The Score Is the Cover
But what if it wasn’t?
Your company ran a pulse survey last quarter. Belonging scored 3.4. Someone put it in a slide. A leader noted the trend. The quarter ended.
That is not a follow-through failure. It is how scores are designed to work.
Survey requests to employees jumped 71% since 2020. Response rates crashed in many organizations over the same period. The common read is survey fatigue. The real problem is the format.
The number is the permission slip.
A 3.4 is designed to compress complexity. That compression does not just simplify the data. It creates the political conditions for inaction. Scores can be trended, benchmarked, and moved past in a meeting. McKinsey’s analysis of more than 20 academic studies found the primary reason employees stop participating in pulse surveys isn’t volume. It’s that nothing happens after. The score goes in, the number comes out, and leadership moves on.
What rarely gets examined is why moving on is so easy. The score is an average, and averaging is how you flatten the fact that a 3.4 on belonging can mean completely different things depending on who you ask. A team of engineers buried in a remote structure and a group of coordinators who don’t know where they stand with their manager can both average to the same number. The score papers over that divergence. Leadership never has to sit with the specific human situation underneath it.
This is where voice AI makes a real argument.
Not as a better survey tool. But as a format that produces data that is structurally harder to deprioritize. When someone says “I don’t think my manager knows I exist,” that is not a score. It is a sentence. It resists being averaged. It cannot be benchmarked against Q3. It has to be answered.
We have an opportunity to get this context at scale by using voice AI to run surveys. I’d much rather speak about my points of view and allow my train of thought to run, rather than typing it out or selecting an answer to a question.
There are real concerns worth naming before anyone deploys this. AI transcription is less accurate for non-native English speakers, with error rates running 10 to 22 percentage points higher than for native speakers. Voice data raises genuine questions about privacy, especially when employees don’t know how recordings will be used or whether candor carries professional risk. These are implementation problems, not arguments against the format. But they are not small, and any rollout that skips them will corrode the trust it was trying to measure.
The case for voice AI in pulse surveys is not just that it collects richer data. It is that it changes the political economy of what leadership can do with the data. A score can be tabled. A trend line can be monitored. An anonymized voice transcript saying, in specific words, what is wrong at work is harder to explain away in the next all-hands.
What does your organization do when a score drops? Would the answer change if leadership had heard the voice behind it?