Data Science Primer: Clone Your Best Salesperson—Machine Learning in B2B Sales

Author by Brian Goodwin, PhD

When you use AI to help guide customer relationship management, machine learning helps you close more leads and optimize the revenue you’re getting from each one.
 
Whereas some AI projects are focused on reducing expenses—for example, by reducing manufacturing downtime—CRM-oriented AI projects are focused on expanding top-line revenue. CRM is a very strong fit for machine learning, because normally you have already gathered a lot of data on your market and your customer base.
 
If your company has been around for a while, you have data that describe past interactions. Information about the length of a sales relationship, the industry your client is in, the nature of the relationship, and the length of your typical sales cycle can all be leveraged into a predictive model. As a result, you can increase revenue and make more sales.
 
The Past Predicts the Future
 
Historically, CRM products assumed the sales team knew their target audience. They were designed as a reporting tool. With a CRM system, you could document relationship status and order history, but you would not be able to estimate the probability that a lead would become a customer. It wouldn’t tell you which customers you should be focusing on or which new products should be sold to them. And it didn’t help you set prices or priorities.
 
Consider all the variables that a good sales leader has to juggle. Some companies have budgets that run out at the end of their fiscal year. Certain account executives pair better with certain leads. Some leads prefer phone, others prefer email. Sometimes a slow-to-buy client has a massive budget and is easy to deal with, but other times they’re likely to be a headache later. Some are open to upsells or cross-sells and others may be interested in an early payment discount. There are a lot of factors you can extract value from—so many that no one person could connect all the dots.
 
Until now, that is. Machine learning can’t close a deal for you, but it can duplicate the intuition of your best salesperson and help you prioritize your sales efforts. An artificial intelligence takes factors like which account executive is responsible for a sale, the timing and length of calls and emails, and even the prospect’s level of enthusiasm as measured by their tone of voice and predicts the probability that the deal will close.
 
“What Does an Elephant Have in Common With a Tree?”
 
They both have trunks! Even simple riddles like this used to be extremely difficult for computers to solve since they rely on human intuition. But modern machine learning setups mirror neural networks in a way few theorists thought was possible. As a result, machine learning is surprisingly effective at “soft skills” like sales.
 
Consider sentiment analysis. Natural language processing (NLP) is a process by which computers can sense the feelings people have around a given product. By instantly “reading” thousands of product reviews and tens of thousands of social media posts for emotionally laden words like “awesome” or “disappointed”, an AI can sense who’s buying what. By sensing subtle “buy” or “won’t buy” clues in emails—or even monitoring tone of voice on a phone call—a machine learning system can sense where a conversation is going and assign probabilities that a deal will close.
 
Now imagine applying this level of analysis to every other data point. Time of day, weather, what the news headlines say about your nearest competitor, how the stock price of your main supplier is moving–it all gets thrown in the mix. The end result is like having a team of excellent salespeople reading every clue to find connections. The AI learns from your past pitches, associated external data, and text-based notes when they succeeded or failed, then it derives a model to predict future sales. Then it back-tests the model against historical data to “simulate history” as a validation of the general system that represents your CRM process.
 
This AI can be built using a variety of approaches such as survival analysis. With a survival analysis, you’re estimating the time to completion as well as the probability of winning a deal. This probability decreases over time, but depending on the type of opportunity, that probability may stay high for quite a long time. With other types of deal, if you don’t close it right away you’re not going to win the deal.
 
Better than “Minority Report”
 
Now you have a number of very realistic and very powerful estimates. First, you have the probability that the deal is going to close in a win. And, it can estimate the time to failure. For example, if you have passed the estimated time to failure and you are still engaging in a relationship with the client, the probability you are going to close that deal is pretty low.
 
If someone on your sales team is spending too much time with a client with a very low probability of closing a deal, you can encourage them to chase better leads. Conversely, a deal likely to close may need just a bit of extra attention to push it over the edge.
 
But AI goes beyond just “closing the deal” and actually gives you information about negotiation strategy. If you have enough historical data, you can use that to estimate what price you should try to negotiate for without losing the customer.
 
You train a machine learning model to look at the data. The AI looks at which clients have generated the most revenue and what deals worked best per client type.
 
Maybe in the past this type of customer has been likely to agree to committing to a year of service in exchange for a price break. Or maybe a better strategy is a price break after a certain purchase threshold.  AI is far more sophisticated than earlier pricing models because it is distilling the actions of all your previous salespeople, then creating a new approach for each client based on similarities with past clients. It accounts for the “human element” and understands how far is too far to push.
 
AI doesn’t replace good account executives, but it supplements them. You could show this model to seasoned account executives and they’ll say, “yeah, I knew that”. But a new salesperson won’t.  Furthermore, this provides a shortcut for sales team leaders and VPs of finance to guide their sales team through hundreds of transactions at once.
 
Ultimately, you’re getting back to hard science and reducing the need for human intuition. Machine learning will help you make more sales by catching subtle correlations, optimizing the specific offers you make, and helping you get clarity on the motives that drive customers to buy or not buy. It’s using the past to predict the future.
 
Author

Brian Goodwin, PhD

Data Scientist

Tags in this Article