We Were Watching the Wrong Signal

Yesterday I wrote that on the provider side of administrative work, a wrong answer often generates no signal at all. The patient who didn't get the callback doesn't call back.

Here is the version of that problem that sits inside your own operation.

We had an account that looked healthy for over a year.

Daily productivity reports came in. Login and logout times were tracked. Communication was regular. Every time we asked the client how things were going, they told us they were satisfied.

Zero escalations across the entire engagement.

The first indication that anything was wrong was the client asking us to replace the person on the account. When we sat down to understand why, we heard about performance concerns that had been developing for some time and had never reached us through any of the channels we had built to hear about exactly that.

Two weeks later, before the replacement was in place, they terminated.

The monitoring was not missing. It was running the whole time, and it was reporting accurately. It tracked whether the person was logged in, and they were. It tracked whether reports were submitted, and they were.

What it did not track was whether the work was being done well.

Those are different questions, and only one of them was instrumented.

A low escalation rate is not a measurement of how much went wrong. It is a measurement of how much crossed a line somebody drew in advance. When the line is drawn around presence and output volume, because those are the things that are easy to capture, a quality problem can run for months without ever touching it.

The number does not just fail to surface the problem. It closes the question. An empty dashboard sends someone looking. A clean one does not.

We were not complacent. We were watching the wrong signal, and the report told us we were fine.

What we changed was not more monitoring. It was defining, in advance, what counts as an indicator that something is wrong, and what an account manager checks to get information that isn't just the client saying things are fine and the person on the account saying things are fine.

That gets written down. Every exception we hit is researched, classified, and turned into a documented standard with its indicators, its checks, and its escalation path. Our account managers are required to know it and are tested on it.

I want to be careful about what that buys. It is not coverage. You cannot build a check for every failure, and this case involved more than one thing going wrong at the same time. What you can build is redundancy, so that a failure has to get past more than one person to reach a client.

Neither people nor software get this to zero. Anyone selling you a system with no error rate is selling you the demo. The question is whether a failure has anywhere to surface, or whether it just runs until a client stops calling.