Discover more from Three Data Point Thursday
Probably a good investment
Invest in probable data handling, DbtLabs lays off 15%, Google will crush gAI.
I’m Sven writing this to help you build things with data. Whether you’re a data PM, inside a data startup, internal data lead, or investing in data companies, this is for you.
New market: invest in probable data handling.
DbtLabs lays off 15%
Google will crush gAI
(1) Invest in the probable approach to data
Lauren Balik will probably write another offensive blog post in the future.
That probably is all I need to decide not to write about her stuff.
Probably is almost always enough to make a decision.
The upside of probably: Probably is faster! Much faster!
The downside of probably: No one is used to probably in the data space.
Subscribe (free!) or someone will steal your data business & (data) users.
How do you use that?
People are starting to get into probable data structures and probable data querying. “When probably is good enough” is a talk at Buzzwords Berlin next month.
Apache DataSketches is a library for stochastic streaming.
=> Stuff is on the move.
What do you do about it?
As individuals? Probably nothing. Stay put; the tech isn’t mainstream yet.
As a startup, PM, or investor? I’d consider the probable approach everywhere.
The only downside to a probable approach is convincing people to use it. How hard can it be?
(2) gAI does nothing for your productivity
… but it will make Google a ton of money.
What you need to know: (1) as a company, integrate gAI into your existing products (like Google does) to actually turn a profit on gAI.
(2) as an individual, use gAI to summarize stuff. Don’t try to do more with it. However tells you they just 10x their productivity with AI is lying (unless they are in the summary business).
The juicy details…
Google’s demoed some impressive gAI stuff at Google I/O. They showed…
Integration into search
You ask, “is Bryce canyon or Arches better for a family trip of five?”
You get a new fat snippet on top with an answer
You also get a lot of explanations and exploration options.
Lots of links to how the answer got generated and links to follow-up.
Oh, and it is integrated into Google’s Ad system, so Google turns a nice profit on gAI, as Packy from NotBoring points out.
Okay, Sven, that’s how companies can turn gAI into money. What about me - you ask?
Enter Sam Altman (CEO of OpenAI)…
“[...] like most of us, Altman mostly uses ChatGPT for summarizing stuff and hasn’t fully adopted the other use cases or plugins.” (Packy at NotBoring)
Yes, true. We already know what to do with ChatGPT, so please stop reading those “10 ways to become more productive with ChatGPT” articles.
(3) DbtLabs lays off 15%
A week ago, Tristan Handy announced that DbtLabs would lay 15% of the staff off across all departments.
What you need to know:
Tristan thinks the tech market will take longer to recover.
He offers a fantastic leave package for his employees (including removing option cliffs and giving away laptops).
Another hoard of qualified people will hit the data job market.
DbtLabs handles laying off people smoothly. Kudos!
Hmm, so you thought DbtLabs was doing great? We got some random guesses at how good it is doing…
20,000 companies are using dbt in production (acc. to the announcement)
We guess, roughly 15-20% of dbt users are also paying customers, which makes about 3,500 paying customers for DbtLabs.
DbtLabs current valuation is at $4,2b with something like $150m in revenue (our guess extrapolating the numbers from last year).
That’s a 28-fold revenue multiple.
Doesn’t sound too bad, right? Databricks, at one point, was at a 500 revenue multiple.
Our guess on where the problem is…
We assume companies with 1k+ models are on enterprise plans, that’s about 20% of them. Making 700 enterprise customers for $33,6m yearly revenue.
Since we don’t see DbtLabs with many enterprise features yet, we assume this is a good guess, although likely slighty too low.
This means, something like 70-80% of DbtLabs profit come from smaller projects, huge amounts of smaller projects.
This likely makes for a big problem with servicing and support for lots of small fish.
Fun fact: Databricks turned the ship around by focusing in on enterprises…