Keatext is using AI to track consumer feedback

Until a few years ago, if you had a problem with a company, you would probably have given them a call and spoken (or at least tried to speak) with a human being.

Now, around 65 per cent of customer feedback that companies receive is text — things like emails, Facebook posts and online surveys, said Narjès Boufaden.

Boufaden is the founder and CEO of Montreal-based Keatext, which uses artificial intelligence to help companies navigate this feedback, giving them a sense of customer sentiments and common issues customers are experiencing.

But computers aren’t built to understand language; “computers are really made to work on numbers,” Boufaden said.

While it’s easy to use a computer to find averages or plot a chart to see trends, finding a way to “change the words, the meaning and the intention into numbers so you can actually process those numbers and make decisions and see trends” is a challenge, she said.

Adding to that challenge is the fact that different people will use different words to express the same sentiments and some words can have different meanings in different contexts.

That means Keatext’s software has to do more than understand the meaning of words. It has to understand the context in which those words are being used.

It’s a problem that software developers have been trying to solve for decades.

Older systems looked for keywords, but that sort of association isn’t very accurate, Boufaden said. For example, a negative word in an online review of a hotel might describe the weather during the reviewer’s stay, not the room. But a computer looking for keywords isn’t able to understand that.

More modern systems are better but, until recently, computers still struggled to understand conditional statements.

Keatext’s system uses multiple “layers” of processing that build on each other to “try to infer new knowledge, new information (and) new relationships,” Boufaden said.

“This is really the way to capture, layer by layer, the different linguistic phenomena that you have in the sentence to, at the end of the day, get the right sense,” she said.

Keatext’s product isn’t just being used to gauge customer feedback. Some of its customers, like NASA, are using it to measure employee sentiment.

Next, Boufaden said, she wants to add the capability to give predictions and recommendations instead of just analysis.