Imagine feeding known data into a black box and then analyzing the output. When you didn’t get what you wanted, you simply adjusted a few knobs on the box and fed it a different data set and monitored the output again. You would repeat this process of adjusting the knobs on the box until the black box gave you the output you expected. This is what is called ‘supervised learning’ in the world of Machine Learning (ML) a subset of Artificial Intelligence (AI).
Once the box did give the expected outputs, it is considered ‘trained’. This finely tuned black box can now be used to predict the output of other data sets.
In sales, imagine feeding a similar black box information on past customer purchases. After a few iterations and fine tuning of the knobs, the black box would be able to ‘predict’ who is more likely to buy (i.e., customer personas) and what are they most likely to buy (i.e., upsell or cross-sell).
For example, If you have over 1,000 customers who have purchased a given product over the last 6 months, you may want to know who out of those 1,000 customers is more likely to buy your new product. Now, you could just market to all 1,000 but that might prove to be costly and diminish you Return On Investment (ROI).
So you’re faced with three big questions:
- Who should we market to?
- How should we reach out to them?
- What should we send them?
Another feature of a Propensity Model is that it allows you to segment your customer output by likelihood of buying using what are called ‘deciles’ which allows you to group customer in 10% increments. The top 10% of more likely to buy than the next 10%. and then the next 10% and so on.
In the case of 1,000 customers, the top 10% (100) are very likely to buy which means that you need to put your best sales foot forward. You may decide that this group is worth sending your best product catalog.
The next 10% (second decile) is likely to buy but to a lesser extent than the top 10%. So maybe instead of sending the next 100 customers a product catalog, you may want to call them.
The next 10% and down (third to tenth decile) may or may not buy so you may just want to send them an email promotion.
Propensity Models allow marketers and salespeople to hone in on likely buyers which will reduce marketing cost and reduce Client Acquisition Cost (CAC) when targeting new buyers, respectively.
It’s worth highlighting that Propensity Models work far better than the old-school RFM model which simply look at a customer’s past history along three dimensions: Recency (last time they bought), Frequency (how often they bought) and Monetary (how much did they buy). A Propensity Model, with the help of AI, can look at tens if not hundreds of parameters when making a prediction. An RFM can tell us what happened in the past but does little to inform us as to what will happen. Propensity Models are on average 30%-50% more accurate than an RFM model.
What does all this mean to you as a salesperson? With the help of marketing and a data scientist, salespeople will be able to target their customer base more effectively and spend less time on poor opportunities. You’ll be able to:
- Develop relevant and timely customer communications
- Increase your bookings and wallet share in your vertical market
- Target your customer with the ‘next best’ product or service solution
The future of Cross-Selling and Upselling will be the use of ‘trained black boxes’ (i.e., Propensity Models) that are driven by data from your Customer Relationship Management (CRM) system and other data sources within your organization.
Victor Antonio, Co-Author of 'Sale Ex Machina: How Artificial Intelligence is Changing the World of Selling.