These A.I. Startups Want to Automate Sales. Can They?

Most people know about one of the greatest salespeople of all time, Steve Jobs. When Jobs introduced the first iPhone, in January of 2007, he had to convince a skeptical world to pay a then-outrageous sum of $600 for a phone made by a company that had never produced a handset before, with a slick back and no physical keyboard to peck out emails. It was a creative act on Jobs’s part, an ability to craft a vision of how the world would be and to convince people their lives would be better in that world if they bought his shiny new object.

That creative act poses a challenge for a raft of software startups trying to use artificial intelligence to reinvent sales. Companies including Vymo, InsideSales, SalesLoft, and Outreach have gotten hundreds of millions in financing in the last few years, in hopes that by mining historical data such as emails and customer call logs, they can figure out what the best salespeople do.

Sales observers say it’s possible the software programs could bring incremental benefits that will nevertheless prove valuable to companies, but they’ll be unlikely to replicate the finer art of closing transactions.

“The human element is not going away,” says Frank Cespedes, a senior lecturer at Harvard Business School and author of Aligning Strategy and Sales and Re-Thinking Sales: What’s Changing, What’s Not, and Why Knowing the Difference Matters. “The opportunity is to make salespeople even better and more productive at what they do.”

Jonathan Atwell, assistant professor of organizational behavior at the Stanford University Graduate School of Business, agrees. “Count me skeptical that the creative act of people talking with each other to create shared visions of the future can be captured by the current approaches,” he says.

Selling, Atwell notes, “is about people talking to people.” The finer points of the salesperson’s job generally come down to “social skills like creating trust and emotional and social resonance,” observes Atwell. “We can’t make an algorithm understand it yet, because a lot of that behavior looks like statistical noise.”

But that hasn’t stopped entrepreneurs from trying. Having reinvented the “back-office” parts of industry, enterprise software makers are now focused squarely on the front office, mostly “SG&A,” selling, general and administrative costs, which now stand out as conspicuously high with the streamlining of the cost of goods.

Vymo, a New York City startup founded by former McKinsey consultant Yamini Bhat, is focusing on ways to improve the one-to-one relationship of the sales person who has to visit lots of clients on a regular basis. Its customers include many salespeople in the “advisor” type of role, at places such as global bank Allianz and Liberty Mutual Insurance. Vymo’s software can automatically prompt a salesperson when it might be a good idea to check in on a customer, or follow up on an email exchange, among other things.

Premised on the notion that by engaging people with an app, Vymo’s software is able to observe statistics of behavior. “You have to make sure the activity is captured on a regular basis, so you have something to work with,” says Bhat. Vymo was initially developed by shadowing top salespeople for six months when the company was in its development phase.

A slightly different approach is coming from InsideSales of Provo, Utah, which has received $317 million in funding in multiple rounds from Microsoft and from Salesforce, among others, and which is selling to large clients such as American Express. Using statistical tools such as “random forests” and “decision trees,” the InsideSales cloud-based software “knows the patterns of when a client is likely to answer an email” based on historical patterns of thousands of interactions. The goal is to predict behavior. “Are there things that are suggesting buying intent way before they are ready to buy,” is the kind of question the software sets out to answer.

But instead of automating sales, the biggest opportunity for services like these may be improving the performance of people at the middle grade of performance.

“Incremental improvements to the abilities and focus of the average performers usually have, in the aggregate, much bigger impacts on sales productivity than efforts at either end of that spectrum of performance,” says Harvard’s Cespedes. Managers know what to do with their stars and their laggards, he says. “It’s the people in the middle that managers get paid to manage, and that’s also where the technology is likely to have its biggest impact.”

Another low bar is the current software landscape. Every software startup notes that Salesforce’s software for customer relationships is ubiquitous but also almost unused. It’s a filing system, for call logging and such, but salespeople are loath to do the clerical work of logging data to the system. Tools like Vymo are a way to automate that, to remove some friction. “In a CRM world, people are not even using the tool, and a manager is asking every week, where is the deal, and it’s not working,” says Bhat.

But ultimately, tools using A.I. are no better than the goals set for a sales program, at least that’s what software startup Aible of Foster City, Calif. proposes. Operating on the premise that most companies need some hand-holding to figure out their goal, Aible’s program, called “AutoML,” will pick a machine learning model only after helping the company articulate a likely return-on-investment.

“Most A.I. is based on accuracy,” observes Aible founder Arijit Sengupta, a veteran of Oracle and Microsoft. But accurate predictions can nevertheless be misleading. “Take a situation where you have a 10% win rate in sales,” Sengupta says. “The most accurate prediction of a machine learning model would be very conservative, to predict that most customers are not going to buy from you.” But that model would “destroy the business” by encouraging the pursuit of far fewer customers.

Even with better objectives, it will be hard to capture a complex dynamic that eludes simple quantification. “The most important thing about selling is the buyer,” says Harvard’s Cespedes, going back to the days of the ancient Agora in Greece. For a complex item such as capital goods, the bar is especially high. “Those buyers still want to deal with knowledgeable sales people who can work with them in navigating the process and justifying the decision to and with colleagues,” he adds.

The gift of gab, knowing when to listen and when to speak, when to gently nudge or when to apply pressure—these remain the mysterious elements of the salesperson’s art that no amount of analyzing historical data can capture. At least not yet.