Predicting Customer Behavior with AI Will Transform Sales

(This article written by Scott Nelson was published in the NEFA Newsline Nov/Dec 2025 issue)

Artificial intelligence is reshaping the equipment finance sales process from a numbers game into a precision science — changing not just how we sell, but who we sell to.

Hollywood would have us believe that being in sales is basically running across a big field with Mel Gibson wearing a suit, carrying a briefcase, and screaming the company mantra.

“A-B-C. A-Always, B-Be, C–Closing!”
– Glengarry Glen Ross

“People don’t buy stock. It gets sold to them. Don’t ever forget that!” – The Wolf of Wall Street

“There is no such thing as a no-sale call.”
– Boiler Room

Although scripted for drama, these quotes mostly speak truth about the standard operating procedures for sales organizations today—sales is a profession of continuous effort, a contest of conviction, a battle of attrition.

The harsh reality is that an equipment finance business loses roughly a third of its revenue generating portfolio every year to roll off, so sales originations, are critical to maintaining the business—say nothing of growing it. The need for sales is critical in every industry, so sales organizations are often operated with almost military precision and regiment with “demand gen” and SDRs (Sales Development Resources) scouting the market to identify targets for business development and application expert follow ups. Sales is the focus of millions of dollars of training, process analysis and technology spending every year. But a new technology, AI, is going to change how the sales process because it is going to change both the who and how of customer engagement.

Sales Prospecting is always defined the same way: “the process of identifying and contacting potential customers (prospects) who are likely to be interested in buying the company’s product or service.” Yet, most sales organizations approach the challenge in reverse—they contact as many Ideal Customer candidates as possible and then work to identify the ones who are motivated and ready to buy. The process is managed statistically, action-oriented and costly because of the manual nature of converting prospects to buyers. As Hollywood has so succinctly stated, sales takes a lot of hard, manual work to convince customers the time to buy is now.

I argue that AI is poised to change this paradigm.

Good sales and marketing teams define and use an Ideal Customer Profile (ICP) for their offer to drive outreach. An example ICP for an equipment finance company, per Grok, leverages business characteristics that improve the probability that the salesperson finds a match for their offering, e.g., industry codes, asset types, deal sizes, credit scores, etc. Most sales teams use a variety of databases to assemble the longest possible list of companies that fit, or mostly fit, the ICP and then the contacting begins. Ironically, the inefficiency of the process is often celebrated as a demonstration of hard work—“We sent out 2,355 emails resulting in 58 click-throughs and 4 requests for a call.” Always Be Closing!

But any youth hockey coach knows that while shots-on-goal is a good leading indicator, only good shots that score win games. Four calls out of 2,355 “shots” (~ 0.2%) is recognized as an industry average for the SDR (Sales Development Resource) method. No wonder there are so many SDR companies bragging about lower-cost outreach—the throughput sucks.

An irony of sales technology is that software like Salesforce and ZoomInfo has made reaching out to 2,355 “potential” ideal customers not only possible, but relatively easy and affordable. Or at least so it seems today. Software disrupted both the bookstore model—“Come see what we have”—and the traveling salesman’s frequent flyer status—“I visited 15 customers last week.” AI is software that will disrupt this “battle of attrition” model as well.

Start with customer behavior

Note what’s missing in the ICP. Nowhere does it say, “Customer X in Des Moines is planning to buy an excavator around the second week of July.” Next consider the brand-impact of sending 2,355 “ideal customers” emails that they don’t want—repeatedly because ICPs don’t change very often and software never gets tired or discouraged.

What if outreach, the contacting phase, targeted customers who not only fit the ICP but are also planning to buy what you’re selling in the near future? What if customers were identified by predicting their buying behavior? Predicting human behavior has always been the real challenge for sales—“Will they buy this? Are they ready to buy this?”

Predicting human behavior is AI’s superpower. Indeed, predicting human behavior defines AI. Gen AI Large Language Models (LLMs) predict the next letter, word, or phrase that a human user expects based on the human-generated data upon which they were trained. Machine Learning AI models predict human behavior based on past recordings of a specific behavior, e.g., making payments, renewing a contract, or funding a deal. All AI models are trained on data from human activities, so naturally one of the most effective applications of AI is predicting human behavior.

Smarter customer engagements using behavior predictions

AI will disrupt sales with the classic “work smarter not harder” paradigm. Throw away 0.2% conversion rates by meeting customers who are ready to buy with the solution that solves their problem. This is the opportunity of AI, and the good news is that a lot of data is already available today to help predict customer behavior.

Data from customers you already know—your data

Whether your company is strictly an originator or a full-service finance company the data from the operational workflow can support predictive models for customer repeats, buyout decisions, renewals, asset type needs and vendor preferences along with timelines, “when predictions,” for those decisions. Front end data, data from an origination or CRM system, provides the most insight into customer profiles and first-time buying behavior, but servicing data can help with predictions of customer behavior along the life of the contract, e.g., repeat customer or renewal predictions. The following figure shows an example of repeat customer timeline data for first, second, third and fourth repeat scenarios. Note how as the borrower-lender relationship grows, i.e., additional repeat purchases, the decision timelines shorten. Once borrower and lender have come to know each other, both make repeat purchase decisions faster and with more confidence. The normal distribution of the data makes repeat prediction very accurate and as the relation matures the “when” prediction becomes more precise as well.

A lender or originator with such comprehensive repeat customer data set can achieve prediction accuracies in excess of 90% due to the quantity and clarity of the data, i.e., repeat decisions are unambiguous. When-predictor accuracies are strongly affected by the period over which decisions are made, i.e., number of months covered in the  distribution. The data shown above supported accuracies of plus or minus three to five months for the first repeat and tighter timeframes when customer moves to second, third, and fourth repeats. Similar results can be expected for buyouts, renewals and multi-asset decisions. A sales team can use AI-based buyer behavior prediction to make the right call at the right time.

Data from potential customers

The equipment finance ecosystem has terabytes of data online for customers and competitors, all of which the open LLMs are very good at ingesting and analyzing. This data includes public data from news reports, UCC filings, building permits issued, equipment operator job postings, as well as enterprise-connected social media postings. The right prompts can get the LLMs to identify customers of a given industry, using certain assets, in a specific geography, who might need more equipment. Some quick research exposed posts like “We are excited to have been selected to build this new data center and will be ramping up our site prep operations 3x.” Social media can hold a wealth of data on buyer intent that when combined with regulated or state filings goes way beyond ICP identifications.

LLM prompting does not have the precision of models built on an enterprise data set, but they still provide the opportunity for a sales organization to have insights on the buying behavior of new customers withing a marketplace. Of course, the competitive advantage of these tools will, at least initially, be limited to the prompting skills of the investigator, but identifying a customer-is-ready-to-buy is still only the first step, the shot-on-goal. But only those sales teams working smarter with AI are going to find themselves on the field in a position to score to close.

AI will not disrupt sales with higher efficiency and cost savings. That war has already been fought and won by traditional sales software tools and databases. Cheaper reach outs are incremental and won’t capture market share. AI will disrupt sales by completely changing who is contacted to improve the prospect-to-buyer conversion productivity by orders of magnitude. AI enables sales to work smarter in ways that will require new strategies to compete. When asked to “find buyers who are already buying” some may retort, “We manage what we can control and we can’t control customer behavior.” That may be true, but AI can predict it and help the organization stop wasting time on customers who fit a profile but aren’t ready to buy. Those who acknowledge and embrace this AI capability will find themselves meeting buyers at the door with the product they came to buy.