Sometimes, the very idea of sales force automation (SFA) can seem like a contradiction in terms. Thirty years into the SFA revolution, a stubborn remnant of the sales process is still human-labor-intensive. Frustratingly, sales force automation is not yet completely automated.
“Nobody wants to listen to 25 phone calls to get caught up on a deal, nobody wants to read 30 notes. Nobody wants to read through all your emails. Everybody wants the Cliff Notes version of that story,” argued Kris Hartvigsen, founder and CEO of Dooly, which offers automation services for Salesforce.
The unpaved last mile in SFA is this manual, tedious dimension. It is the point at which the information stored in the sales person’s brain must be ingested by, integrated with, and stored in a CRM application. It is the point at which the salesperson sits down – usually in front of her laptop – to distill the results of dozens of in-person and virtual meetings. The point at which the salesperson must review and take action on the basis of dozens of SMS messages, emails, or voicemails.
It is the point at which, so far as most sales people are concerned, sales automation technology shifts from being assistive to being tedious, entailing hours of robotic data entry, pointing and clicking, etc.
The good news is that this seems poised to change. Hartvigsen and Dooly propose to automate at least a part of this process: to keep pushing that stubborn, human-labor-intensive remnant further and further out. To this end, Dooly uses ML-powered speech and text analysis capabilities to parse voice mails, SMS messages, emails, etc.; identify and capture events; create or update contacts; harvest data that is required for specific predefined fields; create action items; and capture a salesperson’s notes.
Behind the scenes, Hartvigsen claimed, Dooly’s service transfers or syncs this information with Salesforce: populating fields, updating contact information, scheduling tasks, etc.
“The idea that we thought [about] was, ‘Okay, well, instead of having an Evernote which is another data silo or, you know, Salesforce [which is] another data silo, why don’t we actually just make it so all those data silos actually talk to each other and broker the information around?’” he told host Eric Kavanagh during a recent Inside Analysis podcast. “We wanted to create this … connected workspace that ties together all of the people and systems behind the scenes that need access to the information that you’re capturing and try to deliver to them, without you having to leave that high value mode of selling.”
The thing is, Hartvigsen and Dooly are hardly alone.
Getting the most out of the human factor: intelligence and empathy
Automation in the context of sales force automation works best when it is used to complement human creativity, intuition, perspicacity, etc. This is something SFA software sometimes got badly wrong. So, for example, analytics-assisted SFA software sought to assist sales people by “prompting” them with insights – in some cases, overwhelming them with prompts. In some cases, a well-timed prompt is useful; just as often, however, experienced sales people took them to be banal, irritating, or woefully misguided.
This approach to intelligent SFA sought to reduce, rather than to accommodate and promote, the role of human empathy and intelligence in the sales process. This is actually a well-known problem.
Almost 40 years ago, cognitive scientist Lisanne Bainbridge identified this as one of several “ironies of automation.” In practice, Bainbridge observed, most ostensibly “automated” systems are actually maintained by human supervisors. This is one irony. Another, Bainbridge said, is more subtle – insidious, even: in the most advanced automated systems, the role of the human operator is of paramount importance; ironically, however, the remit of human supervision tends to be more tedious than not.
In other words, Bainbridge argued, these systems delegate to their human operators “an arbitrary collection of tasks, and little thought may have been given to providing support for them.”
This irony is not lost on Howard Brown, CEO and co-founder of ringDNA. Like Dooly, ringDNA uses ML-powered speech and text analysis software to parse vocal statements and textual communications to glean insights, generate facts, and identify possible events, tasks, etc. “[ringDNA] tracks emails, tracks SMS, tracks phone calls, tracks the words that are said in the phone calls, as well as the sentiment. It also tracks the marketing content and then intent data from third-party engines like 6Sense or Demandbase, so we take all of that information to really inform the rep to better serve the buyer,” he told Kavanagh. “At the same time, we’re taking a historical snapshot of those phone calls, so … it’s not just in the moment, it’s also post-call analytics.”
Brown emphasized that the challenge with automating SFA is two-fold: first, avoid inhibiting the instincts of sales people; second, do not get in the way of the sales process itself.
“What’s really important is understanding the specific buying situation and [that] not all buying situations are the same. When you have a sales development rep and they’re making a first outbound call to one of their prospects, the information that’s important to them is very different than an account executive that’s dealing with a procurement officer and trying to get that deal over the line,” he noted.
Anticipatory sales force automation
Aisera, developer of what it dubs a “conversational” AI platform, wants to shift the posture of SFA from what co-founder and CEO Muddu Sudhakar sees as a largely reactive, partially labor-intensive model to more of an active, anticipatory enterprise. One way to shift from a reactive to an active posture is to chip away at SFA’s stubborn remnant of human-labor-intensive tasks. And one way to shift to an anticipatory posture is to automate the identification of verbal or textual signals that indicate interest in a product, demand for a product, or intent to buy a product. “You want to capture the signal right in the wild, wild, west – right when the people are [there] on your webpage or [right where] whatever conversations are happening. Conversation is not happening in Salesforce,” Sudhakar told Kavanagh.
