Kai · Voice-based organizational listening
Kai doesn’t run a questionnaire. Kai converses.
Kai is an organizational listening platform: it interviews the people of an organization by voice, from their phone, in an 8-to-10-minute narrative conversation, and turns hundreds of stories into aggregated evidence for decisions. Designed for financial services, where one more survey says nothing new.
The problem it solves
Climate and culture surveys measure perceptions with scales; people answer quickly and teams receive averages that explain nothing. What really happens in an organization lives in concrete stories: what was prioritized, who decided, what went unsaid. Kai goes after those stories, at scale, without turning listening into an interrogation.
How it works
1. Invitation and consent
People join from their phone with a link or code, with federated access through Microsoft Entra ID when the organization requires it. Before any text is kept, they review and consent to the exact version that is stored.
2. Narrative conversation
Kai runs a voice interview around real working days and situations. It does not improvise: every intervention is selected from a reviewed script, one question at a time; an independent review can veto an intervention and Kai rephrases it or simply listens.
3. Analysis with human oversight
Stories are consolidated into facts, narratives and counts. Results are published only in aggregate (a minimum of five people per cut) and with human review before reaching the organization.
4. Decisions, not rankings
Kai delivers maps and patterns to decide about the system of work. It never produces evaluations, rankings or decisions about individuals.
Data handling and product limits
- Audio is never stored. It is processed in memory during the conversation; only the text the person reviewed and consented to is kept.
- Encryption at rest for the people database and each campaign’s packages, with the key under the service’s control; without the key, the system does not start.
- Bounded retention: authorized text is kept for a maximum of 90 days, or less if the organization defines it, with a recorded purge.
- Personal-data minimization in the text (names, emails, phone numbers) before analysis. We do not present it as guaranteed anonymization: that is why results are published in aggregate and with human review.
- What Kai does not do, by design: no camera or video, no biometrics, no emotion analysis, no lie detection, no personality or performance inferences, no personal memory across campaigns, and no use of the data to train models.
- Every client isolated: campaigns, instruments and data separated per organization; a conversational application with human oversight that does not operate client systems or decide about people.
Kai is delivered as a managed service by Agility Changes. Every deployment goes through a laboratory with synthetic data and a validation of the instrument and the models with the client before listening to real people.
Who it is for
Meet Kai
We show you an example conversation with synthetic data and the kind of results a campaign delivers.