Artificial Intelligence

How JD.com Uses AI To Connect Consumers With Commerce

JD Talks Next-Gen AI Consumer Experiences

There is nothing worse than trying to get customer service from a less than fully functional chatbot or automated voice system.

Ask it a question it gives an answer, just perhaps not a relevant one to the question asked.

Re-phrase and re-ask, and the consumer might do better — but it’s likely they get the same bad answer, or the bot digs deep and finds a way to give a worse one.

It’s why all the pomp and circumstance chatbots entered the market a few years ago as the next great idea in contextual commerce enablement. In the U.S., the tech has more plateaued than exploded. The experience still isn’t that good, and consumers are likely to be driven to try to talk to a live person instead of trying to fight it out with glitchy artificial intelligence (AI).

The situation in China, Dr. Xiaodong He, deputy managing director of JD AI Research and head of the Deep learning, NLP and Speech Lab at JD.com, told Karen Webster, is quite a bit different. Unlike the U.S. where the modern era of physical and digital retail evolved separately and have converged in the last decade — the modern era of retail in China is synonymous with the digital era of retail. Omnicommerce, social commerce, chat-based interfaces for transactions, and digital authentication are all part of the baked-in experience.

“Basically, when the customer presses order and the item is delivered, they’re part of a wholly-formed process on the front- and back-end,” he said. “If they have any questions, they’re easily connected to the customer service, they are authenticated, they are updated. I think that in China all these services are more integrated because I think, it’s just that the underlying systems are more connected.”

And from a baseline of better connectedness, Dr. He told Webster, JD.com has been able to build AI that doesn’t just understand customers’ questions well enough to give them right answers. It understands them well enough to respond to their feelings — and curate the content they see.

AI, This Time With Empathy

He said it’s not enough to set the bar simply at an AI interaction going well (i.e. the AI understands the question it is being asked, and it answers informatively and correctly). And that’s true particularly if one wants to use it in a customer service function. In most instances, when someone is calling customer service in any country, he noted, they have a problem.

In the U.S., sitting on hold waiting for a representative is realistic. In China, where JD receives millions of customer service contacts a day, it is less so, and consumers are more open-minded about embracing an automated variation for time-saving purposes.

But, he said, only if it is meeting and exceeding all their needs in the experience — and that includes their emotional needs.

“With our idea of empathic chatbots, we want it to be that requests made in text or by voice, we definitely provide the answer, but we also try to give some empathetic response,” He said.

Consumers, he noted, are often emotional when the call customer service. They’re waiting on a late package. They’e asking the same question over and over in increasing volume. These are things that would signal a human operator that the consumer they are talking to needs reassurance as much as they need information. A machine can’t do that on its own. But, He noted, their multimodal chat interface can be trained to.

“So, with the late package, we can check the inventory or even check the logistics and then some more information, and then also have the bot note, ‘Everything’s fine. No, it’s not lost; it’s just on the way.’”

It’s a small connection, he noted, but the difference between a good interactive chat experience and a frustrating one — and also a baseline of what they can use the voice and text chat interfaces they are building to do. In a customer service function, he noted, they can program to avoid the things that make consumers feel upset enough to call in.

In a sales function, he noted, the programming can also zero in on what delights the consumer.

Curating Consumer Desire

For anyone who has ever started out a digital journey looking for something general but relatively simple, the problem of discovery has likely appeared. Lost in the infinite digital shelves, the customer knows what they want, but they can’t quite figure out how to make their merchant show it to them.

JD, he noted, when it is building those AI-enhanced efforts is about giving customers a better ability to make those connections. Visual search, for example, is an arena where a single snapped image can be fed by a customer into the system to have it run through the whole system against the entire inventory catalog at JD to find a match. Those matches are then sent back to the customer who can select their choice.

That’s a successful system, he noted, but it requires specifics of customers. Sometimes, he noted, the more important service is the general discovery guide that understands what one has wanted in the past well enough to guide future decisions.

“We started a project called Alpha Sale, which provides a digital agent sales agent that acts like a personalized assistant to help every single customer, and to help every single product,” He said.

Good help is hard to find while shopping, he noted, in both the physical and digital world — and browsing inventory ad infinitum is an experience killer most. The agent JD is now building, he said, has two missions — both tricky. The first is to understand any product, how it works, and who it works for. The second is it needs to be able to understand any and every customer.

Then, it has to successfully matchmake between them, leveraging its massive duel pools of knowledge. Oh, and the goal is to do it for its own store — JD.com — but also for the merchants it has selling on its platform.

It’s a big job and not an easy project, He told Webster, but among the benefits of having a fully-connected, fully-integrated eCommerce ecosystem that flows between online and offline transitions as a baked-in component is that it isn’t an impossible goal either. JD has the data and the experience with building chatbots that aren’t merely there to know customers, but to understand them a level or two deeper. And as powerful as the AI tech that will be used to bring this all together for consumers is, he noted, the real power here is the interconnected ecosystem running underneath it.

“It’s not just one single technology,” he said. “AI itself is probably not the magic tool to make a big change in the experience that it can be made out to be. There’s some basic infrastructure that has to lie underneath it, and some pretty advanced connections actually, which is actually very important.”

——————————

NEW PYMNTS DATA: HOW WE SHOP – SEPTEMBER 2020 

The How We Shop Report, a PYMNTS collaboration with PayPal, aims to understand how consumers of all ages and incomes are shifting to shopping and paying online in the midst of the COVID-19 pandemic. Our research builds on a series of studies conducted since March, surveying more than 16,000 consumers on how their shopping habits and payments preferences are changing as the crisis continues. This report focuses on our latest survey of 2,163 respondents and examines how their increased appetite for online commerce and digital touchless methods, such as QR codes, contactless cards and digital wallets, is poised to shape the post-pandemic economy.

TRENDING RIGHT NOW