Proof, Not Pilots – How We Run AI for Business Outcomes

Yesterday was about what's possible; this session is about how to run it. The Seagate operating model scales People Analytics and HR AI through a visible service catalogue, a simple pilot→production scorecard, and a monthly value cadence Finance trusts. Two vignettes evidence outcomes and the guard‑rails – privacy, bias checks, audit trails, rollback – that keep Legal and Works Councils supportive. Leave with templates to apply next quarter.

• The service catalogue: 6–8 operated services with Owner | Consumer | SLA | Success metric, plus intake→triage flow and RACI so business asks route cleanly.
• Pilot→production scorecard: Data, Evaluation, Safety and Operate checks – provenance, SLOs, access, PII minimisation; ground truth and bias tests; policy‑as‑code, explainability, red‑team and rollback; owner/SLA, adoption telemetry and bound cost‑to‑serve – to kill, pivot or scale decisively.
• Two case vignettes: manager policy copilot/HR decision service and a throughput or planning uplift, each with problem, minimal mechanism, quantified outcomes and the guard‑rails that kept it safe.
• CFO‑credible value model and cadence: Hours→Capacity→Cost conversion with adoption factor and ranges; a monthly note to CFO/ExCo on what moved, why, confidence bounds and next bets.
• Adoption telemetry and governance that endures: weekly actives, completion and exceptions; AI literacy and UX patterns; model cards, access control, logging, legal checkpoints and rollback, plus one non‑negotiable control and one practice retired.

• Publish a People Analytics/AI service catalogue with clear owners, SLAs and success metrics.
• Apply a pilot→production scorecard across Data, Evaluation, Safety and Operate to decide kill, pivot or scale with confidence.
• Convert time saved into capacity and cost using a CFO‑credible method, and report monthly value ranges with stated confidence.
• Operate governance by design – policy‑as‑code, access control, logging, model cards and red‑teaming – with a tested rollback path.
• Set up an adoption telemetry loop to track weekly actives, completion, time‑to‑decision and exception rates, and use it to prioritise what to scale or stop.

Patricia Frost

SVP and Chief People and Places Officer · Seagate Technology

In global manufacturing, leaders need operated decision services, not isolated pilots. Labour and capital are tight, quality and safety are non‑negotiable, and works councils and regulators expect privacy and bias control. Finance demands value ranges it can defend. Scaling HR AI requires service discipline, governance by design, and auditable outcomes.