Concierge Startups Tap Cognitive Technology

Even concierge apps need a little help themselves to get their jobs completed.

Some are turning to artificial intelligence, a sensible solution given that significant VC funding is moving away from social media and towards businesses that utilize cognitive technology. Magic+ and Operator, personal assistant and personal shopper messenger services, use IBM Watson as an integral part of their stack.

The Rise of Text-Message Services

In 2015 and 2016, a wave of startups bet that people would rather send a text than open an app. Instead of tapping through menus, you’d message a request in ordinary words, “book me a table for four on Friday” or “find me running shoes under a certain price,” and the service would handle it.

The big platforms made the same bet. In April 2016, Facebook opened its Messenger Platform to bots, saying that more than 900 million people used Messenger each month to talk with friends, families and over 50 million businesses. It also offered developers Wit.ai’s Bot Engine to build bots that could “interpret intent from natural language.” Concierge services like Magic+ and Operator sat at the high end of that trend, promising to handle requests no simple bot could.

Magic+, which received $12.5M of Series A funding in 2015 from Sequoia Capital, is an on-demand, scalable high-end personal concierge that claims to be capable of handling any task, regardless of size or scope. Users can send text messages that will help users with tasks such as arranging mutual meeting times and finding personal trainers. You can even order helicopter transport.

Requests are fulfilled through a combination of humans working as executive assistants paired with intelligent software. According to Magic, their “unique combination of humans and software is what enables us to consistently deliver an unparalleled level of service.”

Operator, set up by Uber co-founder Garrett Camp, specializes in personal shopping. You can request items by text message in very specific detail from around the world. Among other things, it keeps records of your preferences. Operator was most recently funded by three investors in a $10 million Series A round.

Humans in the Loop

The phrase “humans working as executive assistants paired with intelligent software” is the key to how these services worked. The software did what it was good at: understanding a request, pulling up a customer’s preferences, suggesting options and handling routine steps. People handled judgement calls, unusual requests and anything that needed a phone call.

That model is often called “human in the loop.” It lets a company offer a service that feels smarter than its software really is, while collecting examples that can be used to train the software over time. The open question was always whether the software would improve fast enough for the business to stop depending so heavily on people, because people are expensive and don’t scale like code does.

The same machine learning and natural language processing technology that powers these startups via IBM Watson is available by signing up to IBM’s Global Entrepreneurship Program.

Through the Global Entrepreneurship Program, up to $120K of credits are available to spend on access to IBM services that these startups use, as well as go-to-market support, technical expertise, and business mentorship.

Here’s how to apply.

What Happened Next

Facebook’s own experiment shows how hard that balance was. Its assistant, M, ran inside Messenger and also combined people with AI. In January 2018, Facebook said it was shutting down the human-assisted M after two and a half years. The Verge reported that it had only ever become available to about 2,000 people in California. “We launched this project to learn what people needed and expected of an assistant, and we learned a lot,” Facebook said.

If one of the world’s largest tech companies couldn’t make a human-powered assistant work at scale, it’s no surprise that independent concierge startups found the economics tough.

The Idea Returns as AI Agents

The dream behind Magic+ and Operator never went away. It came back in a new form with large language models, which can follow written instructions far better than the chatbots of 2016. Today’s “AI agents” promise much of what these concierge services offered: tell the software what you want in plain words and let it book, buy and schedule on your behalf.

The lessons from 2016 still apply. A service that leans on people behind the scenes can deliver a great experience, but it has to be honest about its costs and about how much the software can really do alone.

What It Means for Founders

For founders building on AI today, the concierge startups are a useful case study. Starting with humans in the loop is a smart way to learn what customers actually ask for. The risk is building a business that only works while people quietly do the hard parts. The founders who come out ahead track how much of the work the software truly handles, and plan for the day it needs to handle more.