Chatbots still have a lot to learn, but a little shopping they can manage. We show you which solutions are already in the market, and where the technology is likely to develop.Christiane Fröhlich
Butter, salami, ten organic eggs, organic whole milk and cheese, preferably cheddar. Where yesterday we took a grocery list like this to the supermarket, today we can send our preferences to Allyouneedfresh using WhatsApp, and get a link sent back just a few minutes later with a full shopping basket.
The WhatsApp shopping service has been available for the DHL marketplace grocery shop Allyouneed since late July. The response is good: "Today we can already say that the positive acceptance of the service, and the speed at which it is spreading, make it likely that it will continue after its three-month trial phase. Just how remains to be seen," reports Allyouneedfresh CEO Jens Drubel.
Here’s how it works: The customer stores an Allyouneedfresh number in their smartphone and sends their shopping list or a photo of a recipe to the online supermarket using WhatsApp. There, a bot – a piece of software for automating tasks – analyzes the list and identifies the desired products.
The shopping basket is then filled in the shop using a self-developed recommendation engine. The customer receives the link to this basket, which brings them to a landing page at the shop using responsive design. There they can modify the basket, pay, select delivery options and complete the order.
But don't go too fast: things aren't quite that simple – yet. Technology is not ready to fully replace people. "The system makes a preselection. Employees then check the selection and make changes if needed. That's why this is still a pilot. The system still has plenty to learn, with more than 22,000 available products," explains Drubel.
He gives an example: if a customer has several organic products on their list, it makes sense for the system to also make a recommendation for an organic cleaning product. Recipe photos also require plenty of post-checking.
Chatshopper, launched by Antonia Ermacora and Matthias Nannt in Berlin, Germany in 2015, also relies on chatbots. Users can send fashion requests to Chatshopper via Facebook Messenger. Their shop partner is Zalando. Text and voice recognition runs on the cloud-based solution from Wit.ai, a startup acquired by Facebook in 2015.
Chatshopper developed its bot architecture itself. The bot uses "intends" such as shirt, sweater, pants, color or size to recognize what the user wants. If the bot is 100 percent certain it has correctly identified the customer wish, the request is sent on to Zalando. When it is not sure, it is checked by an employee.
The bot architecture is modular, so that it can dock at any shop or messenger system with a public interface (API). "That way we can dock immediately to Instagram, Snapchat or WhatsApp, if they open via an API like Facebook," explains Ermacora.
There are two challenges for providers like Chatshopper or Allyouneedfresh to overcome: First, only a few messengers, shops and service providers currently use public APIs. WhatsApp is rumored to be opening up to such business applications at the earliest at the end of the year.
The system must also run automatically, because processing requests by people requires too much work and is too costly. "It was quickly clear to us that we can't constantly hire more people. People don't scale up," says Ermacora. So the bot has to continuously learn, and constantly improve its ability to interpret customer wishes. "The first step is for the bot to understand what the customer even wants. After that it can learn to optimize and personalize suggestions. We're still in phase one, the training phase," stresses Ermacora. The goal is for the bot to also manage a minimum of small talk.
The degree of automation is what determines success, as shown by the example of the "James, bitte" ("James, please") concierge service. It offered to fulfill all kinds of requests via Messenger – from buying aspirin to organizing a candlelight dinner. "Developing an automation system took lots of time and money. In the end we lacked an investor who would help shoulder that cost," sums up Mateusz Warcholinski, the founder of "James, bitte." After a five-month pilot the service was terminated, and Warcholinski now shares his expertise with other businesses via the Brainhub agency.
Nonetheless, chatbots have a large number of possible uses. You can search for and store job offers at Jobmehappy using a Facebook bot. Berlin-based startup Insurgram is working on a bot with Ergo Direkt that provides information on insurance, and should even make it possible to acquire a policy.
Completely new applications are also being developed in other business areas. Sage, a provider of cloud-based accounting solutions, recently introduced its messenger bot "Pegg," which registers and executes financial transactions sent via Facebook Messenger. "Users should have nothing more to do with the work of accounting. Business transactions can simply be registered and sent via text message. We've spent far too much time having to learn to talk like a computer. Now it's time for computers to learn to talk like us," says Kriti Sharma, Global Director for Mobile Product Management at Sage.
Many of these ideas are still dreams in formation. But with progress in artificial intelligence and machine learning, the digital service landscape will change. Ralf Ohlhausen, Business Development Director with payment technology provider PPro, thinks that bots for various tasks might be offered in bot stores in the future, like apps are in app stores today. He also foresees in-bot payment and – like with app stores – payment using information stored in the bot store – and thus also via a messenger provider such as Facebook. And that is how the butter, salami and organic eggs would be paid for.