Segmentation

Leave a market uncut and every number comes back as an average. The trouble with an average is that nothing real is standing at that value. A survey reporting 3.2 out of 5 on satisfaction with settings may describe no user at all, because it's two groups added together. Cut it and one side is at 4.1 while the other is at 2.3.

In meetings this shows up as contradictory requests colliding. One person arrives with feedback that the settings are too complicated. Another arrives with a request for more settings. Both are reporting real conversations, so neither backs down. And the decision that comes out of that meeting is usually a compromise — a few more settings, tucked into a submenu — which is the one outcome neither group asked for. A company of fewer than 20 people wants less to configure, and a company of 5,000 wants permissions it can slice finely — and looked at as one block, those two only appear to want opposite things.

Segmentation is the act of cutting that block. Instead of arguing over which piece of feedback is correct, you start by accepting that two groups want different things, and only then choose which group gets matched.

Pick the lines you might cut along first, then check whether the groups they produce actually behave differently. Choosing a cut and stopping there produces labels and leaves every decision where it was. Six steps.

1. List the candidate lines. Company size, industry, which tools they already use, how they handle documents, which channel brought them in.
2. Split real data along each line. Pull retention — how many users keep coming back — plus free-to-paid conversion and the mix of incoming requests, separately for each group.
3. Keep only the lines where the numbers move apart. If splitting by industry gives retention of 62 percent and 59 percent, that line didn't cut anything. If splitting by company size gives 41 percent and 78 percent, the line is real. What you are testing is the gap, not the cut: a line you can draw in the data is not the same as a line the market is standing on either side of.
4. Reduce to somewhere between three and five groups. Twelve looks more precise, but nobody has the capacity to build for twelve.
5. Write a name and a condition for each group. The condition has to be sharp enough to sort an incoming inquiry on the spot. "Companies that write a lot of documents" is not a condition. "Software companies under 50 people" is.
6. Size the groups. TAM, SAM, SOM — the total market, the part you can reach, and the part you can realistically win — gets used at this step. A group that turns out to be 3 percent of the whole isn't worth managing separately.

That covers the cutting. Choosing comes next. Writing down the conditions of the customer who fits best and stays longest gives you an ideal customer profile (ICP), and picking the one square to take first gives you a beachhead market. Cut without choosing and you end up facing five groups at once and finishing first in none of them.

Three signals say it's time. Contradictory requests keep colliding in meetings. The average metric hasn't moved in months while something inside it seems to be splitting. Sales needs a shorter list of companies to call. In all three, one average doesn't determine the next action.

It isn't needed on a product with thirty customers. Someone on the team remembers every company by name, so there's no reason to build groups. Hold off, too, when there's no data to split yet. A line drawn from guesswork gets quoted from the day it lands in a document until something proves it wrong, and six months of usage data settles more here than a workshop full of assumptions.

Three limits come with it.

1. Cutting creates work. With five groups, changing one screen means thinking it through five times, and the onboarding copy, the pricing tiers, and the sales deck split along with it. That cost is why step 4 exists — the right number of groups is the number the team can actually serve differently.
2. A line drawn on company traits doesn't reach the people. Inside one customer, an administrator and an everyday user want different things. Seeing that layer takes a persona, a model that turns a group of users into one concrete person.
3. Lines go stale. A gap that split cleanly by size a year ago may have closed since, so set an interval for pulling the same data again and checking.

"That number is an average across everyone, so it points at nobody. Can you split it by company size?"

An average can't determine which group to build for or what to do next.

"I split by industry and the two groups had almost the same retention. Let's drop that line."

A split that produces no difference labeled the market rather than cutting it.

"With twelve groups there's nothing we can actually tailor."

The number of groups has passed what the team can carry.

Pagedeck split its customers by size in year two. Mia pulled retention separately along four candidate lines. Industry and acquisition channel produced no gap at all. Only the line drawn at 50 employees opened up, with retention at 78 percent on one side and 41 percent on the other. That one line stayed, and the groups were set at three.

Since then Ravi writes ad copy per group. For the smaller companies the first line is that there's nothing to configure. For the larger ones it's that permissions can be divided. Same product, different opening sentence.

The most common mistake is reading segmentation as demographic sorting. Survey forms usually ask for age, region, and job title first, so those fields look like the lines. But the purpose of cutting is to find where behavior diverges, not to arrange groups neatly. If people in their thirties and people in their forties use the product at the same frequency and leave for the same reasons, that line has divided nothing. Miss the distinction and a few pie charts end up in a document while product decisions run exactly as before.

The second is believing that more groups means more precision. Cut into twelve and each group's character does get sharper, which reads as better analysis. But a cut only means something once it's used, and using it means a different screen, a different message, or a different sales motion per group. A 45-person company cannot produce twelve of each. Cut past what the team can carry and the document says twelve groups while the actual behavior goes back to treating everyone as one.