01 · Problem
Manual bid factors, inconsistent across teams
Across the organization, client teams were frequently asked to provide manual bid factor recommendations across inventory, audience, device, recency, time-of-day, and other performance dimensions. While these recommendations were valuable, the process was often time-intensive, inconsistent across teams, and dependent on individual analyst judgment.
Because there was no standardized statistical framework for calculating bid factors, recommendations could vary by team, account, or workflow. This created a gap between optimization strategy and mathematical rigor. It also pulled GTM and client service teams away from higher-impact work, as they had to spend significant time manually pulling data, evaluating performance, calculating recommendations, formatting outputs, and translating findings into client-ready narratives.
02 · Solution
A Flask app with Bayesian inference and PowerPoint automation
I designed and built an internal Flask Python-based optimization application that uses Bayesian inference to generate statistically supported bid factor recommendations at scale. The application evaluates campaign performance against benchmarks and applies probabilistic modeling to determine whether a segment is likely to outperform or underperform.
Rather than relying only on raw performance metrics, the app incorporates statistical confidence, sample size, uncertainty, and performance severity to produce more balanced and defensible recommendations. Outputs are translated into actionable bid factor guidance (scale, monitor, bid down, or cut), giving teams a clearer framework for optimization decisions.
The application is directly integrated with TTD's Vertica database, allowing users to securely pull campaign performance data without relying on manual exports. It also supports SSO-based access, keeping the workflow secure and aligned with internal authentication standards.
To make the tool more useful for GTM teams, I also built an automated PowerPoint assistant into the application. After the model generates recommendations, users can manipulate the data in-app, export the results to Excel, or automatically generate a client-ready PowerPoint deck. The deck assistant takes the statistical output and formats it into a polished narrative with key insights, recommended actions, and performance context, helping teams move from analysis to presentation much faster.
03 · Impact
A scalable, statistically rigorous optimization workflow
This project created a scalable, statistically rigorous optimization workflow that reduced the need for manual bid factor analysis and standardized how recommendations are generated across teams. It helped GTM teams move faster, improve consistency, and bring more data-backed confidence into client conversations.
By combining Bayesian modeling, direct database integration, Excel export functionality, and automated PowerPoint generation, the tool turned what was previously a fragmented manual process into a streamlined end-to-end optimization system.
