To Truly Work, AI Needs to Get Really Personal

‘I just can’t stop it/ Every Saturday you see me window shopping/ Find no interest in the racks and shelves/Just ten thousand reflections of my own sweet self, self, self...’ —The (English) Beat (‘Mirror In The Bathroom’)

Although gender stereotypes are usually galling, I am often told that I am an unusual man in that I like going clothes shopping. Whether looking for work shirts for me (absurdly loud patterns and contrasting cuff interiors, please) or skirts for my wife, I revel in the possibilities and make the shopping experience into playtime. I’ve had aunts, friends of friends, and even wives of colleagues ask me to accompany them on their retail expeditions because I don’t get antsy, bored, or cranky.

Since I spend a goodly amount of time in clothing shops, I frequently get to observe the behaviors of retail clerks. And those clerks really do say that a lot of people look “really great” in a lot of clothes. I hear this myself when I try on things, and while I do rock a vest like nobody’s business, I’m fully aware that not all clothes look good on me. The salesclerks use this line in hopes of driving more sales, sure, but at heart they seem to be echoing back what they think prospective clients want to hear. I doubt I am alone in thinking that when it comes to clothes, telling me something doesn’t work on me won’t make me leave the store; instead, I’m more likely to ask the sales staff to help find the clothes that do flatter.

Increasingly, when consumers engage with sales or customer service over the phone or through digital channels, they are interacting with personnel that have some technological augmentation. Not like Roy Batty or Steve Austin, which is a shame because that would be very cool, wouldn’t it? No, these employees benefit from systems like personality matching tools that analyze consumers’ communication styles and then attempt to match them to employees with complementary styles. Or the staff might take cues from a next-best-action engine that predicts the words to use or the actions to take to get us to buy more stuff or to solve our problems. Or, as I’ve written about here before, conversational AI tools may provide suggested conversational snippets to the employees to feed the dialogue they are having with you.

That all sounds great—and it sometimes actually does help make for smoother conversations. But even though these tools are trying to personalize interactions for us, they still end up being too impersonal. Let’s look at the conversational AI example. Those systems may personalize the interaction by including some information about your product or service (“You have 55,000 loyalty points”), but the dialogue is pretty generic; my mom would get the same response, just with point balance substituted in. This feels like the salesclerks saying that I look amazeballs in that muumuu for men.

If we are going to invest heavily in artificial intelligence, maybe we should start trying to apply it to real-time personalization that is well and truly personalized. My communication style may generally be calm and collected, but the day my house is destroyed by a tornado/hurricane/tsunami (hey, climate change is making it possible for weather disaster mashups!), I might not communicate calmly. Don’t let AI just mirror back what it thinks I want to hear; focus on understanding the micro-fluctuations in personality, needs, communication styles, and emotions.

I realize this may sound like one of those “Where is my jetpack?” rants coming from someone feeling shortchanged that our lives don’t look like The Jetsons. Well, yeah, it kind of is. Now build an AI that could have predicted that I would feel this way when I sat down to write this!  

Ian Jacobs is a principal analyst at Forrester Research.