Workforce measurement for AI deployment
Assaya Labs asks every employee how AI changed their work, then shows leadership which part of the resistance they created themselves, and what reverses it.
Know whether a team needs training, a workflow redesign, or a straight answer about their jobs. Three different budgets, three different owners.
See which seats sit untouched, which teams quietly route around the tool, and what it would take to bring them back.
We run the same instrument again after 90 days and report what changed, including the recommendations that did not work.
The business case
A thousand seat deployment runs to roughly this much a year in licences alone, before a single hour of change management.
Seats commonly go untouched. The usual response is more training, more comms, another pilot, and no measurement of whether any of it was the problem.
Most of the time the cause is a workflow nobody redesigned, a question about job security nobody answered, or work quietly pushed onto the team downstream.
Each of those has a cheap fix. None of them has the same fix. Knowing which one you have is the whole job.
Specimen readout
n = 1,240 employees · 4 trigger levels
Traceable to how the rollout was run
62%Illustrative figures, shown to explain the output. Every engagement reports its own.
What you receive
The attribution split, the workflow map and your position against comparable organisations, presented live rather than emailed.
Team level findings for every manager whose team clears the reporting threshold, with the few things that manager personally controls.
Each one carries a named owner, an effort estimate, a target measure and an honest expected size of effect.
What everyone collectively said, what leadership committed to, and the date we ask again. This is what makes the second round honest.
The same instrument, the same population, side by side. We report the recommendations that failed as clearly as the ones that worked.
Where you sit against comparable organisations in your industry, so a score becomes a judgement rather than a number.
Method
Sampling misses the thing that matters most, which is what happens between teams. Agents run every call in parallel, so we interview across the handoffs rather than around them.
A 10 to 20 minute spoken conversation per person, thousands running at once. A published, validated scale is administered inside it word for word and scored as the call runs, so the agent probes wherever a score comes back high. That is how a number turns into a reason you can act on.
Names, clients and specific incidents are removed before anything is analysed. Nothing is reported below a minimum group size, and the check runs on combined filters rather than one at a time.
Anxiety is separated by trigger level, so the share your rollout created sits apart from the share that arrived with the news cycle. The first is reversible. The second is not.
You act for 90 days with our support kept deliberately light. Then we re run the instrument and report the movement, including where nothing moved at all.
For the people being interviewed
Employees answer honestly or the data is worth nothing. That is not a reassurance we offer for comfort. It is the constraint the entire method is built around, and it is why we brief employee representatives before fieldwork rather than after.
Method and provenance
The core instrument is a published, peer reviewed scale from organisational psychology, administered exactly as it was validated. Same wording, same order, same response format. That discipline is what lets a score be compared between one quarter and the next, and between your organisation and its sector.
Privacy and employment law obligations are designed into the method rather than added afterwards, which is why the same architecture holds whether you are governed by the EU AI Act, US state privacy law, or a collective agreement. Aggregate only reporting is a deliberate decision, and we document it as one.
Fit
Design partners
Early partners get the work at a reduced fee and help shape the instrument. In exchange we ask for depth of access: the AI programme owner in the room, and the operational metrics to tie the findings to.