Self-assessment is the astrology of change management. Comforting, widely used, and predictive of nothing. If you can't observe the behaviour, it didn't change.
Every transformation programme ends the same way. A survey goes out. People rate their confidence on a scale of one to five. The averages come back at 4.1, the deck gets a green slide, and everyone moves on.
Six months later, nothing works differently.
This is a measurement choice. Organisations ask people about themselves instead of watching what they do, and that choice fails for reasons that have been documented for decades.
Four mechanisms, all working against you at once.
First, people are poor judges of their own competence. The people who improved the least tend to rate themselves the highest, because the skill required to judge your own ability is the same skill you haven't built yet. That is how self-perception works for everyone, including the person reading this.
Second, surveys measure mood, not behaviour. Ask someone two days after an energising workshop and you get enthusiasm. Ask the same person during a brutal quarter-close and you get frustration. Neither number tells you whether they work differently on a Tuesday morning when nobody is watching.
Third, everyone knows what the right answer is. When your employer runs a transformation and then asks "how confident are you with the new way of working?", the question carries its own expected answer. People are being polite to a system that clearly wants a four.
Fourth, and this is the one nobody warns you about: the survey is also a ballot. When a change threatens something people value, they vote with their scores. We have watched teams in the middle of a change programme rate their mood, their flow, even their customer impact at rock bottom, round after round. The company was losing money; the team had enjoyed a good run together, and a low score was the only sanctioned way to say we liked it how it was. The distortion follows the agenda, whichever direction the agenda points. Sometimes that produces a polite four. Sometimes it produces a strategic one.
Stack these together and a self-assessment score is roughly as informative as asking a restaurant's chef whether the food is good.
The alternative is simple. It just takes more work. You decide, before the programme starts, which observable behaviours would prove the change is real. Then you count them.
If the change is about AI adoption, look at whether tasks moved from manual to AI-assisted this month, whether teams are documenting and sharing their workflows, whether unofficial workarounds are shrinking because the official path finally works. Every one of these is countable. None of them depends on anyone's opinion of themselves.
If the change is about a new way of working, the same logic applies. Are decisions being made in the forum the new model prescribes, or in the hallway afterwards? Are the new meeting structures producing outputs, or attendance? You can observe all of this. Most organisations simply don't, because observation takes effort and surveys take an afternoon.
There is a physics to this. Momentum needs mass and direction, and you cannot steer what you cannot see. A transformation tracked through self-assessment is flying on instruments that measure how the pilot feels.
Here is the part most change programmes get structurally wrong. The training, the workshop, the kick-off: these are the cheapest and easiest parts. What decides the outcome comes after, in the weeks where new behaviour either hardens into habit or quietly reverts to the old default. Knowledge is rarely what was missing.
That window, roughly the first 90 days after any intervention, is where measurement matters most and where it is most often absent. The trainer has left. The energy has faded. Whatever tracking exists is a satisfaction survey filed somewhere.
This is exactly the window where behavioural data earns its keep. Weekly signals tell you where momentum is building, which teams are drifting, where a detractor is quietly recruiting. You can act on that. You cannot act on "average confidence: 4.1."
Measuring behaviour instead of sentiment has a cost, and the cost is honesty. A survey can always be read generously. A behavioural baseline cannot. When the data says a team hasn't changed how it works, someone has to own that, usually someone senior who sponsored the programme.
This is why organisations retreat to self-assessment: it never delivers news that anyone has to act on. Inflated scores confirm the programme is working. Sabotaged scores get dismissed as resistance and filed away. Either way, the survey changes nothing, which is precisely why it survives.
So the real questions before your next transformation: which behaviours will prove this worked, how will we observe them, and what will we do when the numbers tell us something we don't like?
If you can't answer the third question, the first two don't matter yet.
We build behavioural baselines and measurement systems into every change programme we run. It is the layer that stays behind after we leave. If your last transformation was measured in survey scores, we should talk.