Introduction

  • 4 minutes

Most organizations improve efficiency by predicting events that could affect their operations. These predictions often guide decisions about things like inventory levels, marketing budgets, and other areas where better forecasting leads to better results.

In this module, you’ll learn how the AI Builder Prediction Model can help you calculate the likelihood of an event happening based on historical data.

AI Builder Prediction Model

The Prediction Model is a type of custom AI model in AI Builder. It works by analyzing patterns in historical data during a process called training. Once trained, the model can predict future outcomes based on new data.

You can use the prediction model in three main ways:

  1. Two-option questions – Where users choose between two possible answers, such as:

    • Were you satisfied with your stay? (Yes/No)
    • What was the pool temperature? (Hot/Cold)
    • Do you prefer to arrive during the day or night? (Day/Night)

  2. Multiple-choice questions – Where users choose one option from several, for example:

    When was the parcel delivered?

    • Early
    • On-time
    • Late
    • Lost

  3. Numeric answers – Where users provide a specific number, such as:

    • Number of rental days: 12
    • Number of rooms required: 2

Using Historical Data

Like other custom AI models, a prediction model needs to be trained before it can be used.

The first step is to identify which historical data can help predict future outcomes. It’s best to include a wide range of relevant information so the model can identify patterns and trends without bias.

Training data must meet the following requirements:

  • Stored in Microsoft Dataverse
  • Under 1.5 GB in database storage
  • At least 1,000 rows of data with a balanced and realistic distribution of outcomes

Model Performance and Usage

After training, AI Builder will assign the model a Performance Grade to indicate how accurate the predictions are:

  • Grade A – Excellent prediction accuracy; further improvement may still be possible.
  • Grade B – Good accuracy; improvements can be made if needed.
  • Grade C – Better than guessing, but adjustments are recommended.
  • Grade D – Either no better than guessing or too perfect (close to 100% accuracy, which could mean the model is overfitting). Data and model settings should be reviewed.

Once your model has a satisfactory grade, you can publish it and start using it with live data.

You can use the model in two ways:

  • Run now – Manually trigger the model to generate predictions on existing data.
  • Real-time prediction – Automatically predict outcomes as new data is entered.

Now that you understand how the AI Builder Prediction Model works, the next step is to learn how to apply it to real business problems.