proSapient and Oliver Wyman publish survey on agentic AI in sales workflows
A joint survey of 100 sales leaders reports productivity gains alongside a 40% rate of AI-related incidents.
proSapient and Oliver Wyman released a joint survey of 100 sales leaders on how agentic AI is being deployed in commercial workflows, published on proSapient's blog on October 2, 2026. The session featured proSapient Founder and President Margo Polishchuk, Oliver Wyman Partner Gabe Knapp, and Oliver Wyman's Aaron Mallenbaum, who demoed three AI agents live.
The productivity numbers sit at the top end of what vendor-adjacent surveys in this category typically report, and the sample is self-selected to sales leaders already deploying the technology. Readers should weigh the 89% and 87% figures with that framing in mind: these are practitioners reporting on their own rollouts, not an audit of outcomes.
The more interesting data point is the 40% incident rate. That is a high number for a technology category still in early enterprise deployment, and it is consistent with what Oliver Wyman and proSapient flag as the central operational question: governance before scale. The survey does not break down incident severity, which matters. A permissioning misfire is a different problem from an agent executing an unauthorized action on a CRM record.
The funnel-stage finding tracks with what moderators at expert networks hear from sales-ops buyers. Lead scoring, enrichment, and account research are self-contained tasks with clear inputs and outputs, which is why they automate first. Mid-funnel work (multi-stakeholder deals, procurement cycles, pricing negotiation) remains relationship-bound and judgment-heavy. Agents can prep a rep for those conversations. They do not replace the rep in them.
Are your experts using AI to cheat?
Try our free demo to find out today.
What to watch: whether follow-on surveys from the same partnership break out incident data by firm size, deployment stage, and governance maturity. That is where the operational signal lives.
Powering institutional-grade transcription for expert networks.
INFLXD provides AI-powered, human-edited transcription with sub-1% error rates for the world's leading expert networks and financial research firms.
Visit inflxd.com →Keep reading.
How Buy-Side Firms Handle Expert-Network Call Transcripts After the Engagement Ends: 7 Retention and Access Models
A structural map of how multi-managers, long-onlys, and credit shops retain, redact, or purge expert-call transcripts once the research is done.
Halluminate raises $30M Series A for AI training environments
The nine-person startup building simulated finance workflows for frontier labs counts four top U.S. AI labs as customers.
SEC proposes rule changes to widen retail access to private market funds
The Commission's proposal would expand retail investor choice in regulated fund structures that hold private assets.