Kyle Cheung from Greybeam - jumping over bathroom stalls.. as marketing
Greybeam's first ad was Kyle Cheung jumping over a bathroom stall to stop his co-founder flushing — because you should only need the small flush for a small query. That toilet gag is still their most memorable piece of marketing.
Kyle Cheung is co-founder of Greybeam, which cuts customers' Snowflake bills by 70-95% by routing BI queries to DuckDB instead, with no migration required. He walks through the six or seven pivots it took to get there — from an AI data analyst, to a self-populating data catalog, to SQL optimization, before landing on the current approach.
We also get into why he and his co-founder Arsham have built their entire marketing style around being deliberately ridiculous — toilet-flush metaphors, stuffed-animal microphones at conferences, hot-wing customer interviews — and why he thinks that's a legitimate strategy even in enterprise sales, not just something you grow out of.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What does Greybeam actually do?
We help customers cut their Snowflake costs, typically by 70 to as high as 95%. Snowflake charges a 10 to 20x premium on the compute it sells, and it turns out most read-only workloads — business intelligence and ad hoc analytical queries — don't actually need to run on Snowflake at all. We unload the customer's data from Snowflake into their own S3 bucket as parquet files, then query those files with DuckDB, an incredibly lightweight query engine that gets you Snowflake-level performance but can run on your laptop. Where Iceberg, the open table format, comes in is if we need to bring in a third query engine down the line — for us right now it's mostly an implementation detail. Onboarding is just changing the connection string in whatever BI tool you use, from Snowflake to Greybeam — you don't need to know DuckDB syntax or understand what's happening behind the scenes.
How did you land on this specific approach after so many pivots?
I used to lead data teams at a few different startups, and most of our pivots came from hypotheses about pain points I'd personally experienced there. We started with an AI data analyst to answer the repetitive questions that kept coming through our ticketing funnel even with self-service BI tools — that didn't work out. Then we tried a self-populating data catalog that scraped information from PRs, tickets, and Notion docs — but in customer interviews, people either didn't have budget for a catalog or just cared more about cutting costs. Snowflake cost kept coming up as the first or second highest line item for almost everyone we talked to, so we pivoted to SQL optimization — speeding up queries so they use less Snowflake compute. That worked, but only for a narrow slice of customers with particularly inefficient queries. We went back to the drawing board and landed on the current approach — moving data off Snowflake entirely for the queries that don't need it — around December.
How much do people actually spend on Snowflake?
It varies a lot, but on average people we talk to are spending $50,000 to $100,000. A mature company is probably spending $300,000 plus or minus $200,000, and once you get into late-stage startup or early enterprise, it can be anywhere from a million to as high as $25 million.
Who's the ideal customer for you?
Two profiles tend to work well. The first is a company that serves data as part of its product — one of our earlier customers is a cannabis analytics platform that aggregates point-of-sale and producer data into Snowflake, shapes it, and serves it as dashboards to stores across the US and Canada, so Snowflake is a real cost center for them. The second is a company using data as an internal operational tool — an ops team using dashboards as their ticketing mechanism, for example. For the first profile we usually talk to a CTO or senior engineering leadership; for the second it's more likely a VP or director of data engineering.
Your homepage headline changed from '70%' to '86%' savings. What else changed in your messaging?
The original version was 'route your Snowflake queries to DuckDB and save 70%' — it wasn't specific enough, and we started getting demo bookings from people who didn't actually have the BI-side Snowflake cost pain we solve. We only support read-only workloads right now, not writing transformations back into the warehouse, so I narrowed the messaging specifically to BI queries — companies with embedded analytics or BI as an operational workflow. We also updated the number to 86%, which is our new average.
Where did the toilet-stall video come from, and why keep leaning into it?
That was our first real piece of marketing: a video of Arsham about to flush the toilet after using it, and I jump over the stall wall and stop him — 'you only did a number one, there's no need for the big flush' — while we're actually talking about using DuckDB for your small queries and saving Snowflake for the big ones. It ran for a couple of weeks and it genuinely helped people who don't know Snowflake or databases quickly get the idea. People still bring it up to me in both good and bad ways — some tell me it's not good, but they remembered us because of it. I was ready to drop the whole toilet theme at one point, but the message clicks and everyone remembers us for it, so I actually want to keep it going.
Is being this silly actually okay in enterprise sales?
We haven't hit a wall yet where it's cost us a deal, and I think there might be one eventually — but everyone we're talking to is still just another person. At this very early stage, we need a way to stand out, get remembered, and get engagement on LinkedIn, and being silly gets us there. Supabase and PostHog are both multi-billion-dollar companies now and they're still very silly in their marketing, so I think we've got headroom left. If being this silly ever costs us a customer, we'll deal with that then — but we're not there yet.
How intentional is the content, versus just improvising?
It's usually on the spot — I'm there thinking, what's the most ridiculous thing I can do right now. One time I forgot to film an intro for a set of conference interviews, so back at the office I just rolled out from under a desk and said, 'what's up everybody, this is Kyle from Greybeam.' But we're also deliberate about brand where it counts without spending real money — our swag is designed to be something people would actually wear day to day, not just a plain t-shirt, and our first customer case study wasn't a vanilla 'we saved 90%' write-up. We flew to interview a customer on his own private plane, and the case study itself reads as an actual playbook — this is his exact situation, this is what he tried, this is how he uses Greybeam — rather than a badge on our website.
What do you know now about fundraising that you didn't when you started?
That a successful raise doesn't really describe how good a company or founder you are — it's just how well you can tell a story. When we raised our first round of capital, we were basically in the same position we'd been in a year earlier, we just had a more compelling story and someone believed us. If you're fundraising, you have to talk to hundreds of investors, because it's a numbers game — you're looking for the one to five people who resonate with your story, and most of them won't. Investors are mostly just trying to assess whether you're someone who can solve problems and articulate answers to their questions, and whether you can tell a compelling story about your journey and how the company scales.