Much has been written about chatbots, and the potential is clear. For many actions, text is by far the most convenient and natural interface — in fact, messaging apps have, for the first time ever, surpassed even social networks in popularity. So it’s natural for chatbots to be the next logical step in tech innovation — and they promise to be a boon for not only specific groups like the elderly or those with disabilities, but also the population at large.
We’ve seen nifty chatbots that can keep you updated on election news (Purple), help schedule your dry-cleaning pickup (Rinse), or help get you an Uber while you’re chatting up your date in Facebook Messenger.
We’ve also started seeing early business applications for chatbots. Running a small e-comm platform? Kit (acquired by Shopify) can help handle daily marketing tasks and report sales figures back to you. Don’t like pulling data out of Salesforce? Troops can spit them out in response to Slack commands. Schedule’s a mess? Clara and x.ai can help (our Clara’s name is Pepper Black, much to the befuddlement of many).
Complexity and common sense limit chatbots for now
But looking at this last category of text-interface scheduling services, what becomes apparent is the current limitation to chatbots’ utility in more complex business use cases. Both Clara and x.ai have a human component overseeing the “A.I.” outputs, before they reach the user.
That’s because there is a human at both ends of a scheduling transaction, and human input is endlessly variable and unpredictable. And while I’m the one using Clara to schedule — so, of course, I know the right commands to get to my desired outcome — the person I’m scheduling with shouldn’t have to; that’s the whole point. I’ve had a great experience with Clara, and I’m sure the company’s data model is getting more and more robust every day, but the human fail-safe is testament to the fact that we are still a ways away from a point when the computer can reliably accommodate human quirks and situational variability and translate them to complex input/output based soley on understanding the written word.
And the example above is “just” scheduling. It’s one focused task with a clear outcome, and it’s already infinitely complex. When one starts to consider more complicated business use cases that require making judgment calls and strategic decisions, things get dicey.
Workplace bots need to improve dramatically to be useful
Take sales, for example. Modern sales workflows are famously mired with necessary-evil tools that mean salespeople spend more time using and managing “tools” than they actually spend selling. Chatbot technology could be a clear and easy way for salespeople to offload all these tedious administrative tasks so that they can get to the part that can’t be automated or replicated — person-to-person relationship building.
I’ve heard hundreds of salespeople ask for a sort of a Pepper Potts – a smart executive assistant that always knows better. Just got out of a meeting? Your assistant gives you a call, asks how it went, and takes care of all the CRM logging. Got a few spare moments in a cab? Text your assistant to ask what you’re forgetting about today. About to walk into a meeting? Get a cheat sheet texted to you with some good ice breakers tailored to the person you’re about to meet (though discussing the latest “Game of Thrones” episode is almost guaranteed to work universally.) You get the gist.
What’s it gonna take to make smart chatbots?
As it turns out, that’s really hard to pull off. For a computer to be able to intelligently assist along the entire sales process, it needs to understand a LOT of very fuzzy data. Think about all little idiosyncrasies of every sales use case: where things stand in the sales cycle, quota, type of customer, what has worked in similar cases, what marketing has primed the customer with, and a thousand other variables. Juggling all of these to come up with sound advice (and automate complex sales actions) requires some really high order thinking, and presents a tough problem.
The key lies in the data model. It needs to have seen enough different patterns play out that it can learn from small nuances. It also needs continuous feedback from the user, to make sure that it’s in fact getting smarter in the process. Figuring out how to get there is in my mind the most meaningful challenge standing between us and truly intelligent interface-less software. You see, users — especially in the enterprise — don’t like just handing over their data and letting you watch them do their job for months so that you can improve your engine. Providing enough value out of the box — and making it worth the user’s while to “teach” the system to get better — is key.
As it has many times before, the enterprise world may need to borrow a page from the consumer playbook. One interesting analogy that my team has drawn on while working on an intelligent sales companion comes from dating apps. Hinge claims to learn about your tastes and preferences and better tailor your dating suggestions the more you swipe your way through the app. I don’t know how sophisticated their machine learning algorithm is — but I think they’re onto something. Most people don’t necessarily love telling some company about who their type is. But Hinge has figured out a way to extract this data really painlessly (hey, you just have to swipe left or right) — and make it worth our while (hey, you just might get laid). Whoever figures out how to do this for a workplace use case will be in a great position to develop the next generation of enterprise software for that vertical.
Could minimalist interfaces with immediate feedback like those used by dating apps be the key to building out complex chatbot data engines?
In spite of numerous roadblocks, this will be a meaningful challenge to tackle. Interface-less technology is more than just a fad — in the 2016 Internet Industry Trends Report, Mary Meeker predicts “the rise of voice interfaces because they’re fast, easy, personalized, hands-free, and cheap, with Google on Android now seeing 20% of searches from voice, and Amazon Echo sales growing as iPhone sales slow.” We want to interact with technology the way we interact with each other; and that means fewer pixels, and more intelligence.
Branko Cerny is a software entrepreneur living in San Francisco, CA. Originally from Prague, Branko graduated Dartmouth College with a degree in Psychology. He left Google’s marketing division in 2013 to start Immediately with the aim of making productivity software more human. Immediately’s latest product – Gong – is an intelligent pocket assistant helping salespeople find creative ways to build stronger relationships and close more deals. He also founded the 100 Ways to GetUnstuck movement, providing micro-mentorship at scale – from top business thought-leaders to the new generation of salespeople.
VentureBeatVentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:
- up-to-date information on the subjects of interest to you
- our newsletters
- gated thought-leader content and discounted access to our prized events, such as Transform
- networking features, and more