melwyn.ai
← Lab / Selected work
Revenue operations

Lead Quality Engine

Score how well every lead was handled, then send that signal back to the ad platform.

Built in-house for a sales operations team
The problem

Lead handling quality could not be measured at scale. Reviewing calls and chats by hand was impossible, so whether the team followed the playbook was anecdotal, and managers could not coach on anything concrete. Meanwhile the ad platform optimized for raw lead volume, so it kept buying cheap leads that rarely closed.

The approach

Every night the engine gathers each lead's CRM trail, calls, and chat threads, transcribes the calls with speech to text, and scores the handling on a 100-point card: half from process metrics in the CRM, half from an AI read of the transcript with quoted evidence. A human can flag any score as wrong, and those corrections retrain the model against a gold set before any change ships. The same quality signal goes back to the ad platform, so budget can chase value instead of volume.

How it runs

Walk the pipeline

Click through each stage to see what the system does and what the team gets back.

Stage 1 of 5

Pull the whole story of every lead

Each night it gathers every lead's CRM trail, call recordings, and chat threads. Nobody samples a handful by hand and hopes it is representative.

Every leadCRM, calls, chats
What changed

100-pt

scorecard on every lead

Nightly

across every call and chat

Human

verified calibration loop

Lead handling went from anecdote to evidence. Managers coach on real moments from real calls, the score keeps itself honest through a human calibration loop, and the same quality signal teaches the ad platform to buy leads that actually close. The budget stops chasing volume and starts chasing value.

CRM activityCall transcriptionLLM scoringAd-platform feedback

Want something like this on your team?