How to tell if your bot is smart enough

I asked a weather bot that launched a few weeks ago a few basic questions:

Me: Will it rain in NYC next week?
Bot: It will be cloudy later today in New York, NY.
Me: hmmmm how about next week?
Bot: hmmmmmm

Thanks to high-profile launches by Facebook and Microsoft, public attention has been focused on dumb message bots that answer basic requests such as “What’s the weather?”

But if messaging is the new platform and conversation the new UI, we are going to need smarter bots. Not bots that only give you information, but bots that know about you and help you do important things. We need bots so smart, they have a deep understanding of their areas of expertise.

The field of bots right now is like the early App Store – there are some interesting entries, but media and Internet buzz mostly focuses on big brands offering simple-minded entities that serve up their familiar content or cutesy gimmicks that make bots seem like novelties. You’re excused if you wonder why you need another, conversational, way to check the weather.

To understand how much a bot can do, you don’t need to know about every bot in the world, but you do need to understand how broad the spectrum is in terms of smarts. At one end are simpler chatbots – the type you can find on Kik or Facebook Messenger. They can kind of hold a conversation in the manner of that person at a party who isn’t listening too closely. They’re fun and somewhat useful but don’t really interact or do anything impressive.

At the other end are smart bots. You might trust the weather bot with tomorrow’s weather, but would you trust it to handle your money or your health? Let’s use the example of that emotionally charged and all-important topic – money – and what it takes for a bot to be smart enough.

• Context – The smart bot doesn’t only react to the latest thing you said. It’s aware of the entire conversation you’re having with it, and maybe even past conversations. It understands the context of what you’re asking and, more importantly, understands what you’re trying to do. If you first ask a smart bot, “How much interest did I make on my savings account in 2015?” it would not only know the answer but would also be able to answer followups such as, “How about my checking?” or “How about in 2016?”
• Personalization – The smart bot knows your preferences and habits. That means different people get different responses based on the information and data the bot has permission to know about them. A smart banking bot (it’s my specialty, as you may have gathered) is a good example. With permission, it can give responses to questions such as “How much have I spent on travel in March?” or “Do I have enough money in my checking account to cover my credit card charges?”
• Domain expertise – The smart bot knows more than you do about money (in this case). Siri, a helpful personal assistant, can help you find a nearby ATM, but you can’t ask Siri if you have enough money in your account to cover your credit card minimum payment. Siri will also not suggest moving money from another savings account to avoid an overdraft fee. Siri doesn’t know about your bank accounts or credit card minimums, nor does she understand overdraft fees. This would require a smart bot fluent in banking.

In a nutshell, smart bots can converse intelligently – they are bots with AI and purpose. Without high IQ, bots have very limited utility. And worse, dumb bots risk tainting the waters with disappointment and putting off consumers.

Facebook believes bots will be bigger than mobile apps. First, because messaging has lapped social media as the way we (especially young people) communicate, and companies want to go where their customers are. Second, the AI platforms necessary to build more complex bots are becoming much more accessible to companies and developers. Like the early days of mobile apps, as the platforms matured, so did the apps in their prowess.

To nurture consumers and bring them along on this new digital journey, we need a blossoming of smart, specialized bots that perform many roles from simple chatting to complex assistance in specific fields. These smart bots will use data to offer personalized service and deep domain knowledge that fulfill requests, solve problems, and predict customers’ needs. Someday, smart bots will even work together, perhaps connecting to help you plan all aspects of a trip – flights, reservations, your calendar, your banking, the errands you need to complete – as if you have not just a personal assistant, but a team of pros to take care of your needs.

Microsoft Bot Framework and Facebook Bots for Messenger offer companies seeking a mass audience big platforms to build on. But many early bots seem to have skipped school in their rush to get to work, fumbling basic requests and lacking context, personalization, or deep domain knowledge. To win over customers, bots will need to be much smarter.

Dror Oren is VP of product and cofounder at Kasisto, a fintech company that uses AI behind its virtual assistant banking services and a spinout of SRI International.