Why Bots Aren’t Ready To Replace Mobile Apps

But the hype about the bot revolution is five years too early. It’s being generated by the companies that hope to dominate the space, and it’s designed to get developers on board.

Tragically, however, the breathless promises about bots are just raising expectations among users, who will be disappointed, get turned off by their limited capabilities and may never come back.

The problem is that even the bots that mostly work, don’t work as advertised. Bots like x.ai’s Amy, GoButlerand many others are partly powered by human beings pretending to be software. Behind the curtains, those people answering queries are also serving to train the bot artificial intelligence system so that eventually an increasing amount of the work can be done automatically. The hype and promise is that AI is controlling the bots. But the AI technology isn’t ready.

Today, bots are just the chat equivalents of interactive voice response (IVR) systems—you know, when you call the bank and a recorded voice says press 1 to open an account, press 2 to find out your balance and press 3 to do some other thing you don’t want to do. People hate IVR. And they’ll hate bots even more.

Like IVR, bots work only in very limited circumstances. If you’ve got a shopping bot and there is only a short list of items to choose from, then it works great. But bots are useless for a store full of goods.

Facebook’s Spring shopping bot is a perfect example of how bots fail. To use the Spring bot, you identify the contact ShopSpring and then send the phrase you’ve memorized, which is “go shopping.” The bot grills you about what you’re looking for, then it will give you five suggestions.

Nobody wants to buy clothes or shoes this way. Imagine walking into a shoe store and having the shoe salesperson suggest five pairs and you’re supposed to pick one pair and buy them. The last time I shopped for shoes, I went to five stores and looked at hundreds of shoes, the overwhelming majority of which were not what I wanted.

The 1-800-FLOWERS bot makes more sense because you likely to want less choice. You probably want to order a dozen roses. And if you don’t want that, you’ll want one of a small number of flower arrangements. There’s far less choice, so it should work OK in isolation.

The problem is not only the AI, but also the data. Every conceivable option must be keyed in, and the bot systems must understand the data. While it has taken the Web decades to optimize itself for search engines, the work of optimizing data for bots has barely begun.

Countless stories have emerged this month revealing epic fails when users tried to use Facebook’s Weather Cat bot or the Spring shopping bot. Users are reporting that Facebook’s bots take minutes to respond (even telephone IVR takes just seconds).

Current bots work only if the user sends the exact input expected. But if users deviate from the expected script, as they will, the bot simply can’t respond in any useful way.

Bots are a return to the command-line user interface. You have to know exactly the right words to enter to obtain the result you want.

If apps are like the graphical Mac OS or Windows, bots are like command-line DOS.

Bots won’t work as a consumer-friendly alternative to apps until the gate-keeper application can choose the appropriate vertical bot based on the user query.