Part of an ongoing series on ordinary workspace friction, and what happens when you point an agent at it.
Somewhere in an account list right now, there's a customer quietly heading for the exit. Nobody's noticed yet, because nothing about it looks urgent. Their login frequency has been dripping down for a few weeks. The person who used to reply to every check-in email within the hour has gone quiet. The last two support tickets came back shorter than usual, and a little colder in tone.
None of that trips an alarm. Each piece looks small enough to ignore on its own: maybe they're busy, maybe the tool's been running fine so there's less to ask about, maybe the usual contact is on vacation. It's only after the cancellation email lands that anyone goes back and notices the pattern was there the whole time.
That's the part worth sitting with. The warning signs are almost always there before the churn is. Nobody's paid to sit and watch for them.
The signal is in the data. Nobody's job is to look for it
Account managers and CS teams do watch their accounts, but attention is finite, and it flows toward the accounts making noise. A prospect asking for a new feature, an account escalating a bug, a renewal call on the calendar next week: those get looked at closely, because they're already asking for it.
The account that's going quiet doesn't ask for anything. It doesn't file an escalation. It uses the product a little less, replies a little slower, and eventually stops replying at all. There's no ticket for "this relationship feels different than it did last month," so nobody opens one.
Even a diligent account manager, tracking a reasonable number of accounts, can't hold the last month of usage and support history for every one of them in their head at once. They're not failing at their job. They're not equipped to notice a slow drift across dozens of accounts when nothing about any single week looks wrong. The cancellation email isn't the cause of the loss. It's the first moment anyone was forced to look.
The kind of system we'd build
This is the kind of problem we like: the information already exists, spread across tools your team already uses, and nobody's positioned to watch all of it, all the time. It's the same reason teams come to us for AI development.
Here's what an agent for this looks like. It watches usage and support tone across your accounts, continuously, not just when someone happens to pull a report. It notices when a contact who used to log in every few days hasn't shown up in a while. It notices when the last few support conversations read shorter, flatter, or more clipped than that account's normal back-and-forth. When one of those signals starts sliding, it doesn't wait for someone to spot it by accident. It posts in Slack: what changed, how long it's been going on, and a suggested next step.
Nobody has to remember to check. The account that's drifting gets surfaced while there's still time to do something about it, instead of after the cancellation notice makes the drift official.
The agent flags. A person still decides what it means
The agent can tell you usage dropped and tone shifted. It can't tell you why. Maybe the main contact left the company. Maybe they're mid-reorg and this tool genuinely isn't the priority this month. Maybe it's a real signal that the relationship is souring. Telling those apart takes context the agent doesn't have: history with the account, a read on the relationship, judgment about what's worth a phone call versus a routine check-in.
So someone on the team still reviews the flag, decides whether it's a real risk or a false alarm, and decides how to reach out if it is. The difference is where they start. Instead of finding out something was wrong after the account is already gone, they're starting with a specific, dated observation and time still on the clock to do something with it.
A question worth sitting with
Think about the last account that cancelled and caught your team off guard. Go back and look at what their usage and support history actually looked like in the weeks before. There's a decent chance the signal was there: a login pattern that thinned out, a contact who stopped replying as fast, a tone in the tickets that had gone a little cold.
It's not that nobody could have seen it. It's that nobody was positioned to be watching every account, every week, for a pattern that only looks obvious in hindsight. That's not a people problem. It's a coverage problem, and coverage is exactly the kind of thing worth handing to something that never stops watching.
Next in the series: The Friday Status Update Problem.