Databricks just announced CustomerLake, the agentic CDP
The battle for the decisioning layer is on
Disclaimer: Databricks is a strategic partner for Snowplow, and also an investor in Snowplow.
Data lake >> lakehouse >> CustomerLake
Today the news finally dropped: Databricks is launching its own Customer Data Platform, called CustomerLake.
Before reading further, do check out the official announcement post on CustomerLake, and the Adweek interview with Databricks CEO Ali Ghodsi and CustomerLake lead Tasso Argyros.
The fact that the Databricks team in New York City, including many ex-ActionIQers, was working on a new CDP had been described by multiple folks as the “worst-kept secret in martech”. But until today’s announcement, the product scope and market positioning were much less clear.
On the product scope - it’s now clear that Databricks has an ambitious take on the remit for a modern CDP, and have built a ton since the autumn. Check out the excellent product walkthrough (26 minutes!) from Tasso and Justin DeBrabant on YouTube:
And the framing as an “agentic CDP” is smart - it positions CustomerLake into a known category with budgets and RFPs, and paints a picture of a 10x category improvement in the shape of “infinity campaigns”.
There are already some strong external takes published today, from folk deeply immersed in CDP and martech, like Matthew Niederberger (who got early access), Luke Ambrosetti from Snowflake and Tejas Manohar from Hightouch. But I want to go in a different direction...
CDP is not the main event
At Snowplow, we are adjacent to, but deeply familiar with, the CDP and martech domains. As neighbors, we have a somewhat different take on CustomerLake...
Fundamentally, CustomerLake is not a foray into the CDP category by Databricks, because the CDP as a standalone software category is dying. Tasso himself puts it well in the Adweek article:
“I think the CDP, as middleware, is going to go away”
This has been the “house view” at Snowplow ever since championing the composable CDP movement back in 2023: the CDP becomes the “squeezed middle” between customer data ingestion on the LHS, customer activation & engagement on the RHS and the data warehouse or lakehouse underneath.
Before I receive angry DMs from my friends in the CDP space - I am not trying to detract from the very difficult enterprise-grade problems you solve around identity, segmentation, personalization and the like! I get as frustrated as you do when those problems are hand-waved or underplayed by consultants or vendors. And it’s very clear from the CustomerLake roadmap that Tasso, Justin et al are taking those problems very seriously at Databricks.
But as I said when I brought this blog out of hibernation last month:
“the interesting frontier is no longer just about where customer data lives. It’s about what AI agents do with that data — in real time, in the moment, on behalf of customers and businesses alike. This decisioning layer around customer data is suddenly a Wild West”
If the CDP framing is a distraction, and the ex-ActionIQ have no interest in re-fighting the last war, then what is CustomerLake all about?
What is now at stake is decision gravity...
Decision gravity is the new data gravity
Yali and I both have the habit of calling trends early enough that we often get the trend right but the ultimate name of the trend wrong.
So in this case I am going to hedge my bets: I am not sure if the ultimate name for the new battleground is decision gravity, or workflow gravity, or (business) process gravity. For the rest of this article, I’ll stick with “decision gravity”.
The last battle, around data gravity, was clearer to define, and easier to call - the hyperscalers won, the data platforms (Snowflake, Databricks, BigQuery) won, and more generally composability won. And the losers were vendors in certain categories of packaged SaaS, including in the CDP space - again, from my last blog post:
“many of the traditional players in CDP have been acquired (mParticle, ActionIQ), have pivoted (Treasure
DataAI) or have made architectural concessions (the ‘zero copy’ narratives of Adobe and Salesforce)”
By contrast, the new battleground is still hazy, still caught up in the fog of war. But I believe the new battleground is, broadly speaking, about who controls how and where business decisions get made...
The SaaSpocalypse is a civil war
What has been called the SaaSpocalypse is better understood as a civil war in SaaS, a fight over a scarce resource: which vendors get to own the workflows and decision-making for each line-of-business and functional area, from marketing to IT support to FP&A to procurement.
This area is under unique competition because of the arrival of a new Peter Turchin-style “counter-elite” - the foundation labs, specifically OpenAI and Anthropic. The labs are being extremely aggressive about moving up into the line of business, and they are pushing on multiple fronts:
General agentic tooling for business users, the now-established name for these being “CoWorks”
Specific wrappers for departments and industries such as Claude for Financial Services, Claude for Legal teams
Forward-deployed engineering teams to go into enterprises and solve specific applied AI problems
Joint ventures with private equity firms to deploy AI into their portfolio firms such as the OpenAI Deployment Company
If you take Anthropic specifically: the Amodei siblings have no emotional attachment, sunk costs or technical debt around how decisions are made in enterprises today. Their “hammer” is training bigger & smarter models to take on more and more tasks with less and less harness. And that hammer is finding more and more “nails” to impact: workflows, processes and decisions to power. This is true disruptive innovation in the Clayton Christensen sense.
