AI-Powered Deal Health Score Engine for Programmatic Advertising Performance

Built for a US-based AdTech organization operating in CTV and programmatic space, to automate deal health scoring, surface anomalies in real time, and enable revenue protection before performance issues escalate.

Home

Our Work

Programmatic advertising moves fast, and the tools meant to support it often struggle to keep up. For one of our US-based AdTech clients operating in the CTV, mobile, and programmatic space, static rule-based approaches had quietly become a liability rather than an advantage. Deal tracking was always a step behind, performance monitoring was reactive, and the in-house team spent significant time manually interpreting the same signals, day after day. The client needed something smarter with the help of AI based technology, a way to stay ahead of performance issues rather than constantly catching up.

The client partnered with Rishabh Software to build a platform to track and manage deal performance, leveraging our AdTech domain-based AI/ML expertise and data engineering capabilities. With intelligent scoring and real-time alerts, teams no longer need to manually check data. This shift improved their workflow efficiency and helped them stay competitive.

Capability

AI/ML Engineering

Industry

AdTech

Country

United States

Key Features

The client was looking to unlock a smarter and more scalable way to monitor advertising (mobile, CTV) deal performance. To address this, we built a platform with the following key capabilities:

Intelligent Deal Health Scoring

We have implemented a normalized scoring framework that automatically evaluates the health of every active advertising deal on a 0–100 scale. Teams get a consistent and reliable way to measure performance and instantly identify deals that need attention, without manual review.

Segmentation-Aware Benchmarking

We introduced segmentation-aware benchmarking to provide better performance context across deal types, traffic sources, and campaign conditions. This helped the client compare deal performance more accurately instead of relying on generic thresholds.

Added Explainable Diagnostics

Structured diagnostics that go beyond flagging an issue, explaining why a deal is underperforming and what is driving the score. This gives teams the context they need to make informed decisions quickly, rather than spending time investigating the root cause themselves.

Real-Time Anomaly Detection

Implemented the capability to continuously monitor performance in real time across key bid lifecycle stages including Requests, Bids, Wins, Impressions, and Spend and flag unusual patterns early, helping teams quickly identify and act on underperforming deals.

Challenges

Manual, SME-driven deal analysis with no automated scoring pipeline made performance monitoring resource-intensive, inconsistent, and difficult to scale.

No normalized metric (0–100) to quantify deal performance consistently across inventory.

Benchmarking was not segmented by deal type, supply source, or campaign condition, making it hard to contextualize performance accurately.

No real-time anomaly detection across the bid funnel meant issues across key stages of the bidding cycle were often caught too late.

Without structured root cause diagnostics, teams had limited ability to understand why a deal was underperforming and act on it quickly.

Solutions

Our approach was to replace manual, SME-driven deal monitoring with an intelligent and scalable performance evaluation framework, combining a unique blend of execution, strategy, and engineering. Here is how we built it:

Built an Intelligent Deal Scoring Engine

We developed a scoring system that automatically evaluates every active deal using a normalized 0–100 health score. Using Python-based data processing and Vertex AI, the system continuously analyzes performance signals and highlights which deals need attention reducing manual effort and improving consistency.

Unified Data Across Multiple Sources

Performance data from multiple sources was consolidated into a single, structured data layer using BigQuery and GCP-based pipelines orchestrated via Apache Airflow. This eliminated data inconsistencies and established a reliable foundation for accurate, consistent deal performance evaluation across the entire ecosystem.

Enabled Proactive Monitoring Across the Bid Lifecycle

Monitoring was structured across every stage of the bid lifecycle, requests, bids, wins, impressions, and spend, using orchestrated workflows built on Apache Airflow and Cloud Composer. Rather than waiting for issues to surface, teams now have visibility into exactly where performance is dropping and can intervene before it impacts revenue.

Introduced Real-Time Detection and Explainable Insights

We implemented real-time detection logic combined with event correlation techniques and AI-driven insights (Vertex AI) to identify unusual patterns as they happen. Clear, human-readable explanations helped teams understand issues faster and take corrective action.