AI Pilots to Production – Scale What Works, Kill the Rest

Maddie Oliver

Senior Vice President, People Research & Analytics · Citi

Ketaki Sodhi

Agentic Listening & Analytics Program Owner · Microsoft

Teams are running GenAI pilots everywhere, yet too many drift on without evidence or end up blocked by governance. This session offers a repeatable path from pilot to production – value and risk gates before build, instrumentation that distinguishes deep from shallow use, enablement patterns that stick, and a 90/180‑day scale path. Expect criteria to kill early, design choices that accelerate adoption, and proof that Finance and Legal will recognise.

• Value and risk gating before build: owner, target metric, baseline, eligible data, privacy and bias checks, 90‑day success criteria, funding path.
• Instrumentation that shows impact beyond 'usage': deep vs shallow use, time‑to‑insight, rework, error rate, policy adherence, and network diffusion from telemetry plus EX and people data.
• Kill and scale decisions: baselining, signal‑to‑noise thresholds, governance cost, and change friction that outweighs benefits.
• Design choices that accelerate adoption: agentic assistants, prompt pattern libraries, role‑based access, logging, and minimum viable workflow integration to HCM, ATS, VMS, and ticketing.
• Enablement and operating model: upskilling cadence, guard‑rails that stick in regulated contexts, finance/legal handshake, and a 90/180‑day scale path with monthly value reporting.

• Use a gating checklist to choose and fund pilots with clear 90‑day success metrics and risk controls.
• Set up instrumentation that evidences value beyond 'usage' – including time‑to‑insight, rework and defect reduction, policy adherence, and network diffusion.
• Adopt prompt design and enablement patterns that raise competency and reduce risk across HR and business teams.
• Decide when to kill or scale using baselined signals, owner accountability, and governance cost thresholds.
• Apply one‑page templates – Pilot Triage Card, Instrumentation Grid, Enablement Sprint Plan, and a 90/180‑day Scale Path – to move from pilot to production.

GenAI adoption is accelerating, but HR and people analytics face pressure to show measurable value, protect privacy, and avoid uncontrolled pilots. Finance demands proof beyond usage, Legal needs robust controls, and leaders want scale without risk. Organisations require a clear operating model to decide, instrument, and industrialise.