insights! #84: Between hype and reality: what chatbots can really deliver in e-commerce

Sonja Fuhrmann:

Are chatbots the cool assistants of the digital age, or perhaps just programmed small talk, Joubin?

Joubin Rahimi:

Right now it's certainly still a lot of programmed small talk, or small information providers, but it's going to be a digital assistant, or rather a manifestation of your digital assistant.

Sonja Fuhrmann:

Where are the limits at the moment?

Joubin Rahimi:

When it really comes down to a lot of information and clear recommendations. That's roughly where the line has to be drawn. And the more processes come into play, the more expensive it gets, so the limit there is money and integration into existing systems.

Sonja Fuhrmann:

How are companies already using conversational commerce today?

Joubin Rahimi:

We see that really nicely with cars and car manufacturers. There are lots and lots of these chatbots to answer questions around configuration. Funnily enough, they all always look a bit the same too. But de facto that's an area where it's used well. We also have things in the fashion sector, clothing and so on. But there it's harder, it's still trickier. People trust a chatbot less to answer style questions or other questions, although the aim is to use it to reduce size- and returns-related issues.

Sonja Fuhrmann:

Does that work?

Joubin Rahimi:

Depends on the data. So if you have good data, then yes, then you really can reduce returns, and the question is: how do I collect it? A quick aside: some measure shoes and even feet. That's the most elaborate approach, and others simply take the returns data - which shoes were delivered - plus the answers that go with it, to say this one's too wide, too loose, too narrow, too big for that size, and pass that recommendation on. That's less precise than the other approach, but I think the imprecise one is entirely sufficient in terms of price-performance, because it's a twentieth of the cost.

Sonja Fuhrmann:

I don't know if you've worked with chatbots at home before. Probably yes, I always found it, well, hit and miss. Would you say chatbots have actually improved customer service?

Joubin Rahimi:

I think the question is when … Chatbots do improve it, but as with any personal interaction, if the interaction is simply bad, then it doesn't help. There's also - a completely different example - chatbots in situations that aren't meant for the customer at all, but for the defaulting payer. Imagine you have an invoice that, for whatever reason, hasn't been paid, a reminder arrived, emails arrived, and then you get an automated call with a nice smooth computer voice, precisely with the information: this is outstanding. In ten days we'll initiate the official reminder procedure. That's obviously pretty harsh, but you also know you'd need to pay. And then a chatbot, in a real case, can also say: "Get in touch with us. It seems you're having payment difficulties. We'll find a solution." That's a bit away from the typical discussion, but that's another way these chatbots can be used, and it actually helps both sides.

Sonja Fuhrmann:

Yes, that's one type of benefit. What other concrete benefits do companies get from using chatbots?

Joubin Rahimi:

The main challenge is that in this world we have to make over 300, 400 decisions a day, quick decisions. And when you're about to buy something or decide on something, you often need a bit of support, a little nudge or a piece of background information. And that's where a chatbot really helps. One example where I'd have wished for one is a manufacturer of window seals. They had seven different variants, but which one should I choose? And because of that, I didn't buy that seal for a whole year, a whole season, because I couldn't get that question answered. The information was somewhere on the website, but I could have used a chatbot and said: Here, I've got windows like this. What would you recommend? And if it had maybe asked me a follow-up question, I'd have gone ahead. I wouldn't have called, because I thought: Ugh, do I really want to phone a call centre now.

Sonja Fuhrmann:

Right. But what about the notorious gut decisions? I don't know how you two decide. In the end, I decide intuitively. Don't chatbots kill that?

Joubin Rahimi:

They don't make a decision for you, or for either of us, in fact for none of us — they give you information, just like a good salesperson would too. I don't know what question's coming next. Is it manipulative?

Sonja Fuhrmann:

Where's the line to manipulation? Yes, exactly. Great minds think alike.

Joubin Rahimi:

If you set them up right, you can of course use all these questioning techniques to nudge someone towards a purchase. Even quite heavily. But of course, if the customer ends up unhappy, you get returns again, and no long-term customer satisfaction. It's tricky. Chatbots aren't the problem when it comes to manipulating people either. A really good salesperson — I'm not wearing glasses today — but I've bought four pairs of glasses. Yes, I'm susceptible to manipulation on that front. There was wine, there was beer. He was brilliant, and alongside the two pairs of prescription glasses he then sold me two pairs of sunglasses as well, one for my wife and one for me. And I'd only meant to leave with one pair of glasses. He did everything right, that salesperson. He completely manipulated me there, but in a likeable way. And I think that's the key. I'll go back there for my next pair of glasses.

Sonja Fuhrmann:

Right. How do you see conversational commerce developing over the coming years? What's going to happen there?

Joubin Rahimi:

I think we're still at the beginning. Five years ago we got to build something for a large clothing manufacturer, and let me briefly describe what we built, how long it took, and how long it would take now to get to that same point. We had six cases, as they were called. So there were six, seven million emails coming in, and six cases were handled. Returns, something broken, my checkout didn't work, so the price was calculated wrong. I've got a merchandise voucher. The value's not right. There were six cases like that, and first we had to understand these cases. We had 40 students sitting on it for a week, classifying emails, and then saying: "Okay, these are the cases," and to reach a high probability. We achieved a success rate of just under 82%, so a bit above, which was already great, because that 82% didn't need to be handled manually — including integration into the backend systems, so that a return slip gets created and so on, checking that it really is the customer writing in, that kind of thing. Just under 350 days for the whole thing, not counting the students. So horrendous, horrendous. That was a lot. I'd say integration now is significantly faster, maybe 100 days, but we also don't have 50 students any more, more like 10 person-days, and you'd already have that language model built. That's from back then to now, and it's not linear, it's a lot faster.

Sonja Fuhrmann:

Exponential.

Joubin Rahimi:

Exponential. And then the question is: Is the chatbot just a chat? Is it voice? Is it an avatar? Is it VR in my glasses? So these are topics that are coming very, very quickly.

Sonja Fuhrmann:

Right, so that's the next question: Which technologies will follow next in this space? You mentioned VR.

Joubin Rahimi:

And that'll be one of the next ones. So from — I just have text, which is the simplest form for a system — to understanding voice even better. We all know Alexa, Siri. Exactly. Where Alexa currently gets noticeably better results, in my view, but I think some others feel the same way, than Siri does, but that's going to move forward. And then I have voice, and then comes image, and then moving image.

Sonja Fuhrmann:

Right, so there's still room for improvement.

Joubin Rahimi:

There's room for improvement there, yes.

Sonja Fuhrmann:

But it could move faster than we think.

Joubin Rahimi:

That'll move faster than we think too, yes.

Sonja Fuhrmann:

Thank you very much, Joubin, for the conversation, and thank you all for your attention.

Joubin Rahimi:

If you still have questions, just leave a comment below or share your thoughts, and maybe also tell us where you think conversational commerce, or this whole chatbot thing, is heading.