For Jackie Rocca, who is vice president of product at Slack, a popular communication and collaboration tool, India holds a special place — the family ties to the country, and this being an important market for Salesforce, which acquired the platform in 2020 to the tune of a reported $27.7 billion. Since then, the company has gone from strength to strength. Slack doesn’t share specific numbers for its user base in India, but says the country is among the ten most important markets for the company, globally.
“India is on the forefront of so much that’s happening in technology. It is an incredibly important market for us,” Rocca said in an exclusive conversation with HT. At the turn of 2024, Slack completed 10 years, yet that milestone may just be coincidental timing as the company pivots towards extensive integration of generative artificial intelligence (AI). To that effect, there is a significant extension of the capabilities of Slack AI, a generative AI suite that includes enhanced search, as well as channel and thread summaries, with a new recap feature.
“Time savings is really important, and I think is one place that people are getting a lot of value from slack AI. It’s also in these critical moments where you need that speed and you need access to the right information at your fingertips,” Rocca points out. Those two are the very foundations of the tools Slack AI is enabling within the app, now. This finds a data illustration too, part of Slack’s latest Workforce Lab report, which indicates Enterprise users who’ve been using Slack AI already, are saving about 97 minutes a week, when searching for answers usually hiding in plain sight, within the many messages in their Slack conversations.
The recap feature, for example, delivers summaries of channels on a user’s Slack, in case something has missed their attention. Conversational AI really makes its presence felt with Search Answers, where a user can ask a question much like how they would otherwise speak with an AI chatbot or do a broader search, while citing specific Slack messages for context, as well as other Slack users who may be relevant to your search.
A genuine time saver for Slack users might be Conversation Summaries, which will allow a user to catch up on missed chats, date range choices varying between the last 7 days to a custom option going back an entire year – a summary generated here will link back to important messages in case you want more context and timestamp knowledge of an event or conversation, and negate the need of having to read every message sent in the time you were away, to piece together the puzzle.
There are two takeaways that emerge from this. First, it gives Slack as a platform an arsenal of tools, at a time when its diverse and often limited focus competition including Google Workspace, Microsoft Teams and Webex, are also making similar moves. Google, for example, is integrating Gemini extensively, while Microsoft has the Copilot generative AI across its suite of apps and services. Late last year, Cisco also added an AI assistant in Webex. According to numbers by research firm Statista, Slack is expected to cross 65 million monthly active users globally, that figure expected to be 79 million by the end of next year. A positive trajectory, from 14.6 million in 2020.
Secondly, as Rocca says, the real value of AI is to be found in those critical moments when you need correct information and context, when running against the clock.
Slack AI’s journey started in February, when it rolled out to a comparatively focused demographic of Enterprise subscribers as a paid add-on. Now, Slack says the AI add-on is available to all users globally, at $10 (that’s around ₹837) per month per user, on Slack Pro and Business+ subscriptions too.
LLMs, data privacy and an eye on regulation
HT asked Rocca about the large language models (LLMs) being used for Slack AI, and she confirms that no data of any Slack user ever leaves the Slack infrastructure. That is something they aren’t willing to compromise on, and take pains to reassert. “Trust and security, and data control was really important for us as we built Slack AI. We know that Slack data is some of the most valuable but also some of the most sensitive information you have in your company,” she says.
It is then that she adds on a contemplative note, “humans at your company are the most valuable asset you have. And they’re spending a lot of time in Slack. This was really at the forefront of our mind as we built Slack AI.”
There is a sense that Salesforce and Slack don’t want to take any risks with AI data safety measures, and the final implementation of generative AI functionality, with a serious push towards AI regulation in many countries. Late last year, the wheels were set in motion with the UK’s AI Safety Summit, a US executive order and a G7 order dictating policies of AI companies, towards broader AI regulation. This too will eventually see global implementation, likely in different formats, in the near future.
As things are, Slack AI is building on a certain amount of self-regulation. There are multiple large language models in play, a mix of in-house developed models as well as external models which the company doesn’t divulge the details for. To that effect, all LLMs sit within the company’s virtual private cloud. “No slack data is leaving Slack’s architecture. We do use both in house and external models, but everything is hosted within Slacks architecture, and no model training is happening with LLMs at all,” she confirms.
How does Slack find a balance between adding new generative AI features to the app, whilst ensuring time-consuming safety measures as well as safeguards are in place? That is something, Rocca insists, they will make no compromises with. Irrespective of whether they are behind the curve in new functionality, compared to rivals.
“Safety is incredibly important. We’ve built this from the very beginning with trust and safety in mind. Having the models housed within our VPC (that’s virtual private cloud) did take us longer to get to market with a particular solution, that we thought was the safest and most secure option available,” says Rocca. She insists that it allows Slack to abide by enterprise grade security and compliance requirements that their customers have.
“I think across the board and it’s really important to get that that foundation right,” Rocca says, summarising a no-compromise approach to getting the data privacy and AI safety basics right.
“The space is moving quickly. Our customers are really eager to use Slack AI, and we’ve had a waitlist for Slack AI that has exceeded our expectations. Trust and security is the most important thing. We’re trying to move as quickly as possible to deliver value for our customers,” she adds.
We asked Rocca about specific Slack user data that may be part of the learning sets for AI models, and whether a user has the choice to turn any of that sharing off. She confirms to HT that no customer data is being used to train Slack AI models. Instead, the company uses RAG, or retrieval-augmented generation, which broadly references a knowledge base outside of its training data sources, before generating a response. In Slack AI’s case, a lot of the information in replies to questions you have asked, will pick information from data that is already yours – private conversations and public channels, files, canvases and clips.
Slack, in due course, is looking to build region and country specific relevant functionality as well. The first step for that is taken by Slack AI’s support for English, Japanese and Spanish languages, and there are more languages that’ll be added to this list in the coming months. Rocca mentions plans to build a translator tool within Slack, with broader language support.