The world of the salesperson is filled with maxims of success: "You don't sell a guy one car, you sell him five cars over fifteen years." In the rough and tumble world of David Mamet's Glengarry Glen Ross, the play from which that line comes, such sales tactics are part instinct, part hard-won savvy.
A startup company named InsideSales argues such expertise can be figured out over time by mining sales data, and applying some rudimentary forms of machine learning.
David Elkington, founder and chief executive of InsideSales of Provo, Utah, was in New York last week for his own customer conference. He sat down with ZDNet to explain his thinking. The venue for the meeting was Elkington's suite at the top of the Beekman Hotel in lower Manhattan.
The pointy-peaked tower duplex suite in the elegant 19th century hotel typically rents for $6,500 per night. The inside of the tower resembles some sort of cozy Swiss chalet, or, as Elkington observed, something out of Hogwarts -- a milieu that seemed to please his nerd side.
Elkington, who studied philosophy as an undergraduate, was in the midst of pursuing a masters degree in computer science, with a focus on machine learning, in the late '90s at Brigham Young University, when he put aside his thesis to form InsideSales. Back then, his ambitions for AI were grand. "I assumed we could emulate the human brain," he says with a grin. "I was naive, perhaps."
His thesis advisor for that masters is now working for Elkington at InsideSales. The goal today, however, is more prosaic than back then, but to Elkington it is of no less value, perhaps more so, to the company's customers.
InsideSales plugs into companies' customer relationship management, or "CRM" systems, such as that of Salesforce.com, and ingests vast amounts of customer data -- 70 million company profiles, 200 million buyer profiles, and data on 10 billion sales transactions, a total of seven trillion data points. "We consume the entire CRM data set," he says.
Elkington likens the system to Google's Waze traffic app: "Waze is only good because there are all these people ahead of me" on the road. Likewise, transactions data shows what deals closed and perhaps provides clues as to why they were able to be closed. Are there things that are suggesting buying intent way before a customer is likely to buy? That's the kind of question the software is intended to ponder.
His chief technology officer, Ryan Allphin, explained: "If there was a sales transaction with you, the customer, how many times have they called you? And what time of day, and what season of the year?"
"How did you proceed down the sales path as the customer? Maybe they first reached you on the phone, and then they sent a couple emails, and then you answered the second email." The company's software is able to do things such as track if a recipient opened an email and when.
By looking at a network of interactions from the millions of buyer profiles, over years, the system finds patterns, things such as which email prompted a response, which did not.
Over time there's something of a predictive aspect that arises. "Our system knows the patterns of when they're likely to answer," says Allphin.
The data here is key, says Elkington; the actual statistical procedures are simple, and they may vary from one task to another in InsideSales's software.
"Data is definitely the new oil," he says. In contrast, "Most of the math [underlying AI] has been around for 60 years."
Elkington knew from the outset that the problem of sales was about data, not algorithms. The first decade, however, was spent mostly selling productivity tools to run on top of CRM, and gradually amassing data.
"After about ten years - it was interesting - there was this tipping point where there was just enough data," he says.
That's when the company began to land big customers such as American Express and Cisco Systems. To fuel that march into the enterprise, the company has received a hefty $317 million in multiple rounds from a stellar collection of tech giants and venture capital firms: Microsoft, Salesforce, Hummer Winblad, Josh James, Polaris Partners, Kleiner Perkins, U.S. Venture Partners.
The work with each client goes into a pool of anonymized data that can then feed all of InsideSales's customers -- an increasingly common practice among startups in machine learning.
"I'm comparing your data to the corpus of data I have gathered over 15 years." The actual AI technology is not as complex as deep learning. "We use random forest algorithms to see co-occurrences of multiple data points," says Allphin, "a certain decision tree might be better for this customer in this industry."
And there's a certain purposefulness with which the company applies basic statistics that has more to do with the data and the problem to be solved than with machine learning per se, Allphin indicates.
"Via association rule mining, we can go beyond simple binary suggestions such as, here's a prospect you might want to call. In an organization, you might have only one contact, but we can tell you the others your colleagues also contacted, at that same company, and if they have a propensity to buy."
The single "highest correlation" among the various data the system has turns out to be a salesperson's notes, says Elkington. "Just a salesperson's notes on the prospect, there is rich data there."
All of this, says Elkington, dovetails with companies' increasing propensity to hire data scientists to do just this sort of pattern-deduction. "Companies are saying this is the problem we are trying to solve, and with them we are looking for those common elements."
Through A/B tests, where one group of salespeople are assisted by AI, while another serve as a control group, doing their normal day's work without AI, companies can see the benefits of the technology, says Elkington. Some customers have seen a $50 million revenue "uplift" after implementing the technology, he claims.
But can it truly recreate the instinct of sales?
Elkington is very aware that machine learning is nowhere near replicating higher forms of reasoning that humans possess. But by tracing patterns such as when calls or emails can be answered, some of the acquired knowledge of sales people can be passed along to the new recruits.
"The good news is, people are really bad at selling," observes Elkington. "People are really awful at it, for a lot of reasons. Last year, 53 percent of sellers hit their quota, that's down from 63 percent just five years ago."
"So for us, it's easy to create value."
Some of the problem is simply the very green nature of the sales force: As of a year ago, more than half the people who are selling are millennials, he notes.
"A lot of the best sales people know the best practices," says Elkington, who counts Salesforce founder and CEO Marc Benioff as a friend and mentor. "They have so much intuition that a millennial doesn't have, but they can't be on the floor all the time, coaching each seller."
Perhaps, then, a machine can't on its own be a salesperson, but digitally savvy millennials, or digital natives, may be equipped to be better sales people over time with smarter software. At least, that's the idea.
Previous and related coverage:
An executive guide to artificial intelligence, from machine learning and general AI to neural networks.
The lowdown on deep learning: from how it relates to the wider field of machine learning through to how to get started with it.
This guide explains what machine learning is, how it is related to artificial intelligence, how it works and why it matters.
An introduction to cloud computing right from the basics up to IaaS and PaaS, hybrid, public, and private cloud.