Yoni Leitersdorf's company, Solid, sells a semantic layer: the plumbing that helps AI agents understand an enterprise's data well enough to do effective, valuable work that passes a customer's evals. As platform software, this is about as invisible as a product gets, and by Yoni's count there are probably 100 semantic layers a buyer can choose from right now, including the ones Snowflake and Databricks already ship. So I wanted to know how a startup like Solid can get market share in a crowded category, and how it decides which deals are worth chasing. One of his answers: the prospect who shows up saying "we know we need a semantic layer" is not the deal he's hoping for.

• What a POC has to prove before a data leader can take it through the finance/procurement gauntlet
• Why Yoni thinks startup speed stopped being a moat against Snowflake and Databricks given how fast bigger companies are now shipping features
• What he'd change about his customer discovery process if he started a third company

If your product is a horizontal platform or infrastructure your buyer can't see, I hope this episode sheds some light on the overlap between product, UX, and GTM for enterprise AI.

• GetSolid.ai
• Yoni on LinkedIn

Full transcript and show notes: designingforanalytics.com/resources/episodes/205

Logan Havern's company, Datalogz, sits on top of an enterprise's BI platforms (Power BI, Tableau, Qlik, Databricks Genie, and the SharePoint spreadsheets nobody admits to) and finds the duplicated, conflicting, and unused reports that AI is now generating faster than anyone can clean up. At one large pharma customer, executives "talking to their data" get wrong answers and then argue with the data team, because they believe Claude over the humans. The product that fixes that, which Logan calls semantic intelligence, is the one he thinks makes Datalogz a multibillion-dollar company. However, this is not what he sells first—and this is the topic we dug into today on this episode of Experiencing Data.

What Logan sells first is a 30-minute connection and a cost-savings business case finance can sign off on. That wedge took him a long time to find, and it sets him apart from other companies trying to sell “better AI answers” and “talk to your data” solutions. We also dove into the manual work [of 5-to-30-person admin teams] his product replaces, and my skepticism that usage data on a dashboard can tell you anything about whether it delivered value. If your analytics product's biggest value is also the hardest to explain, this one is for you.

• datalogz.io
• Linkedin: https://www.linkedin.com/in/loganhavern/

What happens when your product team suddenly moves six times faster at delivering features, but your sales and business traction is unchanged? AI, and particularly AI for coding, has made it easier than ever to ship features, but faster development doesn’t automatically create better products or better outcomes. The real question is whether what you’re building changes anything for the people who buy, use, and benefit from your product.

Today, I get into why AI evals and acceptance criteria measure the product, not its impact, and why your buyers are probably judging your POC on success metrics your team never wrote down. Toward the end, I make the concepts concrete by applying them to a fictional AI-powered pricing tool to show what defining quality looks like for both the fiscal buyer and the end user, and what to do with all that extra engineering velocity instead.

Your goal? Better; not just faster.

Links

• Episode 192: Analytics Product Usage Does Not Equal Value
• Episode 202: Surface Tension