Measurement tools
Your employee survey is averaging away the answer
The company number is the one figure in the dataset that cannot tell you where the problem is.
Six tools, all free. Price your attrition gap · Five Systems Check · Signal Loss Check · Offloaded Judgment Check · Not Everyone Sees This · Seven questions on AI savings · what's coming in 2027
Eight of these teams are fine. One manager's team is at 2.3, roughly a point and a half below every peer. Nothing in the reported 3.8 distinguishes this organization from one where every team scores 3.8. The two require completely different responses, and only one of them is expensive.
The problem
Averages are where organizational costs go to hide
A company-wide 3.8 out of 5.0 can conceal a 4.2 for one group and a 2.6 for another. It can conceal a single manager whose team sits at 2.3 while every other manager clears 3.8. Both organizations report the same headline number, and the survey vendor's dashboard will show them as identical.
This is the same error that makes two different turnover problems report the same rate. An aggregate attrition rate of twelve percent is meaningless if one group is leaving at eight percent and another at sixteen. And that spread, not the rate, is what the departures cost you. The disaggregated view is where the actionable intelligence lives, in both instruments.
What follows is the measurement approach from The Projection Fallacy: what to measure, which validated instrument to use, and how to read the result so it points at something you can act on. The six elements it sits inside are set out in the SHIELD framework.
What to measure
Six metrics that make the cost observable
Each one makes a specific mechanism visible, and each connects to a business outcome. None of them requires anyone in the room to agree about why the gap exists before the number can be acted on.
Differential turnover
The gap between your best-retained and worst-retained groups, priced. This is where the largest cost hides, roughly $5.5M a year in a 1,000-person company.
Psychological safety, disaggregated
Edmondson's seven-item scale, analyzed by manager and by group rather than reported as a company mean.
Promotion velocity
Time-to-promotion by group. Equal promotion rates with unequal timelines is still a filter, and it compounds over a career.
Hiring funnel conversion
Conversion at every stage, by group. Applications 40% women and hires 15% women is not a pipeline problem. It is a process filtering at each step.
Meeting participation
Who speaks, who doesn't, and who gets talked over. Two weeks of noticing produces data even the most concrete thinker can verify.
Idea attribution
Whose contributions get credited, and whose get restated by someone else and credited to them. This is the mechanism behind month-six credit theft.
Two of these six now have instruments. The Signal Loss Check measures what happens to an accurate concern between someone raising it and a decision changing, which is the mechanism underneath both meeting participation and idea attribution. The Offloaded Judgment Check measures what a team stopped verifying once a tool started answering.
The instrument
Use a validated scale, not questions you wrote yourself
For psychological safety, Amy Edmondson's seven-item scale has been in use and under scrutiny since 1999. Run it anonymously, quarterly. It measures something narrower than the word has come to mean, which is worth reading before you run it: what psychological safety is not.
- Are team members allowed to make mistakes?
- Can they raise problems and uncomfortable issues?
- Do people on this team reject others for being different?
- Is it safe to take a risk here?
- Is it difficult to ask other team members for help?
- Would anyone here deliberately undermine another's efforts?
- Does this team use and value everyone's skills and talents?
Adapted from Edmondson, A. (1999). Psychological Safety and Learning Behavior in Work Teams. Administrative Science Quarterly, 44(2), 350–383. Items 3, 5 and 6 are reverse-scored.
How to read it
The five steps that turn a survey into a decision
- Pull the results and set the company mean aside. You will come back to it only for trend reporting.
- Split by manager first. This is almost always the cut that produces the longest silence in the room.
- Split again by group: function, tenure band, level, and demographic, to whatever depth your data supports.
- Take your highest-scoring team as the baseline. It is proof of what this organization can produce, which makes it a target rather than a hypothetical.
- Price the gap. A fifteen-point psychological safety spread and a differential turnover figure are the same finding expressed in two currencies, and only one of them gets a budget line.
Then make it consequential. A dashboard presented quarterly to the board with no attached decision is measurement theater, and people can tell the difference within one cycle.
If you would rather not run this yourself, the leadership team engagement does the measurement and the redesign inside one quarter, and measures again at ninety days.
Tools
How the costs connect
Navigation tax
The effort of working inside systems built for someone else: decoding unwritten rules, reworking a message until it lands, preparing twice for a meeting that rewards the fastest talker.
Then comes the calculation: is saying this worth what it could cost me? The higher the tax, the worse that bet looks.
Withdrawal
Concerns that are never raised. The people who can see the problem stop saying so, and a quiet room looks like agreement.
Signal loss
Concerns that are raised and then lost: never routed to the decision, not credited, not acted on, not written down.
Differential turnover
The group that leaves faster than everyone else. Usually the people who saw the problems first.
Explicit systems lower the cost at every stage: criteria written down, agendas sent ahead, the same questions for everyone, decisions recorded with the reasoning.
Six tools you can run this week
Free · 5 minutes
Differential turnover calculator
Enter headcount, replacement cost, and your attrition spread. Returns your annual cost and a one-page summary you can take to your CFO.
Calculate your numberFree · 10 minutes
Five Systems Check
Five questions on which of the systems that decide who stays still live in people’s heads. Then ask two people hired in the last year the same five, and start where your answers differ most.
Run the checkFree · 6 minutes
Signal Loss Check
Twenty questions on what happens to an accurate concern between someone raising it and a decision changing. Scored at the narrowest point in the chain rather than the average, because a concern only has to fail once.
Run the checkFree · 6 minutes
Offloaded Judgment Check
Twenty-four questions on what your team stopped verifying once a tool started answering. Returns where judgment moved and whether anything still checks the result.
Run the checkFree · 4 minutes
Not Everyone Sees This
Ten images, and people see different things in them. Not tricks. You answer what you see, then find out why, and how you compare.
Take the testFree · 45 minutes, with your team
Seven questions before you bank the AI savings
For the conversation with a leadership team that has already sized a headcount against a productivity gain nobody measured. Each question comes with what a managed answer sounds like, so you can tell one from a real one.
Read the sevenComing 2027
Full cognitive architecture tool
Five personality domains, thirty facets, and how they interact with your organizational environment through the Projection Fallacy lens.
The evidence behind the six metrics above, every figure traced to a primary source and the weak ones named as weak: Key Research Findings (PDF).
Two architectures, side by side: how neurotypical and autistic thinking differ, and where each has the edge at work (PDF).
Related reading: what your attrition spread costs, reducing employee turnover, and the four cuts to run against your HRIS.