The Microsoft Business Applications Applied AI group built its tournament bracket by applying the same data-driven, predictive techniques it uses in business applications. Here’s how the bracket was constructed: - **Historical data**: The team used over 10 years of historical data on college basketball teams. - **Player and roster factors**: They analyzed current player composition and explicitly excluded injured players who would not be available for the tournament. - **Coaching performance**: They factored in the historical competitive success of the head coaches leading each team. - **Game context**: They looked at how teams performed away from home and at neutral sites, which is critical because all men’s tournament games are played on neutral courts. - **Seeds and conferences**: They incorporated historical analyses of how different seeds and conferences have performed over time. Using these inputs, the model identified underrated teams and likely upsets. For example: - It flagged **5th-seeded Houston** as significantly underrated and correctly predicted their upset wins over **4-seed Illinois** and **1-seed Arizona**. - It predicted **8th-seeded North Carolina** would upset **1-seed Baylor** and **4-seed UCLA**. By the quarterfinals, each of the four teams the model predicted to reach the semifinals was the stronger remaining seeded team in its matchup, and all four advanced. As a result, the bracket was: - **Ranked 21st** out of **17 million** brackets submitted to the **2022 ESPN Men’s Tournament Challenge** after four rounds (semifinal stage). - More accurate than **99.9997%** of all brackets submitted at the start of the competition. This same modeling approach—using rich historical data, context-specific features, and continuous updates—is what underpins many of the AI capabilities in Microsoft Dynamics 365.

The same predictive techniques used to build the tournament bracket are applied to real business problems in Microsoft Dynamics 365. Instead of predicting game outcomes, the models estimate the likelihood of different business outcomes. Typical questions businesses can answer with these AI capabilities include: - **Customer behavior** - What is the likelihood that a customer will churn? - What is the predicted lifetime value of a customer? - Which products should be recommended to a specific customer? - **Supply and operations** - Will there be enough supply of a given product on the shelf? - How should inventory be managed from procurement through fulfillment? - **Customer and market insights** - What are customers saying about our products? - How do we identify key business entities in text (e.g., companies, products, locations)? - How do we surface the most relevant news about suppliers? - **Process optimization** - How do we analyze business processes, automatically label activities, find bottlenecks, and suggest improvements? The modeling approach is similar to the sports example. For instance, consider the classic **Recency, Frequency, Monetary (RFM)** model for predicting churn: - **Recency** - In basketball: the model looks at recent performance, such as streaks and last 10 games. - In business: it looks at how recently a customer has made a purchase. - **Frequency** - In basketball: it considers the number of wins a team has. - In business: it considers how often a customer buys. - **Monetary** - In basketball: it can be represented by point differential. - In business: it’s the amount a customer spends. These are just a subset of the features used, but the principle is the same: combine relevant historical and contextual data to estimate the propensity of a particular outcome—whether that’s a team winning a game or a customer churning. In Dynamics 365, many of these models are **out-of-the-box and trained on your data**, so predictions are tuned to your specific business. The goal is to help you: - Retain customers more effectively. - Design loyalty programs that reward your best customers. - Show customers the most relevant products that are actually available. - Improve operational decisions across supply chain, finance, and customer engagement. These capabilities are part of a broader shift toward **agentic business applications** across Microsoft 365, Dynamics 365, and Microsoft Power Platform, helping organizations reimagine how they work, make decisions, and connect data, people, and processes.

Microsoft’s approach emphasizes three things: adaptability, explainability, and human-in-the-loop decision-making. 1. **Adaptability to changing conditions** Business environments and sports seasons both change quickly. In the tournament example, the team: - Incorporated **early results from additional postseason college tournaments** before the main bracket play. - Used those results to adjust the modeled strengths of conferences. For 2022, this included: - Observing **Virginia upsetting Mississippi State** and **Wake Forest winning** in another tournament’s first round. - Interpreting these as signals of a stronger **Atlantic Coast Conference (ACC)**. - Increasing the model’s confidence in ACC teams, which contributed to correctly predicting a **Duke vs. North Carolina semifinal** and a **quarterfinal run for Miami**. In business, the same principle applies: - Models are updated with new data so they can respond to changes in customer behavior, supply chain disruptions, staffing shifts, and market conditions. - This supports scenarios like adjusting inventory strategies, rerouting customer support calls based on real-time staffing, or rethinking procurement and fulfillment plans. 2. **Explainability of AI models** Microsoft invests in making models explainable so users can understand why a prediction was made. This is important for: - Marketing teams planning campaigns. - Operations teams building material requirements plans. - Leaders making decisions about inventory, customer engagement, or financial strategy. Instead of presenting a prediction as a black box, Dynamics 365 AI capabilities: - Highlight the key drivers behind a prediction (for example, which behaviors are contributing most to churn risk). - Provide insights that users can interpret and challenge. 3. **Human intuition and guidance** The goal is not to replace human judgment but to guide and enhance it: - AI models are designed to be **not fully self-service** in the sense that they welcome human oversight. - Users can combine model insights with their own domain knowledge, context, and strategy. This approach extends across Microsoft’s business applications: - **Dynamics 365** uses AI to help organizations reimagine inventory-to-deliver processes—from procurement and production to fulfillment and customer satisfaction. - By linking people, data, and processes, organizations can drive efficiency, reduce costs, and build more connected operations. - Independent studies, such as those from Forrester, indicate that organizations using Dynamics 365 ERP can see meaningful ROI and relatively short payback periods (for example, more than 100% ROI and around a 16‑month payback for midmarket organizations), underscoring the practical value of these AI-driven improvements. Overall, Microsoft’s AI in Dynamics 365 is built to adapt as conditions change, explain its recommendations, and support people in making better-informed decisions rather than replacing their expertise.