INCREMENTAL Rx · INDEXED 100 → 107 VISITED vs NOT-VISITED
- Scope
- top-5 pharma, CIS
- Baseline
- not-visited control (indexed)
- Role
- AI lead / builder
// CHALLENGE
At a top-5 pharma in the CIS, commercial effort spanned a dozen channels — F2F, email, SMS, messengers, remote calls, webinars, conferences, the HCP portal — but the contribution of each channel to sales was not reliably known: too many moving factors to account for at once. A single audience was cross-covered by multiple brands, so spend collided with itself. Layer on data heterogeneity, COVID, geography and seasonality plus each HCP's own channel and timing preferences, and 'where did the growth actually come from?' had no defensible answer.
// APPROACH
I led an AI/ML omnichannel platform built around the HCP, not the channel. Two linked engines: a Next-Best-Action model in CRM that picks the next message, channel, format and time per HCP; and a marketing-mix 'Sales-Impact Prediction' model that decomposes sales by channel, forecasts the impact of a given mix and computes ROI. Around them: a 5-tier HCP microsegmentation, source-data verification to survive the heterogeneity, and a hypothesis-testing layer that turns each cycle into the next insight. The headline win wasn't a model — it was finally being able to attribute sales to channels and act on it.
- Microsegmentation — 5 HCP response tiers decide who gets what
- Next-Best-Action in CRM — per-HCP message · channel · format · time across a 9-channel mix
- MMM 'Sales-Impact Prediction' — resolve each channel's contribution to sales, forecast the mix, compute ROI
- Verification & insight loop — survive data heterogeneity; realized Rx retrains the engine