Skip to Content
Artificial intelligence

Slack Hopes Its AI Will Keep You from Hating Slack

The fastest-growing business app is relying on machine-learning tricks to fend off a deluge of messages—as well as competition from Facebook and Microsoft.
January 16, 2018

If you work at one of the 50,000 companies that pay to use Slack for workplace collaboration, you probably spend hours on it, swapping information, bantering, and sharing files with your colleagues. It’s a casual, flexible way to interact—you tap out brief messages in group chat rooms (called channels) instead of sending e-mail, and it feels more like a smartphone app than typical office software.

But while it can be an efficient way to collaborate, keeping up with Slack can become a full-time task, particularly when you return from a few days away and find thousands of status updates, scattered across dozens of channels. Slack estimates that the average user sends 70 messages per day. How can you know which are must-reads and which can be skipped?

Slack’s solution: artificial intelligence. In early 2016, the startup hired Stanford-trained computer scientist Noah Weiss to make the platform smarter and more useful. Over the past year and a half, Weiss’s group has used machine learning to enable faster, more accurate information searches within Slack and identify which unread messages are likely to matter most to each user. Eventually, Weiss aims to make Slack function like your ruthlessly organized, multitasking assistant who knows everything that’s going on and keeps you briefed on only the most salient events.

Noah Weiss is head of Slack’s AI team, the Search, Learning, and Intelligence group.

Slack says its platform, which launched publicly in 2014, is the fastest-growing business application ever, with more than six million daily active users. The company also predicts it will be bigger in the workplace than e-mail by 2025.

But e-mail is not its only competition. Facebook, Google, and Microsoft, with their large existing user bases, have all released office collaboration tools in the last 15 months. Microsoft says that 125,000 organizations use Microsoft Teams, its group-chat platform, which is bundled free with some Office 365 plans. Facebook says that more than 30,000 organizations, including Walmart, use its Workplace by Facebook service. (These numbers aren’t directly comparable to Slack’s 50,000, though, since neither Microsoft nor Facebook would say how many daily users their platforms have, while Slack wouldn’t say how many organizations use the free version of its service.)

These chat products deliver not only steady revenues from monthly and annual service fees, but also troves of data that show how people interact within companies and what types of files and applications they use to get work done. Slack’s larger competitors also see an opportunity to increase usage of their existing software. Companies like Microsoft “will tie these tools in with their other enterprise-wide platforms,” such as Office 365, says Jeffrey Treem, an expert on communication technologies at the University of Texas at Austin. “All of these large technology companies are pursuing this same space because it’s a very rich market.” 

Slack is not worried. “We think we have a bunch of important advantages, among them traction in the market, sharp focus, and a really deep understanding of our users,” says CEO and cofounder Stewart Butterfield.

The work graph

To understand how Slack intends to improve work through AI, I visited the company’s New York office, where the team is based. The space, at the edge of Manhattan’s East Village, is an eclectic mix of Zen-like décor (tall green fronds planted among polished stones) and cartoon kitsch (flat-screen monitors broadcasting emoji animal faces).

The team gives members balloons to celebrate their anniversaries with the company.

Weiss built the 19-person group by recruiting engineers, designers, and product managers from companies like Facebook, Google, and LinkedIn, many from big-data projects. He boasts a similar résumé: one of his first jobs, after studying computer science and economics at Stanford, involved developing display ads at Google. Later, he became a product manager there. After three years, Weiss moved to the startup Foursquare, where he led the product analytics team. It suggested local businesses to users based on the way they explored their neighborhoods.

At Slack, Weiss is applying what he learned at Google and Foursquare to refine search queries and give people recommendations when they open the app. The work incorporates multiple AI methods, including different types of machine learning and natural-language processing.

Some of the technology is already live. One feature shows which people within a company talk about particular topics most often in Slack and where those discussions take place. The information, which appears when users conduct searches in Slack, is meant to pinpoint subject experts so people can direct questions to their most knowledgeable and accessible colleagues. Another feature, added last year, evaluates all of a user’s unread messages, across all Slack channels; highlights up to 10 of the ones its algorithms deem most important; and presents them in a single list.

