Business Impact

Slack CEO: How We’ll Use AI to Reduce Information Overload

Stewart Butterfield talks about how machine learning can help your work productivity.

More than six million people use Slack every day as a hub and platform for communicating and sharing information with their coworkers. The San Francisco–based company also just raised $250 million more in funding, in a round that puts its value at a lofty $5.1 billion (see “10 Breakthrough Technologies 2016: Slack”).

But new competitors—among them Microsoft Teams, Atlassian Stride, Google Hangouts Chat, and Workplace by Facebook—are pushing into the workplace-collaboration market, challenging Slack’s dominance and, in some cases, specifically marketing themselves as more productive options. During a trip to Boston, Slack CEO Stewart Butterfield spoke to MIT Technology Review about the ways the company plans to use AI to keep people from feeling overwhelmed with data. Below is an edited excerpt from the interview.

Three and a half years after its public release, Slack continues to grow, but people have also started questioning how efficient it is as a workplace tool.

Well, the complaints that people have about modern office life are usually about not being in the loop, not understanding what decisions are being made, not being aware of what’s going on. That feeling of alienation [actually] gets reduced with Slack. If you work at a 10,000-person company and you’re using e-mail as the primary means of communication, then you probably have access to a couple hundredths of 1 percent of all the communications happening across the company. But if you use Slack you might have access to 10 or 20 percent.

The flip side is there’s a lot more information. We’ve already put in hundreds of little things that collectively reduce the impact of massive flows of information in Slack. For example, when you first start using Slack, we send you a notification for every new message, but shortly after, we prompt you and say, how about you switch to our recommended settings [that only notify you when you receive a direct message or your name or specified keywords are mentioned]. I think we can provide [even more] tools and human-centered design techniques to help people better navigate Slack without feeling overwhelmed.

Right, Slack established an AI and machine-learning division called Search, Learning & Intelligence early last year to reduce that “information avalanche.” Slack has already incorporated a few features that SLI developed. What are your specific goals for that group?

One of them is just [improving] search. Within search, there are two things. Basic keyword search, which is when people know that something exists and they want to find one specific thing. There’s also a more general kind of search, like learning about a topic. Most companies use some jargon or code names, which can be confusing, especially when you first arrive.

The second category [of goals for SLI] is proactive recommendations or alerts. The third is trying to make sense of the whole corpus [of information in Slack] and have that improve over time, ideally in a way that doesn’t require any manual input [from users]. The computers will do it all; people can just communicate the way they would normally communicate. You could imagine an always-on virtual chief of staff who reads every single message in Slack and then synthesizes all that information based on your preferences, which it has learned about over time. And with implicit and explicit feedback from you, it would recommend a small number of things that seem most important at the time. So whether you’re waking up in the morning or getting out of a meeting or getting off an airplane, when you check Slack, there could be this virtual chief of staff waiting for you, ideally with a near-perfect list of those things that are important to you.

That’s an intriguing idea. What else is SLI researching?

Organizational insights. We had this really fascinating internal project that looked at the strength of connections [within Slack] between the different departments in our company and the ratio of public to private messages along those axes. So if you look at Slack from the perspective of our facilities team, [it turns out] they talk to finance, HR, and security but almost never talk to engineering. If you look at Slack from the perspective of our marketing team, they talk to sales, product, and finance.

We want to be able to provide those sorts of insights on an organizational basis and also on an individual basis, so the SLI team is taking some Slack data [with the permission] of customers and showing them how things happen at their own company.

I would—and I think everyone would—like to have a private version of a report that looks at things like: Do you talk to men differently than you talk to women? Do you talk to superiors differently than you talk to subordinates? Do you use different types of language in public vs. private? In what conversations are you more aggressive, and in what conversations are you more kind? If it turns out you tend to be accommodating, kind, and energetic in the mornings, and short-tempered and impatient in the afternoon, then maybe you need to have a midafternoon snack.

That “organizational insights” service sounds like it could be ready in the near term. When do you anticipate releasing the “virtual chief of staff”feature?

I don’t know when [we’ll reach] that threshold. It could be a long time. We’re still doing a lot of experimenting. In the early days we worked quite a bit with Microsoft Research on basic research like question-and-answer bots. We also have a partnership with IBM, with the Watson team, which is more like collaboration on an ongoing basis. But a lot of those things are just very hard and far off. I think what we have right now is good. 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.