InsightSquared provides a flexible model configuration and report generation process. This allows tweaking the model to be specific to your business, to increase accuracy and clarity.

Note: Managing your ML models does require your CSM and IS2 Support, but it's a straightforward process we've documented here.

Managing Your ML Models

There are several components of your ML models that must be managed:

Input to the Model

• What field do you split by?

• What fields do you incorporate in the ML model generation?

Output

• Review the output report for accuracy and predictability

Input to the Model

• You can read more about changing your split by here.

• You can read more about adding fields to the model config here.

• You can also add and remove fields from the model generation process.

  • Note: Contact your CSM or IS2 Support for this

InsightSquared provides an interface to manage the ML model for you. In the example below, you can see the fields available to be added or removed from the model.

Note: adding and removing fields from the model should be targeted to IMPROVE THE ACCURACY OF THE MODEL. Let the data tell you what to use.

Adding a field to the model is as simple as checking the box.

Note: if you don't see a field you want, review this.

Before adding "responded campaign names" to the model:

After adding "responded campaign names" to the model:

InsightSquared will automatically generate a new model when changes are made, or once a week. Each model produces a new output report (you can request this from your CSM).

Output Report Review

You will receive a model output report from your CSM. There are two sections that will need to be reviewed:

Model Accuracy Assessment

Note: >90% is acceptable for management purposes, >75% is directionally correct, but will have too many false positives/negatives for management.

• >75% accurate directionally correct models can be used for rollup reports, but not record level inspection

• >90% accurate models can be used for record-level inspection and coaching

Model Details

This section of your report contains more details of the model built for each split.

• Feature: The name of the opportunity attribute. These are derived from your data and buckets by the most ideal intervals as determined by the algorithms.

• Weight: How impactful the feature is in the calculation of the prediction.

• Count With: Number of Opportunities of this type that contain the feature.

• Count Without: Number of Opportunities of this type that do not contain the feature.

• Amount With: Sum of Opportunity amount of this type that contain the feature.

• Amount Without: Sum of Opportunity amount of this type that do not contain the feature.

• Win Rate With: Win rate for opportunities of this type that contain the feature.

• Win Rate Without: Win rate for opportunities of this type that do not contain the feature.

Review the features and splits for reasonableness. Machine Learning isn't smart enough to understand business context-sometimes the data is predictive, but not reasonable.

Note: Data that changes after opportunity close MUST be removed from models (i.e Contact roles that are added after a deal closed-won will distort model predictions).

Note: Any feature with 100% win rate should be reviewed and potentially removed from the model.

If you run into issues, please contact support.