From innovative vision to enterprise scale.

The Situation

Executives and brand managers set advertising strategy in one world; campaigns ran in another. Recommendations arrived strategy-blind — less relevant, less trusted — leaving revenue unrealized across 30K+ advertisers.

The Calls I Made

1. Interviewed sales teams, agencies, and brand managers to map planning-to-execution, and traced the leak to one structural cause.
2. Wrote the three-year vision and won VP approval before building.
3. Proved the hypothesis with a 3–4-week prototype and a difference-in-differences test (~50 bps adoption lift).
4. Cut scope deliberately: one advertiser tier, four of six features, one metric per funnel stage.
5. When beta advertisers rejected the platform over the missing trust loop, owned the call publicly, delayed launch two months, and shipped forecasting and closed-loop reporting.

Results

Year one beat plan: adoption reached 30% against a 27% plan (from a 23% baseline), revenue per adopted recommendation more than doubled, and the multi-year growth target was raised from 10% to 12%.

The Situation

Sales teams spent hours stitching together data, insights, and narrative before every pitch. A single copilot wouldn't fix it — the work needed planning, retrieval, analysis, and rendering, coordinated.

The Calls I Made

1. Started with a planning agent that shows a user-validated work plan before any agent acts.
2. Built a semantic layer over first- and third-party data so every agent retrieves the same truth.
3. Shipped an analytics agent that maps plain-language requests to data and runs the workflow end to end.
4. Sat in user testing with the engineers, then shipped a simple link-out fallback instead of waiting on perfect rendering — protecting the Prime Day-season launch.

Results

Reported stage-level gains sum to a 50% cumulative productivity improvement, and the rendering agent lifted pitch adoption 1,000 bps. Persistent preference memory is in flight — not yet shipped.

The Situation

Recommendation volume was never the problem — relevance was. A salesperson's attention is the scarcest resource in the funnel, and generic suggestions burn it.

The Calls I Made

1. Launched anomaly-based lead generation to find the accounts that actually needed intervention.
2. Combined predictive ML with generative AI to recommend new campaigns.
3. Used LLM embeddings and unsupervised learning to surface optimization opportunities in active campaigns.

Results

Three products delivering 18%, 13%, and 8% incremental annual revenue, respectively.

The Situation

Model confidence tells you what's likely; it doesn't tell you what's worth doing. Ranking needed causal expected impact, and adoption needed forecasts a salesperson could believe.

The Calls I Made

1. Built causal estimation of expected incremental revenue for every recommendation, at scale.
2. Re-ranked all recommendations by expected impact instead of model confidence.
3. Added seven-day outcome forecasts so users could see what acting would earn.

Results

Average salesperson quota fulfillment rose 1,000 bps from causal ranking; recommendation adoption rose 1,200 bps after forecasts shipped.