And so, what we are now seeing is rapid, rhizomatic adoption of Claude across industries, departments and use cases. Anthropic, and also OpenAI, represent a very real displacement challenge to incumbents in every workflow area.
Going up the stack
If we come back to Databricks and Snowflake - both of these players are hungry for continuing growth, and one of the key pathways, following in the Oracle footsteps, is to go “up the stack”.
Hugo Lu of Orchestra wrote a very prescient post on this a year ago:
Hugo’s post was right about Snowflake - a few months after publication, Snowflake acquired Observe, climbing up the stack into the observability space:
But the post got the direction of travel wrong for Databricks: it is now increasingly clear that Databricks is joining Snowflake in going up the stack.
Ali Ghodsi has been candid recently about his lessons from building the open-source DBRX model back in 2024, coming to the conclusion that going down the stack with LLM model training was “not a good business for us to be in”:
If going down the stack is off limits, and if new entrants like Anthropic and OpenAI are aggressively going up into the lines of business - then Databricks needs to head up there too…
The marketing decisioning layer is a unique opportunity
Just like Snowflake, Databricks can pick and choose from the different opportunities up the stack. Shortly after the Observe acqusition, Databricks announced Lakewatch, its own security lakehouse (“agentic SIEM”), and just today announced the acquisition of Panther, an AI SOC platform, to support this.
At the same time - there is something uniquely attractive about the CDP space:
The incumbent vendors in the CDP space have already been heavily dislocated by the composability “border wars” of 2023-2025.
If you want to go and compete for CRM workflows, you have to fight Salesforce. If you want to disrupt the SDLC, it’s Atlassian. With IT service desk, it’s ServiceNow. For game dev, it’s Unity. Advertising, it’s The Trade Desk. ERP it’s SAP. All of these incumbents are extremely wise to the disruption that’s coming, and have no intention of ceding their decision gravity to challengers.
By contrast, there is much less “market lock” in the CDP space: who exactly is the 800-pound gorilla of the market, or even if there is one, is up-for-debate. All of the walled garden players in CDP were significantly weakened by the very composability trend that Databricks - and Snowplow and Hightouch - helped pioneer (example post from 2023).
This makes the decisioning layer for marketing workflows far more of an open playing field than some other spaces. This creates an opportunity for Databricks - but also for everybody else:
Nature abhors a vacuum, and if a CustomerLake does not fill the gap, then a Claude for Marketing Teams likely will.
Don’t fade on composability
One important thing I wanted to say:
Don’t think from this post that I believe it’s a “winner takes all” market around decision gravity - either globally or in a single workflow area. The future is still heteregeneous
This is partly because the genie of composability is happily very much out of the bottle, and partly because major players are worried about a monopoly emerging, perhaps owned by a single foundation lab.
Most notable here is Satya Nadella’s astonishing intervention on Sunday, his first article on X:
The last thing any of us want is a world where every company across every sector is ceding value to a few models that eat everything they see. If all the value is accrued by only a few models, the political economy will simply not tolerate it. There is no societal permission for an AI future that hollows out entire industries.
From our side, at Snowplow we continue to be deeply bought in to Scott Brinker’s composable canvas thesis, and plan to play our part with our own customer context layer.
Above all, Snowplow believes that enterprises have to be able to bring together diverse, differentiated, best-of-breed products and services (both built, assembled and bought) in order to compete.
A last word on CustomerLake
While I couldn’t resist the urge to go much broader brush on this topic, the last word should go to CustomerLake.
At Snowplow we are excited about the opportunities for our clients on Databricks to leverage their first-party customer data & contextual intelligence in CustomerLake.
Deep first-party integrations from inside Databricks can reach parts of the lakehouse - like Unity Catalog - that third-party CDP vendors struggle to support. And with CustomerLake, the set of building blocks for data-sophisticated organizations wanting to build great AI-enabled customer experiences just got a lot richer.
We all benefit from Tasso, Justin and band delivering their “difficult second album” in the marketing decisioning & workflow space. I’m looking forward to seeing what they do next!