See InsightAgent in action

Hi, thanks for joining today. Before we get started, I just need to quickly confirm a couple of things for compliance. You led the data and analytics platform at Meridian Retail until recently — is that correct?

Yes, that’s right. I ran the data platform team there until about eight months ago.

Thank you. And are you under any active agreement that would prevent you from discussing your experience with data infrastructure vendors at a general level?

No active NDA that covers this. I’m comfortable speaking at a general level — I’ll just steer clear of any specific dollar figures tied to the company.

That’s perfect. Let me quickly introduce myself — I’m Riley, an AI interviewer with the research team. I know speaking with an AI can feel a little different, so please take your time and expand wherever you’d like. We’ll cover about five topics on the cloud data warehouse market, and it should take around ten minutes. Does that sound good, or do you have any questions before we start?

Sounds good, no questions from me. Happy to dive in.

Great. To start broad — which cloud data warehouses did you actually run at Meridian?

We started almost entirely on Redshift, since we were already deep in AWS and it was cheap to get going. Over time we hit real concurrency and scaling pain, so around 2021 we migrated core analytics to Snowflake. Separately, the data-science group adopted Databricks for ML and data engineering. So by the end we were effectively running all three.

That’s helpful context. What specifically drove the move from Redshift to Snowflake?

Concurrency was the trigger, but the bigger draw was the architecture — separating storage from compute. With Redshift we were constantly resizing nodes, vacuuming, managing queues. Snowflake let us spin up independent warehouses per team and scale elastically, so we basically stopped doing that maintenance. For a lean team, the near-zero administration was the real selling point.

Makes sense. How did the economics compare once you were actually running on Snowflake?

The credit model is flexible, but costs can balloon if you’re not disciplined — an analyst leaving a warehouse running overnight adds up fast. We put real governance around it: auto-suspend on idle, resource monitors, separate warehouses so we could attribute spend. Net-net we landed maybe twenty percent above what reserved Redshift would have cost, but the agility was worth it.

Got it. And how did you divide workloads between Snowflake and Databricks — did you see them competing for the same use cases?

Initially it was a clean split — Snowflake for BI and SQL that business users touched, Databricks for engineering and ML. But that line blurred fast. Snowflake added Snowpark and AI features, Databricks pushed hard into SQL warehousing. By the time I left, procurement was asking whether we could consolidate onto one — they’d started to overlap that much.

Interesting. How would you characterize switching costs and lock-in on each?

Both are sticky, just differently. With Snowflake the SQL is fairly portable, but your pipelines, roles, and UDFs aren’t — so moderate lock-in. Databricks is stickier in practice, because of the notebooks, MLflow, and Delta tables your team builds around. Realistically, migrating off either is a six-to-twelve-month project. We wouldn’t have done it without a very strong reason.

Last question — where do you see pricing and competition between these vendors heading over the next couple of years?

More pressure, in buyers’ favor. As Snowflake and Databricks overlap, Databricks is using price aggressively to win SQL workloads, and Snowflake is bundling AI to defend its base. I’d expect buyers to push much harder on consumption transparency — the days of just trusting the credit meter are ending. The leverage is shifting toward the customer.

That’s a great place to wrap. Thank you, Daniel — this was exactly the kind of perspective we were after. That covers everything on my side; I really appreciate your time today.