Slack is using machine-learning algorithms to highlight the most important messages you missed while away from the platform.
SLACK

Both innovations rely on a data structure that Weiss calls the “work graph.” It essentially looks at companies that use Slack and analyzes how the people within them are interrelated, where in the app their discussions are taking place, and what topics are being discussed. If the term sounds familiar, it’s because Google and Facebook have similar structures—the “knowledge graph” and the “social graph,” respectively. But while Google studies public data and Facebook promotes the idea of a single, global network of relationships, Slack thinks of the work graph as specific to each company—a representation of how work is structured within it.

The work graph emerges mainly through a type of machine-learning algorithm called collaborative filtering, which predicts a person’s interests and preferences by collecting information about those of many other people. For example, when people start using Slack, the algorithms will look at the channels they’ve joined, who is active in them, and where else those people are active, in order to suggest several more channels to the new users.

“We’ve spent a lot of time building models that understand what you care about and what content you interact with,” says Jerry Talton, who helps lead the team’s technical work. “In the future, we’ll take that same understanding and apply it to content you don’t know about that could make you better at your job.” 

Keeping an eye on you

Another Slack goal is to help management keep a better eye on its employees. One of the team’s newest initiatives crunches data to construct online dashboards that give executives a bird’s-eye view of how employees are interacting, which topics are trending, and how sentiment changes over time.

“You’d be able to see what your European set of offices are paying attention to versus your U.S. set of offices, or what people who have long tenure at your company are paying attention to versus people who are really new,” Weiss says. Slack is still working out the details—it is unclear, for example, whether companies will be able to access data from the past 24 hours or just the most recent week or weeks—but the AI team plans to roll it out in the near future.

The very idea of “organizational insight” analyses shows how far Slack has come from its early days, when it was regarded as a startup beloved by other startups but out of sync with the demands of large companies. Two years ago, Slack’s biggest customer had just a few thousand employees. Today it has more than a dozen customers with more than 10,000 active users, and a few customers with more than 50,000. These corporations can generate millions of messages a day.

But does mining employee communications invade their privacy? Weiss says his team hopes to assuage concerns by parsing activity only in public Slack channels (rather than the private ones where people can conduct confidential conversations). He also says Slack won’t turn on the feature unless companies request it.

Still, employees may balk, particularly if they think they will get assessed on the basis of how active or popular they are on Slack. Adam Waytz, who researches social psychology and ethics at Northwestern University’s Kellogg School of Management, thinks the feature sounds invasive. “Given the increasing public unease about employers’ control over their employees’ lives and what gets said at work, this product could result in backlash or paranoia,” he says.

Slack also needs to gain trust for its existing AI features. “AI can be tremendously beneficial in matching the right people with the right information to do the right tasks, but it’s not a perfect solution,” says Treem, the University of Texas communications professor. “If you were relying on algorithms to get you the most important messages and you find out a week later that you missed something particularly important, you’re going to lose confidence in Slack’s ability to do what you need it to do.”

To gauge user satisfaction with its new tools, Slack includes thumbs-up, thumbs-down, and “dismiss” buttons with each message its algorithms highlight. It uses this feedback to refine the algorithms.

Early statistics seem promising. Weiss says algorithm tweaks by the AI team last year made searches 50 percent more successful, and also made people 30 percent more likely to accept suggestions about new Slack channels to join. If all goes as planned, the intelligence layer the team is building on top of Slack will morph into a digital assistant that can make people more productive.

Butterfield, the Slack CEO, sees AI as a long game. “I think what we have right now is good,” he says. “In a couple of years, it will be very good. In about five years, it will be excellent. And in 10 years it will be impossible to work without it.” 

Deep Dive

Artificial intelligence

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

Providing the right products at the right time with machine learning

Amid shifting customer needs, CPG enterprises look to machine learning to bolster their data strategy, says global head of MLOps and platforms at Kraft Heinz Company, Jorge Balestra.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.