MadKudu’s AI scoring surfaces insights from your historical data to determine what makes a lead likely to convert into a customer. To generate accurate insights and train a reliable model, some foundational setup is required.
How Audience and Conversion definition interact
When defining your conversion criteria, follow these best practices:
Choose a conversion definition as far down the funnel as possible — ideally "Closed Won" customers. This reduces the risk of the model learning from noise or biases in your sales process. An earlier funnel stage may be necessary if there are volume limitations.
Separate New Business from Expansion/Renewal opportunities. This avoids double counting and ensures the model is trained on net-new customer acquisition.
Filter by deal size if your product offers a wide range of plans or pricing tiers. For example, if some deals are <$100/month and others are $10k/month, set a minimum deal size in your conversion definition to avoid the model optimizing for small, less valuable deals.
Ensure you have enough volume. Your conversion cohort should ideally include 200–300 opportunities. This provides a large enough sample for statistical relevance while keeping the model focused.
Consider the Quality and Recency of Your CRM Data. If you've recently completed a data cleanup or CRM migration, it’s often best to start your dataset from that point forward. Historical data that predates standardization or process changes might introduce inconsistencies or noise that negatively impact the model.