A predictive score model is a formula to calculate a probability.

There is a 70% chance that you will read this entire article. How do I know this? Because I used a predictive score model. The score is the probability of you reading to the end of the article (or one minus the likelihood that you will not – the exact opposite).

My example in the paragraph above is a well-known application of predictive analytics in marketing. The most common examples in business-to-business (B2B) sales are lead scoring, churn (or customer attrition), cross-selling, and pricing.

Score models can be rule-based, machine learning-based, or a combination of both. And finally, remember, a model itself is only as worthy as its application. In other words, developing the model is only half of the job. Applying the model is what makes it valuable.

Let’s review these points in detail.

Predictive Score Model: Technical Definition & Example

A predictive score model is a mathematical methodology to calculate a probability of an event happening in the future. This score is usually a number between zero and one describing the likelihood of an event. The steps how to calculate this probability can be seen from the model itself.

To create a predictive score model, you first need to understand what it is that you are predicting. You also need to select the data to make your predictive score model. If you are using machine learning, this last step is known as “training a model”.

What you are predicting with the score is typically discussed in terms of classes. What is a class? A class defines an element belonging to a group. For example, will customer ABC buy the product XYZ? Or will customer CDF churn? The customers are the elements and the classes are buying product XYZ or churning respectively. There are, of course, more sophisticated ways of defining classes.

To create any score model, you will need historical data regarding your element and your classes. In the example I mentioned above, you will need customers who have bought the product XYZ or do not buy it anymore. These customers are your targets or labelled elements. Your score model will now take more data and predict the likelihood of your customers belonging to any of these classes.

For example, a predictive score model for customer attrition will give a score (a value between zero and one) of all your customers belonging to the class “churned”

Similarly, a predictive score model for cross-selling will be a value between zero and one for any combination of customer-product. Both scores are helpful to retain customers or to increase sales (or together in cooperation).

Let’s see how you can create your predictive score model.

How to Create your Predictive Score Model?

Historically, you might be acquainted with predicted score models for lead scoring, churn, or cross-selling. The list does not end there.
For example, let’s use lead scoring to illustrate how the principles of the predictive score model can be applied.

“Lead” is usually the name we give to a viable new business. In this case, the score would be the probability of this new business happening in the future. Most B2B sales practitioners are familiar with the lead scoring concept. They use this score, for example, to prioritise customer contacts.

You can calculate a lead score in different ways, depending on the amount of data you have available and your technical resources. For example, if you have ten leads – ten possible new business opportunities – the first thing you need to know is how to differentiate them from one another.

In machine learning terminology, the different ways of distinguishing these leads are called “features”. A feature is a characteristic that sets apart one element from another. For instance, a feature could be how big the prospect company is or how many Euros you assign to the opportunity. Depending on the number of features and the number of leads, you can create different models.

What is a Rule-Based Scoring Model?

The most used scoring model is usually rule-based. Rule-based models work, as their names imply, using rules. Humans, as opposed to computers, create these rules. They could be simple or complex rules. A rule is a way of assigning a value to the score.

We are all familiar with rule-based models. For instance, if your lead belongs to a particular industry, you might assign a higher likelihood to that lead. This probability might come from your past knowledge or the experience of your sales managers.

Rule-based predictive models are straightforward to create and could be helpful for a limited amount of data.

If the number of elements or leads and the available features increase, a rule-based model might not be sufficient.

AI-Based Predictive Score Model

Imagine you are a company serving 50,000 customers closing around 5 to 10 new businesses per month with each of them.

Furthermore, imagine that there are numerous features and characteristics for these companies. In this case, only specialized software can create the necessary rules to calculate the probability of closing new businesses.

This methodology is known as machine learning – an AI discipline.

Machine learning is, in simple words, a much more sophisticated way of deciding the likelihood of an event happening. There are different machine learning methods one can use to calculate a probability or a score. It does not have to be a neural network. Decision trees, boosted trees, or gradient trees methods will do the job.

To increase the performance and efficiency of these methods, data scientists frequently create hundreds – if not thousands – of models. Finally, models compete and complement one another to find an optimal combination.

Empirical research has shown that the best predictive score models use a mixture of rule-based models and machine learning once complexity increases. This combination supports a human agent implementing the prediction and gives salespeople with the ability to look into the future. An oracle of numbers at the service of your team.

The question now arises: Once you have a perfect oracle, should you fire your salespeople? Or how should you use a predictive score model in sales?


How to use Predictive Score Models in Sales

Companies deploy predicted score models in different ways. The typical strategies of using predicted score models are automatic, manual, or in augmentation.

A product recommender in e-commerce is, in essence, a predicted score model working automatically in sales. The recommender system is a predictive score model that returns the likelihood of each customer buying each product. Amazon, the e-commerce titan, among others, employs these kinds of models to suggest which additional products you could buy. They earn a quarter of their revenues in this way.

Companies in distribution and manufacturing apply them as well. They use them to predict which additional products each customer can buy from their extensive catalogue.

Using predictive score models automatically means that the outputs of the models will not be reviewed by people. They will go straight to an application where a person (or a machine) will consume them.

Another example of automatically deployed predicted score models are pricing models for air tickets. With the volume of predictions that the model needs to create, humans can only supervise the model itself but not each prediction.

On the other extreme, manually applied predictive score models are models that will not be automatically deployed. Humans will review the score beforehand. For example, in sales, some companies create models to predict pricing classes. Sales personnel then check the pricing classes, categories, or discounts. Finally, they negotiate with each customer. The end-customer never sees or uses the score model itself. As for the salespeople, this predictive score model only takes away a part of the job they must do. Negotiating and setting the appropriate pricing technique is still their job.

The Autopilot for B2B Sales

Finally, augmentation is somewhere in between: a predictive score model’s successful application and employment to support a human-based decision.

Like the manual application, the critical difference resides with the idea that a human will have to decide “operationally” based on the model (or several models). “Operationally” deciding means deciding continuously.

Think about them as an autopilot and a pilot. The autopilot will fly the plane. On the other hand, the pilot will continuously review its instruments and determine the best course of action.

Investing in B2B sales is a complex task. Predictive score models are usually used for augmentations to assemble models. Examples include customer churn, pricing, and cross-selling. A scoring model only predicting attrition will not be enough to make a decision. So, the human actor still needs additional predictive models to select the best course of action.