Methodology, made visible

Key Driver Analysiswatch it actually work.

Most firms hand you a ranked list of "what drives engagement" and ask you to trust it. We'd rather show you the machine. This live model runs the exact method on simulated employees with known drivers — so you can watch it separate what truly moves engagement from what merely looks like it does.

Interactive demonstration · simulated data with known answers
The simulated survey

We generate a bank of employees who each rate 9 areas of their experience, plus their overall engagement. Engagement is built from 8 real drivers we control — and one decoy. Move the controls and the whole page recomputes.

800
Medium
Respondents800
Areas measured9
Model fit (R²)—
Variance explained—
1The raw material

We never ask people what matters

That's the trap most surveys fall into — asking "how important is recognition to you?" People are poor judges of their own drivers. Instead we measure two things: how satisfied each person is on every area, and their overall engagement. The drivers are inferred from how those move together, never self-reported.

A peek at the dataset

Each row is one employee, scored 0–100. (Showing 8 of 800.)

…and every other respondent, each rating all nine areas. From here on, no human opinion about importance enters — only the data.

2The mechanism

Does this area actually move engagement?

Pick an area. Each dot is one employee: their score on that area (across) against their overall engagement (up). If the cloud tilts steeply upward, people who score it high are far more engaged — it's a real driver. A flat cloud means the area barely moves engagement, however much people care about it.

Correlation with engagement—
Slope—
↗

—

3The crucial distinction

A high score is not the same as high impact

This is where averages mislead. An area can score well yet barely move engagement, or score poorly while quietly driving everything. KDA reports both numbers separately — and the gap between them is where the budget decisions live.

How well it scores—average across the bank (performance)
How much it moves engagement—isolated impact (relative weight)
—
4The whole picture

What actually drives engagement — and the trap of correlation

Run every area at once and you get each one's share of what moves engagement. Here's why the method matters: switch between a naïve correlation ranking and the isolated impact KDA produces, and watch a plausible-looking driver collapse.

!

—

5The decision

From numbers to a priority map

Finally, plot every area by how well it scores (across) against how much it drives engagement (up). The top-left — low score, high impact — is where to act first. The bottom-right is where money is wasted on things that score well but barely matter.

Priority matrix

Computed live from the simulated survey above.

About this demonstration. The data here is simulated with drivers we set in advance, so you can verify the method recovers them — and catch the decoy. On real engagement data the principle is identical, but we use relative-weights / Shapley-value methods to fairly split influence between overlapping areas, report confidence intervals, and re-estimate every wave (drivers shift over time). Key Driver Analysis shows association, strong enough to prioritise on — not proof of causation — and needs an adequate sample, which is exactly what the sample-size control above lets you feel.