When building chatbots, think about the entry points

What do you think of when you’re hungry? Bots want to know. Image credit: Pexels.

“I’m hungry. Shall we do Japanese or Italian today?” You message your teammate.

Within a few seconds, you receive a reply. “Would you like to place your order with these restaurants?”

Except that the reply isn’t from your friend. It’s from a bot.

At Tech in Asia Singapore 2017, we gained surprising insights into artificial intelligence (AI) bots and assistants.

From left: Tak Lo, managing director of AI accelerator Zeroth.ai, Jeremy Seow, senior product manager of Zendesk, and Virginia Yang, head of product of Mimetic.ai. Image credit: Tech in Asia.

Think of how customers will start connecting with your bot

“AI is the next electricity,” says Tak Lo, managing director of Zeroth.ai. Zeroth.ai is an accelerator program for early-stage AI startup.

If AI is electricity, then natural language processing (NLP) would be its juice.

It’s the fundamental building block of bots, which could range from customer service chat bots to Evie, the AI scheduling assistant by Mimetic.ai.

To get bots right, Jeremy Seow, senior product manager of customer service software Zendesk, says that on top of studying NLP, bot builders need to think of entry points – the moment that a customer gets hooked onto your product.

“In my experience, the more successful bot builders are those that really thought through the ways people connect to the bot,” observes Seow.

Facebook introduced an AI bot with Delivery.com where you can order food in a chat with a friend. Using this example, Seow points out that when we speak with friends, we don’t usually say, “I want to order food.” Instead, we say phrases like “I am hungry” or “I want to pick up pizza”.

Seow gives another example – ecommerce. An online shopping platform may want a bot to answer “shipping returns” and build NLP around that specific phrase.

But that’s not what customers say. Instead, they might say, “Something’s not quite right with the product I received.”

Expanding the bot’s recognition to different commonly-used phrases also increases customer entry points.

The hard part isn’t coming up with the algorithm, it’s gathering data

One difficulty with bringing bots to the next level is that businesses don’t share conversations with each other. Data gathering is difficult.

“I don’t think businesses want to share conversations. You have to build conversation models based on only your own customers’ data,” says Seow.

Moreover, even if there is a large body of data available through conversation apps such as Facebook and WeChat, Yang thinks it’s often not possible to tell what a conversation really means.

Instead, Evie uses emails to get really good at assisting people. “Evie gets better the more meetings she schedules,” says Yang. “We’re focused on people, time, and location.”

The email entry point helps Evie collect the right data to become a better assistant for scheduling and managing appointments. This would possibly allow Evie’s creators to bring her onto Slack to automate appointments eventually.

Bots won’t take over the human race – yet

While bots aren’t powerful enough to be our overlords yet, they’re still immensely helpful. Seow says the robot vacuum Roomba automates something repetitive we don’t want to do, like cleaning our rooms. “AI won’t replace this panel because people still want to do this,” says Seow.

“Bots are still here to assist. The best combination is when you mix AI with human creativity,” says Yang.

This is part of the coverage of TIA Singapore 2017, our conference taking place on May 17 and 18.

This post When building chatbots, think about the entry points appeared first on Tech in Asia.

from Tech in Asia https://www.techinasia.com/build-bot-tips
via IFTTT