Smart Communications - ML-Driven Personalization
Launched ML-driven personalization for notification timing as a two-way door decision while engagement was declining. Validated quickly, then expanded the framework to frequency, channel, and content.
Context
User communications at Games24x7 relied on fixed send times and frequencies based on experiments run several years earlier. User behavior had evolved, and early indicators suggested personalization could improve engagement. Validating this through traditional experimentation would have taken weeks while performance was already declining.
Approach
- Prioritized Smart Communications as a two-way door, calculated risk decision.
- Narrowed initial scope to Smart Timing: using machine learning to personalize notification send times based on past user engagement patterns.
- Aligned stakeholders directly using analytical rationale, external research, and competitive benchmarks, without waiting for formal sign-off.
- Managed risk by limiting blast radius: launched on a single parameter, ensured fast rollback capability, positioned the release as a learning mechanism.
- Upon validation, expanded to optimize frequency, channel, and content.
Result
- 30% increase in notification open rates.
- 20% increase in click-through rates.
- Established a repeatable framework for ML-driven communication personalization across multiple parameters.
What I learned
When the cost of waiting is greater than the cost of being wrong, ship narrow and reversible. The Smart Timing launch was a single parameter with a clear rollback. That made it a two-way door, even though the broader Smart Communications bet was much larger.