Meta buys AI analytics startup Sundial and brings its ex-Meta founders home

A quiet acquisition with a familiar cast

Meta has acquired Sundial, a startup that builds AI tools for data analysis, a company spokesperson confirmed on Friday, 23 October 2026, to Axios. Terms were not disclosed. The deal is small next to Meta's spending on data centres and models, but it says a lot about where the company thinks AI will change its own operations: in the way it measures, tests and monitors everything it ships.

Sundial was founded by two people who know Meta well. Julie Zhuo was Meta's vice president of product design. Chandra Narayanan was director of analytics at Facebook and later ran analytics at Instagram and at Sequoia Capital. Both are rejoining Meta along with other members of the Sundial team, several of whom are also former Meta employees.

"I left Meta nine years ago," Narayanan wrote in a LinkedIn post announcing the deal. He named Zhuo, Stanley Watt, Asa Levi and John Maier among those joining, and said the group would "build the next generation of AI analytics tools and agents." He described the move as "coming home."

Sundial will stop serving outside customers when the deal closes, according to Axios. That makes this an acqui-hire with technology attached: Meta is buying a team and its software for internal use, not a product line it plans to sell.

Where Sundial lands: Meta's DATA unit

The team will join Meta's new data analytics organization, the Data Analytics Tooling & Agents unit, which Meta nicknames DATA. It is led by Alex Schultz, who became Meta's first chief data officer in July. According to his Meta leadership profile, Schultz joined the company in 2007, ran its analytics organization from 2015 and served as chief marketing officer from 2020. He now oversees analytics, data science, data engineering, competitive intelligence and user research.

In his own LinkedIn post, Schultz said Narayanan will lead Meta's data analytics tooling and agents work. "Harnesses are cheap these days. Vision, taste, and experience are not," he wrote, arguing that the Sundial team had a clear view of what analytics becomes and had shipped against it for demanding customers.

Axios reports that DATA is responsible for descriptive analysis, including identifying questions the company has not yet asked; for experimentation to test and improve Meta's data and AI products; and for monitoring the company's critical AI systems and issuing alerts.

Schultz's three-part plan for AI analytics

Schultz's post is the clearest public description so far of what Meta wants from AI in analytics, and it goes beyond dashboards. He listed three parts. The first is descriptive analysis that answers the company's questions before it thinks to ask them. The second is experimentation in which AI proposes tests, ships them, explains the results and decides what to do next. The third is alerting and monitoring that surfaces only the signals that matter, "at a token cost that makes sense."

"Done right, they become one loop," he wrote. Descriptive analysis works out what to ask, experimentation acts on it, and monitoring catches what changed and feeds it back, so that each turn is better informed than the last. He called the goal "a company that gets better at improving itself."

The experimentation piece is the most ambitious. Meta runs huge numbers of A/B tests across Facebook, Instagram, WhatsApp and its ads business, and deciding what to test and what to ship has traditionally needed data scientists and product managers. An AI system that proposes and ships its own tests, and then recommends decisions, would move a lot of that judgement into software. Schultz did not say how much human review would remain.

"AI has created effectively unlimited demand for insights, and we want to define the future of AI analytics," he wrote. He also said that in internal surveys about 8 in 10 new hires tell Meta its AI analytics tools are the best they have used, and that the company is not satisfied with that.

What Sundial built

Sundial's product let teams ask questions about their data in plain language and get back analysis, without waiting for a central data team. Axios describes it as turning natural-language questions and complex data into simple, actionable insights through a chatbot plugin, so that teams across an organization can do their own analysis.

According to Crypto Briefing's summary of the deal, Sundial was founded in 2021 and raised $23 million in total, including a $16 million Series A in July 2025 led by DJ Patil of GPV, with Sequoia Capital among its backers. It also joined the Open Semantic Interchange, a Snowflake-led effort to standardize how business metrics are defined across AI tools, in April 2026.

Narayanan's post adds detail on who used the product. He thanked David Sasaki and his team at OpenAI, "who pushed us harder and further than almost anyone," and named Alfred Lin at Sequoia Capital, Mike Vernal at Conviction and David Fischer at 01A Advisors among Sundial's investors. OpenAI is a direct competitor of Meta in AI models, and the reporting does not say how its use of Sundial will be wound down.

Why it matters beyond Meta

The deal fits a pattern in Meta's AI push, which has involved heavy spending on infrastructure and talent. The company recently created a non-voting class of shares to help fund that push, and its consumer agent, Muse, has been expanding to new markets such as Japan. Sundial points to a less visible front: using AI agents to run the company itself.

For the analytics software market, the acquisition removes one independent player and is a reminder of how much large companies value teams with experience in self-service analytics. Many vendors, from cloud data platforms to business intelligence tools, are adding chat-based analysis. Sundial's customers now need to find a replacement.

There is also a cost angle in Schultz's mention of token cost. Running AI agents constantly over a company's metrics is expensive, and Meta is signalling that it wants monitoring that is selective rather than exhaustive.

What to watch

Meta has not said when the deal will close, how many people are joining, or what it paid. The next signs of progress are likely to come from inside the company, through tools its employees use, rather than through new products for outside customers.

The bigger question is how far Meta lets AI drive product decisions. Proposing experiments and summarizing results is one thing; letting a system decide what ships across apps used by billions of people is another. Schultz has set out the goal clearly, and the Sundial team will be judged on whether that loop holds up at Meta's scale.