain the purchase on their own.

Second, there are operating conditions. A company may be expanding locations, moving upmarket, adding a new sales channel, replacing a key system, or dealing with a compliance requirement. These conditions often create the pressure that turns a possible buyer into an active one.

Third, there are observable signals. Hiring activity, leadership changes, new locations, acquisitions, product launches, funding events, or changes on the company website can indicate that the operating condition is real and current.

Don't force every field into a neat table. A useful profile might say:

Our strongest customers are privately held US logistics companies with 150 to 600 employees that have added a facility or regional operation in the past year. They usually have a small operations systems team, use separate tools across locations, and are hiring for process or systems roles.

That description is more useful than a row of filters because it tells sales what to verify. It also tells marketing which accounts may not be worth targeting, even if they match the employee range.

Keep the data honest

An ICP can only be as good as the records behind it. If half your customer records have stale employee counts, inconsistent industries, or the wrong headquarters location, you may find patterns that aren't real.

This is where company data collection matters. Workloom crawls and enriches companies instead of simply reselling a fixed database. Its discovery process searches across four engines, map data, aggregator sweeps, and waterfall discovery. If one path fails, another can try to resolve the company record.

The collection layer also uses two browser engines for different conditions. The light engine strips out 20 tracking domains and moves a cursor along human curves. The hardened engine uses persistent profiles, pooled sessions, and device-fingerprint masking. Those mechanisms help a source resolve when a page behaves differently or blocks an initial request. They don't decide whether a company belongs in your ICP. That still requires judgment.

The pipeline separates discovery, writing, and normalization. Thirty-five discovery workers find and gather company information. Fourteen writers turn that information into structured fields. Thirteen normalizers then settle the output into one canonical company record. Writers and normalizers are separate stages after the workers, not part of the worker count.

That separation is useful when sources disagree. One page may report 240 employees while another says 310. A company may list offices in three cities or appear under two names. Conflict detection and confidence scoring make those disagreements visible instead of silently keeping whichever value arrived last.

Don't hide uncertainty from the sales team. A contested employee count may not matter for one campaign, but a disputed industry or ownership field could completely change account priority. False precision is worse than an incomplete record because it gives people confidence they haven't earned.

Turn the ICP into a working filter

Once the profile is based on real wins, use it to prioritize accounts, not to declare who will buy.

A rep should be able to look at a company and answer three questions: does it resemble our best customers, is there evidence of the buying condition, and is the evidence current enough to act on? If the answer to the second question is always "we don't know," the account may fit the firmographics but still be a weak prospect.

Keep the profile separate from the target account list. The ICP describes the type of company. The list contains specific companies that appear to match it. Mixing them creates a predictable problem: the team chooses familiar accounts first, then writes an ICP explaining why those accounts are good.

Instead, derive the profile from won deals, apply it to current company data, and inspect the resulting accounts before outreach. A 220-person logistics company that just opened a warehouse is not equivalent to a 220-person logistics company that has made no visible change in three years.

The profile should also change when the evidence changes. A new customer outside the current ICP might be an exception, a new segment, or a sign that the profile is too narrow. A string of losses in a supposedly strong segment is evidence too. Don't explain it away just to preserve the original definition.

The useful test is whether each major part of the profile can be traced to specific won accounts and reliable company records. If the team can't show that trail, it's still a hypothesis, no matter how polished the document looks.