Amplify — AI seller-insights tool
Amplify is a seller-insights tool that reduced average seller-call prep from 45 to 5 minutes in a 10-account-manager study. It serves approximately 1,600 account managers across North America and Europe.
1
Proof
~1,600
NA/EU account managers
Primary seller-insights tool
45 → 5 min
Average seller-call prep (−89%)
Before-and-after time study, 10 account managers
< 500 ms
P95 latency
Production
$3,000 → $250
Monthly inference spend (−92%)
Same workload; recommendation and email quality stayed above release thresholds
2.1
The problem
Marketplace account managers had seller data, dashboards, and playbooks. Before each seller call they still had to piece together what mattered and decide what to do.
2.2
Why it mattered
Account managers felt it on every call: preparation averaged 45 minutes in a 10-person time study.
The original request was a dashboard aggregating seller signals. They already had dashboards. Another reporting surface would have left them doing the same synthesis.
2.3
What I did
Changed the brief from reporting to preparation. Amplify brings seller context, a recommended action, and the evidence behind it into one place.
Rules decide, models explain. Deterministic rules handle eligibility, permissions, freshness, and routing; models synthesize and explain the evidence.
Precompute instead of generating live. Pre-computation, task-specific models, and batch processing keep it fast and cheap.
I wrote the PR-FAQ, made the case to leadership, and built it with AI-assisted coding. North American data engineering built the pipelines for the NA launch; European engineering adapted it for Europe.
How a recommendation is produced
The workflow combines structured seller signals with policy and playbook context. The service checks permissions, seller state, eligibility, and freshness before retrieval and selective generation. The interface shows the recommended action, its evidence, and a feedback path.
2.4
What went wrong
Our anchor customer changed leaders shortly before the PR-FAQ review. I didn't confirm the new leader's decision criteria and support, and the customer withdrew during the review.
I formed a cross-functional working group, sent weekly updates on decisions and risks, and expanded the pilot to two other account-management teams. Those pilots exposed gaps in the campaign workflow, which became a campaign-performance module. The original customer came back and asked us to launch.
Adoption slipped three months.
I now brief key stakeholders before any formal review, especially after a leadership change.
2.5
Stack
Rules engine with task-specific models, precomputed and batch generation, built with AI-assisted coding.
3
Contact
Hiring for a product area?
I am looking for Principal / Staff IC ownership of a product area.