What a Dashboard Cannot See

Some things are unmeasured because nobody bothered. These are unmeasured because they cannot be instrumented, and pretending otherwise produces a worse number rather than a better decision. For a related reference, see further details.

The seven

Whether somebody is being carried. A team member whose work is quietly absorbed by two colleagues shows normal output and normal activity.

Whether the good people are disengaging. Disengagement precedes departure by months and looks identical to competence in every metric — the work still arrives, on time, without the extra.

Whether a decision was right. Systems record that a decision was made and never whether it was any good.

Whether the work is interesting. The single largest factor in whether skilled people stay somewhere, and it has no field.

Whether a manager is trusted. The largest controllable factor in retention and entirely invisible.

Whether a problem is being sat on. The thing psychological safety exists to prevent, and by construction it leaves no trace until it emerges.

And whether somebody has stopped learning, which they will notice about a year before anybody else does.

Why these resist measurement

They are states rather than events.

Systems capture events — a commit, a ticket, a login, a swipe. A state has to be asked about or observed, and neither produces a row in a database.

And several are only knowable by the person themselves, which means the measurement is a question and the answer depends on whether it is safe to give.

How each becomes visible

By asking, in a setting where the answer is safe. Which is what the one-to-one is for, and why its cancellation rate matters.

By noticing what changed. Not the level, the change: somebody who stopped volunteering, stopped arguing, stopped staying for the interesting part.

And by the slow measurements, which are proxies rather than the thing itself and are the best available substitute.

The temptation to instrument them anyway

Every one of the seven has a product claiming to measure it.

Sentiment analysis on messages. Engagement scores. Collaboration network graphs.

These measure something — usually message volume and who talks to whom — and attach the name of the unmeasurable thing to it. The precision arrives, the accuracy does not, and the organisation now believes it has covered the gap.

A known blind spot is safer than a false reading, which is the whole argument for this page existing.

What to do about it

Write the seven down and accept them as unmeasured.

Assign them to attention rather than to instrumentation. A manager whose job explicitly includes noticing these is doing something different from one reviewing a dashboard.

And check the proxies against outcomes, because a proxy that does not move with departures is not a proxy for anything.

The short version

For additional context on this topic, see DPReview.