Listening as a Product – Signals, Open-text and Agentic AI

Operate Employee Voice as a product – not a reporting project. The session outlines an in-house model spanning engineering, comms and change; a weekly micro-pulse with UX and nudge experiments; and the shift to agentic AI where clean, central data replaces dashboard sprawl. Expect a 90-day plan, an instrumentation grid, and a model card pattern for safe AI-assisted listening, plus practical guard-rails for privacy and explainability.

• In-house operating model: cross-functional team, intake and backlog, success measures tied to decisions and time-to-signal.
• Weekly micro-pulse design: cadence, embedded delivery, UX/nudge experiments, coverage and fatigue monitoring.
• Open-text with AI: when to favour comments over scales; minimal evaluation to avoid hallucinations; red-team for sensitive themes.
• Agentic listening architecture: centralised quality data, access controls, policy-as-code, logs; retaining anchor surfaces during transition.
• Behaviour-of-response telemetry: interpreting who replies, when and to what; adapting cadence, channels and value exchange.

• Stand up listening-as-a-product: define roles, intake, backlog and success measures; choose in-house build where it makes sense.
• Design a weekly micro-pulse and run UX/nudge experiments to improve signal quality, not just response rate.
• Apply an instrumentation grid covering response coverage, time-to-signal, open-text signal strength and decision uptake.
• Prepare for agentic listening: centralise quality data, reduce dashboard dependency and add model cards and audit trails.
• Read behaviour-of-response as signal and adjust cadence and content for a more self-preserving workforce.

Kristin Saboe

Head of Employee Voice · Google

Survey fatigue, fragmented tools and rising expectations for privacy and explainability are colliding with rapid AI adoption. Leaders need faster, safer decisions from employee input while employees behave more self-preservingly, altering participation patterns. Operating listening as a product – with experiments and governed AI – has become a commercial necessity.