Real-Time Enablement

The problem

Convin’s product told agents what they had done wrong after the call ended. Useful for coaching, but useless in the moment that mattered — while the agent was still speaking with a customer.

The opportunity was not a feature request. It was a category shift: surfacing guidance during the call moved the product from post-call QA into real-time agent enablement.

The business strategy shift

I worked with leadership to reframe the spend: post-call analysis is a compliance and reporting purchase, while real-time assist is revenue enablement. That repositioning expanded the addressable market from roughly $200M to an estimated $3–4B+.

The latency problem

Early builds surfaced nudges 4–8 seconds after the relevant moment. By then the conversation had moved on. This was not polish; it determined whether the product worked.

Partnering with engineering, we introduced a new recording and processing approach to cut delay below one second. I defined where nudges needed to appear, how much an agent could absorb mid-conversation, and what fast enough to be useful meant in a live call rather than a demo.

My role

I owned the interaction model for when a nudge surfaces without disrupting a conversation and connected it to the Rule Engine, so the same logic powering audits could power real-time guidance. One platform, two applications.

What I’d take forward

Real-time features live or die on latency. Building on the Rule Engine rather than a parallel system was what made this fast to ship.