Why most Agentforce pilots stall at day 90

Across the Agentforce rollouts we've benchmarked, the failure mode at day 90 is nearly identical: a working agent, no clear owner, and a steering committee asking whether it 'paid back'. The model isn't the problem. The decisions made before the build are.

Three decisions, made early and on the record, separate the pilots that scale from the pilots that quietly disappear.

An agent without a P&L owner is a science project. Pick the line leader whose number moves when the agent works — service VP for case deflection, sales ops for lead qualification, supply chain for proactive case creation. Their bonus structure must reflect the agent's outcomes. If no one's comp changes, no one cares at day 91.

Define the cutover metric before you start. Not 'CSAT will improve' — 'first-contact resolution moves from 62% to 70% on tier-1 cases within 60 days, measured in Service Cloud reporting'. Vague success criteria guarantee a vague day-90 review.

Most stalls we see trace back to a Data Cloud or sharing-rule conversation that should have happened in week one and happened in week ten. Pre-clear the data scopes the agent needs with security and the data steward before kickoff. Treat it as a procurement decision.