If you ask me whether artificial intelligence is overhyped or underhyped, my answer has to be: both.
AI is not the universal solvent that makes all problems melt away, and many of the most promising technologies are still in their infancy. For all the Big Data available to us, we are only making sense of a small fraction of it.
On the other hand, while I see vast swaths of unexplored territory ahead, I am fortunate to work on one of the frontiers where AI (and its machine learning little brother) are proving their practical value. It turns out one of the most valuable things software can do is understand its users a lot better. In somewhat the same way that autonomous car software developers are working their way up from the basics of understanding driving and traffic – first winning mainstream trust for their ability to assist drivers with accident avoidance and parallel parking before progressing to fully autonomous driving – my team is leveling up with how well we understand sales and sales management and the tasks our users need to accomplish.
A big part of making those interactions more natural and productive is understanding the way people speak and communicate in free-form text, rather than structured forms designed for the convenience of the computer.
I believe what is even more important than voice recognition is for the software to understand your words in context. That means going beyond the literal meaning of words and sentences in isolation to understanding ongoing interactions. It’s the difference between dictation and transcription software, versus truly capable voice software. The software must also understand the context of you – who you are, how you talk, what kind of work you do, whom you interact with, and what your goals are. Ultimately, we want a digital assistant to know you well enough to understand statements that a stranger might find ambiguous – to know what you mean, as opposed to what you say.
In other words, that app on your phone or your PC should really “get” you – and not treat you as a generic “user.”
To be clear, I am not talking about “personalization” if by that we mean simple ways of personalizing the user interface for specific users. That’s been a goal of software developers since at least the late 1990s. No, I’m talking about a deeper level of understanding of context based on ongoing interactions and all sorts of other clues like location and time of day and the deadline for a sales quota to be met.
I don’t need a digital assistant to be smart enough to pass the Turing Test and convince me it’s a real person. In fact, I think you could argue some digital assistants work too hard at trying to simulate a human-like artificial personality. My goal is human-friendly, rather than human like. Helpful. Resourceful. Anticipating what I need before I ask.
The Executive Assistant
My company, Tact.ai, has been pursuing this human-friendly vision of what a digital assistant should be, specifically a digital assistant for sales and sales management. We deliver it on phones, PCs, and edge computing software that integrates with CRMs and with personal devices such as Alexa. Beyond offering a convenient interface to sales and customer data, it aims to learn the goals and habits of the people it works for and anticipate their needs, rather than making them jump through hoops to get the information they want.
In that context, I don’t need C-3PO (who talks too much, anyway). I don’t want or need an omniscient computer intelligence, just software that is smart enough to understand customer contacts, and sales targets, and ongoing dialogs about the steps needed to close a deal.
What I want is more like an executive assistant – a digital helper that books appointments and keeps projects on track, not only following instructions but making intelligent suggestions. An executive assistant has the situational awareness to know when it can fill out a form on your behalf, versus would be easier to hand you a form to fill out yourself, rather than trying to do it for you. If you just booked travel to Los Angeles, then a few minutes later ask about the weather, your assistant knows you’re asking about the weather in Los Angeles, not wherever you are sitting at the moment.
A good executive assistant also knows when to nudge you to take an action you might otherwise forget, something we can simulate with push notifications to your phone or desktop.
Having worked with some excellent executive assistants over the years, I don’t mean to suggest those people are replaceable. On the other hand, not everyone is lucky enough to have a human executive assistant, but everyone has a phone. A digital executive assistant that is always with you has tremendous potential as a productivity accelerator – in our case, an accelerator for making sales and expanding deals.
In Gartner’s recently published Hype Cycle for CRM Sales Technology, 2020 (subscription required) Tact.ai is mentioned as an Example Vendor for three high business impact technologies – Voice-Driven Sales Apps, Sales Bots, and Mobile Sales Productivity. And despite putting these technologies in the category of being hyped and still needing to mature, what Gartner about voice apps for sales is this: "When effectively deployed, such implementations can improve individual productivity by offering a more accessible alternative for documenting sales activities and outcomes, thus freeing up more time for selling."
That’s exactly our goal. Salespeople tend to resent being asked to use CRM systems designed around screen after screen of data entry forms that serve the cause of management reporting, perhaps, but not sales productivity. Forms are convenient for the computer, forcing data into a structured format, but often not convenient for the user. We think software ought to be getting smart enough that it can meet the user – the person – halfway, rather than trying to make people behave like computers.
The Context-Aware Digital Assistant
I am proud of what my team has accomplished with digital assistant software but also far from satisfied.
As I mentioned, using voice as the new user interface is only part of the task of creating human-friendly software. Understanding context is the most important thing, and there is always more to do there. By focusing on the knowledge domain of domain to sales and sales management, we have made our digital assistant smart about those topics. Ultimately, we will want it to understand an even broader context of businesses, markets, and the world in which they exist.
You might expect a good executive assistant to read the news about the market you sell into and alert you if one of your customer contacts just changed jobs. A digital assistant should do the same (and ought to be able to “read” far more widely).
We are only starting to explore the possibilities. Steps my team has taken in the right direction include tailoring the Tact.ai platform to industries such as pharmaceuticals which have distinct sales and marketing and regulatory requirements.
As much as possible, we want your digital executive assistant to learn your business organically, meaning that you train it much as you would onboard a human executive assistant by introducing them around the office and showing them where to find the stationery and the coffee maker. When we began working with the enterprise sales team at a major manufacturer of computing equipment, we trained their sales assistant implementation to check warranty records prior to every sales visit so the team would know to ask for a renewal if the expiration date was coming up soon.
What I said earlier about a good digital assistant understanding what you meant, as opposed to what you said, is a difficult problem computer science calls “intent extraction.” That is something we continually work to improve. When a word can mean two or three different things, we want the software to understand what is intended. One of the ways we improve on that is by keeping track of ongoing dialogs, rather than trying to understand words or sentences in isolation. That helps with examples like the one I used earlier about booking travel to Los Angeles and then asking, “What’s the weather like?” A digital assistant needs to remember the previous conversation (about booking travel) to interpret this question correctly.
To return briefly to the question about whether AI is overhyped or underhyped, I should acknowledge at least briefly the need to avoid the “when all you have is a hammer, everything looks like a nail” syndrome. AI and voice are the perfect answer for many problems, but not all. Simplicity is one of the greatest virtues in software, and software development teams should resist the urge to apply complicated solutions to simple problems. There are still times when giving the user a simple button to click or form to fill out is the simplest and best answer.
That said, I firmly believe one of the greatest challenges in computing is making software serve its human masters rather than expecting the user to kowtow to the needs of the software. That is the best use we can apply AI to – and it’s no hype.