Lead Scoring Prediction

Lead scoring prediction is the process of estimating the outcome of a deal close lost or closed won, with the goal of better understanding, analyzing and acting on the deal. It is based on historical data, and uses machine learning algorithms to provide you with accurate predictions and confidence.

Confidence Parameter

Confidence is a measure of the reliability of the prediction. It can be changed at each model invocation and represents the minimum confidence level that the prediction must have to result as confident. For example if you set the confidence level to 0.8, the model will only return predictions with a confidence level of 80% or higher. You will receive as output the predicted class of a deal, only if that prediction meets the confidence level you requested.

This is very useful for example to have a more reliable budgeting process, where you can set a confidence level of 0.9 to only consider the predictions with a confidence level of 90% or higher, building a more reliable budget prediction.

• Higher Confidence Values (e.g., 95%, 99%):
  • Provide stronger guarantees that the true value is captured.
    • Super useful for example in budgeting processes.
• Lower Confidence Values (e.g., 70%, 80%):
  • Provide bolder predictions, with a lower guarantee of being correct.
    • Useful for example in a lead scoring process, where you want to have more opportunities to close a deal and highlight the most promising ones.

Valid Range of Confidence Values

The confidence value ranges between 0.5 and 1.0, exclusive.

Click to expand

We employ conformal prediction, the confidence score represents a calibrated probability, based on data, that the true outcome belongs to a certain class.

1. Interpreting Low Confidence Values:

Predictions with lower confidence values (e.g., 0.5–0.6) indicate higher uncertainty. While they still favor a specific outcome, their reliability is limited.

Recommendation: Use higher thresholds (e.g., 0.9) in critical workflows, such as financial forecasting, to minimize uncertainty.

2. How Conformal Prediction Affects Confidence:

Conformal prediction ensures that the provided confidence levels are statistically valid and calibrated based on past performance. This means:

If a confidence threshold of 0.8 is set, approximately 80% of the predictions meeting this threshold are expected to be correct, assuming the data distribution remains consistent.

3. Practical Tips for Using Confidence.

High Confidence (≥ 0.9): Use when precision is crucial, such as prioritizing high-value deals or setting sales goals. Moderate Confidence (0.7–0.8): Balances reliability and coverage, ideal for exploratory processes where maximizing opportunities is key. Dynamic Adjustment: Test different thresholds to find an optimal balance for your specific use case.

4. Edge Cases: When Predictions Are Filtered Out:

If the confidence score of a prediction does not meet the specified threshold, the system will return:

• predictedTermination: null
• isPredictionConfident: False

This behavior ensures that low-confidence predictions are excluded from decision-making processes, reducing the risk of errors.

Requesting a Lead Scoring Prediction

To request a lead scoring prediction, you need to have a collection with historical data for deals. You can create a collection and upload your data following instructions in the Collection section.

Once you have uploaded your data, you can request a lead scoring prediction by sending a POST request to the invocation endpoint, with lead-scoring as invocationType. See more at Invoke Lead Scoring.

Interpreting the Lead Scoring Prediction Results

The results of a lead scoring prediction request will provide you with the estimated outcome of a deal and a confidence score for the prediction.

Here is provided an example for a lead scoring prediction.

Suppose you have a set of deals that you want to score.

Example

The outcome of the request to the model will include:

• id: the identifier of the deal.
• predictedTermination: the predicted outcome of the deal, can be closed_won or closed_lost.
  • This field can be null if the model is not confident about the prediction.
• isPredictionConfident: a boolean value that indicates the confidence of the prediction.

The output, in json format, will be:

[
  {
    "id": "1",
    "predictedTermination": "closed_won",
    "isPredictionConfident": true
  },
  {
    "id": "2",
    "predictedTermination": null,
    "isPredictionConfident": false
  },
  {
    "id": "3",
    "predictedTermination": "closed_won",
    "isPredictionConfident": true
  }
]