The users who erase everything and keep coming back

There's a user in your product who logs in three times a week, creates something, deletes it, and leaves. Their account looks empty. Your engagement summary calls them inactive or at-risk. But they've been doing this for nine months.

We've been looking at patterns where deletion and undo activity is high relative to creation — not in the aggregate, but per user over time. The pattern we expected to find was confusion: people who couldn't figure out the interface, made mistakes, backed out. That pattern exists. It's real, and it looks like short bursts of create-delete-create-delete clustered in the first session or two, then nothing. The user churns.

But there's a second pattern that looks almost nothing like the first, even though the raw event counts are similar. In this one, the user returns regularly. They create, edit, sometimes let something sit for a day, then clear it. The cadence is steady. The account, at any given snapshot, holds little or nothing. If you pulled a list of "accounts with zero saved objects" you'd find them right next to the people who signed up and never came back.

The difference is what the product is doing for them. The first group is failing to use the tool as a record. The second group isn't trying to keep a record at all. They're using the product as a thinking surface — a place to externalize a decision, look at it, rearrange it, and then wipe the board clean once the thinking is done. The artifact was never the point. The cognitive loop was.

This reframes something basic about how we read engagement data. Most analytics stacks measure value by what persists: saved projects, completed workflows, stored records. The implicit assumption is that a product's worth lives in its outputs. And for many users, that's true. But for the subset treating your tool as a scratch space, the value is entirely in the process. It's consumed in real time and leaves no trace in your database. You could call it "dark engagement" — activity that is genuinely valuable to the user but invisible to any metric that counts objects rather than sessions.

This matters for retention modeling. If your churn prediction weighs "number of saved items" or "days since last meaningful action," these users score as dead. A well-meaning re-engagement email hits their inbox: "We noticed you haven't created anything in a while. Need help getting started?" They created something yesterday. They just deleted it when they were done thinking. The email doesn't just miss — it signals that you don't understand how they use you, which is worse than not emailing at all.

It also complicates the question of what "active" means. A user who logs in, stares at a blank canvas for four minutes, types three lines, deletes them, and logs out has — by most event-stream definitions — done almost nothing. But if they do it every Tuesday and Thursday for a year, something is clearly working. The product is holding space for a cognitive routine. That's a dependency, and a deep one, even though it would score a zero on any feature-adoption dashboard.

One smaller note: this pattern seems more common in tools that are flexible enough to be misused — spreadsheets, text editors, lightweight project boards, diagramming tools. Products with rigid workflows tend to force the user toward completion and storage. The scratch-surface behavior emerges precisely where the product doesn't insist on an outcome. That's not a design flaw. But it does mean the teams most likely to have these users are also the teams least likely to see them, because flexible tools tend to define engagement around the structured use cases they were designed for, not the unstructured ones people actually adopt.

The presence of a user who erases everything and returns is a quiet signal that your product has become part of how someone thinks. No object in the database will tell you that. The session log might, if you learn to read absence as its own kind of evidence.