By Phil Wainewright

April 22, 2024

Slack AI added a recap feature to accompany general availability of its AI-powered features last week, completing a roll-out that began in February. Pricing for the add-on has now been set at $10 per user per month for customers on Slack Pro and Business+ plans. As well as English, it is also now available to Spanish and Japanese language users, with other languages to follow. The time savings customers are seeing repays the extra cost, says Slack. In a briefing ahead of the news, Jackie Rocca, VP of Product Management at Slack, told us:

We have, through our pilot programs, been quantifying the time savings that people have had with our initial feature set. We've seen that an average Slack AI user is saving 97 minutes per week by using these features.

There are also cases where the value of finding information faster goes beyond simple time saved. One CEO uses Slack AI's channel summaries feature when customers call out of the blue to quickly review what's happening with the account and be able to respond appropriately. At another customer, she goes on, there was an outage that required the attention of the technical lead. She wasn't immediately available, and by the time she had logged into Slack there was a long chain of messages about the issue. With the help of an AI-generated thread summary, she quickly got up to speed and was able to identify the required solution. Rocca says:

The time savings there maybe were 20 minutes, but the customer impact of resolving that issue, turning billing back on, that was even more material.

Helping to reinforce this value proposition, the newly available recap feature actually includes an estimate of time saved at the foot of each summary. Recap provides a daily summary of what's happening in a configurable selection of channels suggested for the user — these might be accounts or projects where they want to stay in the loop without knowing every detail, social channels, product news, et cetera. Having a daily summary means that the user can stay in touch with what's happening, without the distraction of having these channels active in their sidebar throughout the day. Rocca says:

We've gotten great feedback that, 'Oh, it feels so good that I was able to stay up to speed on these things. But I didn't have to spend all day, I didn't have to block [time for it].' A lot of people are spending their hours, after hours, catching up on these things. Now you can do it much more efficiently, stay in the loop on what you need to know about, but do it in a much more effective way that's not sucking you out of your day, all day long.

Existing features include a conversational search bar which provides concise answers to search queries along with citations of the source material, and on-demand conversation summaries — such as the customer examples given by Rocca — which briefly summarize a busy message thread or a specific date range from a channel, again with citations.

To ensure that the AI produces accurate, reliable answers, Slack is using a combination of techniques, including Retrieval Augmented Generation (RAG), which limits the AI to finding answers from within the source material and means it can provide links back to the original source. The technology is also using metadata to help evaluate the authority of the source. Rocca explains:

We are looking at the date of the different messages. We're looking at the engagement of the messages. We're looking at, is this more of an official announcements channel? Or is it a small social channel that doesn't have very much traction? We look at the sender of that message and how much authority they have in the organization based on other things that they post. So we're looking at a whole host of metadata to really rank what we think are the best answers that are available in Slack...

So it's really [a combination of] RAG to ground everything in your company data, it's around citations and sourcing, and then it's around ranking through that metadata to try to get the best results for you.

Upcoming features still to be made available include automatically generated notes from meetings held in huddles, Slack's impromptu audio and video chat tool, and bringing Einstein Copilot, the conversational AI assistant for Salesforce CRM, natively into Slack.

My take

The Slack AI features first unveiled last September are now finally available to all paying customers — at a price. At diginomica, we're not keen on vendors charging extra for AI that's still somewhat experimental and largely unproven. The feature set for Slack AI is not yet complete, with some in the pipeline and others still in development. And the accuracy and reliability of AI-generated answers and summaries is only now starting to be tested in live use. But as Rocca points out, there are quantifiable productivity gains from using the technology, and a lot of work has been done to ensure responses are pertinent and valid. Vendors also have to take into account the not inconsiderable processing cost of running inference on large language models. One final point in favor of Slack AI is that it's designed into the product in a way that doesn't require users to be prompt engineers or AI experts.

Customers will have to make up their own minds whether these capabilities provide enough value to justify the extra cost. It's likely that in many cases they will. But it won't be long before AI will become such a standard feature of enterprise applications that it won't make sense to offer a product without AI built in. The logic therefore for charging a premium for AI features seems short-lived at best.