How ChatFeatured migrated from PlanetScale Postgres to Postgres Managed by ClickHouse to power AI brand discovery<br>Open searchOpen region selectorEnglish<br>Japanese

47.5kSign inGet started

->Scroll to top

BackBlog<br>User stories<br>Copy pageCopied!More actionsView as Markdown Open this page in Markdown<br>Open in ChatGPT Ask questions about this page<br>Open in Claude Ask questions about this page<br>Open in v0 Ask questions about this page

How ChatFeatured migrated from PlanetScale Postgres to Postgres Managed by ClickHouse to power AI brand discovery

ClickHouse<br>May 18, 2026 · 12 minutes read

Summary

ChatFeatured helps brands influence how they appear in AI search engines like ChatGPT, Perplexity, and Gemini, from analytics to content execution.

They were already running ClickHouse for agent analytics with 20x compression, but needed a way to run it alongside Postgres without managing two systems.

Postgres managed by ClickHouse gave them a single platform for transactional and analytical workloads, cutting analytics query times from 2.5 minutes to

When someone asks ChatGPT to recommend an OLAP database, or asks Perplexity what skincare brand works best for their skin, the answer they get isn’t random. It’s shaped by what AI models have read, cited, and learned to associate with authority. For brands, that means a new kind of visibility challenge, and a new kind of platform to solve it.

ChatFeatured is one of the fastest-growing solutions in this space. The Toronto-based startup helps brands track, optimize, and influence how AI models discover and recommend their brand. Where other tools stop at showing brands how they show up, ChatFeatured closes the loop, telling marketers what content they need and actually helping them produce it.

“We talked to customers, and they told us, ‘The data’s great, everyone’s giving me data… but what do I do with it?” says co-founder and CTO Nithiiyan Skhanthan. “They said, ‘I don’t know anything about search engine optimization. You guys are the experts… you take care of it.”

Loading video...

Nithiiyan and his co-founder Farris Nasr built what they call an “embedded AEO strategist,” effectively an AI agent that analyzes the sources AI models are citing, identifies what content would improve a brand’s visibility, and guides marketers through producing it. In just a few months, the platform has already attracted customers from all over the world, including billion-dollar companies like Cooper Consumer Health. “It goes to show how much of a problem this is, and that execution really is the most important part,” Nithiiyan says.

For a fast-growing, analytics-first product like ChatFeatured to deliver on its promise, it needs to handle transactional workloads and power complex analytical queries, ideally without the overhead of managing two separate systems. We caught up with Nithiiyan to learn how Postgres managed by ClickHouse gave them the best of both worlds.

Three providers, same problem #

The recent emergence of tools like Cursor and Claude Code has changed how founders think about technology, prioritizing speed-to-market before long-term scalability. As Nithiiyan puts it, “Most people, especially with the rise of AI coding, aren’t asking, ‘What stack do I need to scale to 1,000 users?’ They’re asking, ‘What stack do I need to get this out as quickly as possible to validate the idea before sinking time into the architecture?’”

ChatFeatured’s story was no different. Like most startups, they began with Postgres. It was quick to get up and running, flexible enough to handle both application and basic analytics needs, and well-understood. “There’s a lot of talk out there saying you can do anything you need for your first version with Postgres,” Nithiiyan says. “Me being the person who had to put this together, I started with that methodology, and I think it makes sense.”

As they built the first version of the product, the team cycled through three managed Postgres providers, starting with Digital Ocean. “It was cheap and easy… for about 20 dollars, you can get a decent-sized instance,” Nithiiyan says. While it worked fine at first, once the platform was ingesting around 1,000 prompts a day across six AI models, it became too slow for real users.

With the second provider they tried, performance improved. But as the customer base grew, CPU usage during nightly prompt ingestion windows climbed to 90%, leaving little headroom for growth. As Nithiiyan says, “If I want to add 100 more customers tomorrow, I’d need to scale the database up significantly just to maintain the same performance.”

Having heard good things about PlanetScale, they decided to try that instead. “Once I switched over, though, I wasn’t super impressed with the performance,” Nithiiyan says. On top of that, they were IOPS-bound, meaning the storage layer was hitting a hard ceiling on read/write operations. Upgrading to PlanetScale’s Metal tier could remove that ceiling, but with how much data the team was...