At Avito — one of the world’s largest online marketplaces, 60M+ monthly users — I led communications and buyer-engagement products. Communications was running as a cost-centre channel — campaign teams, a send tool, open-rate dashboards. I authored the board-approved investment case to turn it into an ML-driven product, defended it through CEO and board review, and led the delivery.
What I built. A system, not more campaigns: a measurement substrate (global control group and a multi-metric outcome stack), a single orchestrator holding one per-user message budget, a ranking-and-filtering stack (candidate ranker, uplift head, send/no-send filter), an LLM content engine with a learned reward model, and deep-linked landing and notification surfaces. The platform spans messenger, email, push, and in-app and operates at billions of messages per month. Measured against a global control group — incrementality, not attribution — it reached a 5% incremental lift in buyers on Android and grew its annual revenue uplift 1.5x.
How that’s measured. The global control group is a permanent holdout of users who receive no communications at all; every result is the difference between them and everyone else, so the numbers count only what the messages caused — not what users would have done anyway. The experiment has run continuously for about four years, at α = 0.001 with CUPED variance reduction, and the headline effects land at 30+ sigma.
My contribution. I defined the direction, wrote and defended the investment case, and connected product, ML, CRM, content, and experimentation teams into a single commercial growth engine — and served as GenAI lead for marketing and communications.
For the full analysis of what a platform like this returns and what it takes to build, see the investment-case article.