Instead, Sudhakar pointed out, conversations are happening in Slack – which Salesforce purchased – in Zoom, in Microsoft Teams, in Google Meet. Sometimes, he noted, early sales signals are embedded in casual discussions, as when the participants in a Zoom call chat amongst themselves prior to the start of a meeting. “So, how do you find these collaborative signals where the user information is there? How do you take the signal? That is what Aisera specializes on,” he explained.
Sudhakar cited Aisera’s work with Zoom, which he claimed serves as a proof-of-concept of this idea.
“What they have done is their billing use case, billing, renewal, subscriptions: all of that is automated” using Aisera’s conversational AI technology, he told Kavanagh. “We automate all the billing invoices and sales requests that come in to them, whether it comes through any channel: it could be through a chat channel, it could be through email … it can be through people going into their CRM system.”
Ironies of sales force automation
At its most ambitious, this vision of fully automated SFA seems to cast AI into the role of an almost literal god-from-the-machine: AI-assisted SFA as a means of ameliorating, obviating, or smashing long-standing economic, technological, physical, even social constraints. This seems optimistic. After all, behind the curtain of what we like to call “AI” there lurks an assemblage of mostly ordinary components: true, some ML modeling methods are highly complex – opaque, even – but the guts of today’s “lite” AI is powered by rules engines, scheduling and automation software, alerting and monitoring software, data movement and integration software, etc.
Individually, none of these things is new; however, their use in combination is new, different, and powerful, as is the scale at which they are being taken up, deployed, or productized in business and society.
That said, absent something like true artificial general intelligence, the “AI” of today is probably not the stuff of a paradigm shift. For the foreseeable future, then, it is likely that the effects of AI-assisted SFA will conform to (and be limited by) well-known constraints, such as Pareto efficiency.
In spite of this, AI-assist technologies really do have the potential to chip away at the stubborn remnant of human-labor-intensive tasks in sales force automation. The thing is, AI-assisted SFA will almost certainly have a human supervisory component, too: for example, human beings will be enlisted to review, vet, and, if necessary, correct the gleanings of AI-assisted SFA.
To be sure, AI-assisted SFA will chip away at even this remnant of labor-intensive activity. So, for example, as sales people reject poor or mistaken insights, AI software will learn (to a degree) from these mistakes; moreover, the companies that market AI-assisted SFA software could develop and refine automated data cleansing and data quality routines that (e.g.) vet contact information against known databases. In spite of this, human beings – although probably not sales people themselves – will still be delegated the banal task of monitoring the operation of AI-assisted SFA software, redressing issues when they can, escalating them when they cannot. In practice, this means human beings will be enlisted to point, click, swipe, or otherwise manipulate different kinds of computer input devices to fix the things the AI gets wrong.
This goes to another of the ironies of automation identified by Bainbridge in her seminal 1983 paper: the fact that the primary human role in any highly automated process is (in effect) to babysit the behavior of automation technology. It turns out that this is at once the most important, the most difficult, and, ironically, the most banal of jobs, Bainbridge noted. “We know from many ‘vigilance’ studies … that it is impossible for even a highly motivated human being to maintain effective visual attention towards a source of information on which very little happens,” she wrote. “This means that it is humanly impossible to carry out the basic function of monitoring for unlikely abnormalities.”
Bainbridge was talking about humans who are tasked with monitoring control systems for anomalies. By definition, these kinds of things happen only rarely, which makes it difficult for humans to sustain focus. But this is true, too, of tasks that engage or appeal to human habit as against human intelligence: tasks such as repetitively pointing-and-clicking a mouse to resolve issues flagged by a machine. And her larger point – that automation tends to eliminate certain kinds of manual tasks that appeal to human intelligence and creativity – seems as valid in an age of “lite” AI as it was in 1983. “By taking away the easy parts of his task, automation can make the difficult parts of the human operator’s task more difficult,” Bainbridge wrote.
About Vitaly Chernobyl
Vitaly Chernobyl is a technologist with more than 40 years of experience. Born in Moscow in 1969 to Ukrainian academics, Chernobyl solved his first differential equation when he was 7. By the early-1990s, Chernobyl, then 20, along with his oldest brother, Semyon, had settled in New Rochelle, NY. During this period, he authored a series of now-classic Usenet threads that explored the design of Intel’s then-new i860 RISC microprocessor. In addition to dozens of technical papers, he is the co-author, with Pavel Chichikov, of Eleven Ecstatic Discourses: On Programming Intel’s Revolutionary i860.