The CDP schism: Salesforce vs. Snowflake in the battle for your data

Last year, at Simon Data, we faced a decision: based on an accumulating molehill of market data, do we build our Customer Data Platform (CDP) to be deployed within a client’s data warehouse?

As background, we’ve always viewed the Cloud Data Warehouse (CDW) as the center of gravity around which an organization’s data strategy and applications should orbit.

About six years ago, we decided to build our platform in the Snowflake ecosystem and aligned early with their vision of bringing the applications to the data vs. data to the applications.

This paradigm (composability) is defined here, for the sake of simplicity, as deploying the software directly within a client’s data environment was a bit of a Rubicon decision. Tradeoffs around product roadmap and resources notwithstanding, this was an architectural decision from which we likely wouldn’t return.

On one hand, we continued to see market interest in this composability. We had quasi-competitors like Hightouch claiming the CDP category as then currently conceived was “dead.” I sat through many hundreds of hours of sales conversations in 2022 where a lack of “composability” was a perceived deal-breaker.

Fundamentally, and the biggest factor in the pro-composable camp, was that the composable vision aligned with how we view the data warehouse within the client’s data ecosystem.

On the other hand, we had just as many conversations with clients who had attempted to deploy a composable CDP and were unsuccessful. There is, practically speaking, a spectrum of whether the product is deployed within or on top of the data warehouse.

For example, when your product is designed to power personalized messages across channels, data does eventually need to leave a client’s CDW before the ad, email or SMS is seen by the customer.

If we were going to fully embrace composability, we had decisions to make around which components of our application would need to be composable, in what order, and by when.

Understanding what composability means in the MarTech space

Underlying this decision of this was a question of whether the market understood what it meant for the software to be composable and by what standard. Ultimately, we questioned whether this was a zero-interest-rate fever-dream like trading ape JPEGs or the Metaverse vs. a lasting trend.

Speaking with self-awareness of my confirmation bias, now that we have delivered a fully composable CDP to market, I can confidently say two things:

1. There is still an uneven understanding in the market of what it means to be composable
2. Composability is not a fad and will have huge implications for the CDP and broader MarTech space

I won’t try to solve #1 here. Braver people than I will continue to lead that charge. What’s more interesting to me (and one of the complicating factors in the decision set forth above) is that there has been relatively little discussion of topic #2.

Considering the implications of composability for the broader market, it’s important to reiterate just how challenging it was to distill market feedback around “composability.”From speaking to clients all day, the composable CDP narrative sounded more like an elementary school recorder lesson and less like an orchestra (i.e. high noise to resonance ratio). The market generally struck some consistent notes like data security.

Beyond that, as someone whose job it is to (1) understand and explain how the machine actually works and (2) translate it into the business value it delivers, I struggled to do both.

Composability often felt to me like a religion. It was something that people just believed to be good or essential without it having to be explained or proven.

A year later, the recorder lesson is starting to sound like improv a cappella. Everyone in the space has figured out on our own what harmony to sing, but without the accompaniment of a 9-figure sales and marketing budget to have Matthew McConaughey broadcast it.

Clients well understand the benefits, even if the terminology is not widely synchronized. Demand is substantial and concerns around data replication or non-composable approaches have formed a beautifully dissonant tone of their own.

Recently, I’ve realized that there’s a schism widening within the idea of composability and data warehouses, and it has huge implications for the CDP space and MarTech more broadly, and that’s what inspired me to write this article.It occurred to me in Mark Benioff’s response to a seemingly platitudinal comment by a research analyst about attracting data science talent during Salesforce’s earnings call in November.

Here’s Benioff’s response:

“Well, I think that, that is very much a primary focus of the company, which is that — when we started this Data Cloud, we thought we were just building a CDP. And a CDP looked like an exciting market opportunity. We’re #1 in enterprise marketing automation. That seems like a great opportunity.But the more we started working on this product, we realized, oh, every one of our clouds needs this Data Cloud. And so Sales Cloud needs a Data Cloud, Service Cloud needs a Data Cloud. Yes, Marketing Cloud needs a Data Cloud, called that CDP. And Slack needs a Data Cloud. Tableau also needs a Data Cloud. If you’ve seen any of my recent demonstrations and propose it with these great Tableau customers in Japan, they all need Data Cloud on the back end of Tableau. And this idea that the Data Cloud will become the heart and soul of the product, be the engine of all of Salesforce’s apps and say you can use our models, our AI models or you can bring your own models into the Data Cloud, which is a very cool feature.This idea that it also has this incredible level of capability. But the amount of data that it’s already managing and the amount of data that it’s already ingested, that is what is shocking to us. And I think that you’re going to see as we get deeper and deeper into this so you can really see the level of data that we’re handling, the trillions and trillions of transactions. This is going to be the key to the AI working for enterprises.”

Overall, on that earnings call, Data Cloud was mentioned 42 times. Marketing Cloud was mentioned 3 times. From his comments, two things are explicitly clear:

1. Salesforce views the Data Cloud as the foundation underpinning the rest of their applications (i.e., similarly to how composable CDPs view the data warehouse and recognize that other applications, like analytics, should do the same)
2. Salesforce is making a long-term bet that their clients will move all of their data into the Data Cloud

If you believe those two things, you must also believe that now Salesforce and Snowflake are competitors.

Snowflake vs. Salesforce in the battle for customer data

It would be at best insufficient and at worst incorrect to say that this now means organizations will have to choose where they want their data to live. As those of us in the MarTech space know, it’s never that simple. Organizations that choose to centralize all of their data in Snowflake may still choose to bring their data into the Salesforce Data Cloud (unfortunately, both now use this term).

Given how Salesforce markets, sells, and then builds (or more often repackages) software, we’re years away from a customer having all of their enterprise data in Salesforce, if that’s ever even possible.

Beyond that, I started thinking about practical applications of Benioff’s statement in our client’s terms.

What if a client uses something other than Tableau for analytics and something in the Adobe ecosystem like Test & Target, as is often the case?

What if a client has invested heavily in centralizing their data in the Snowflake ecosystem (as is also often the case) and is concerned about maintaining parity between systems?

Will the Salesforce Data Cloud be extensible to support applications beyond its scope like data sharing, clean rooms, enrichment and identity resolution, or will it revitalize the corpse of Krux and try to make it compatible?

It struck me how the bet Salesforce is making obviously runs counter to the composable CDP vision and less obviously represents a complete divergence from the underlying trends that led to its emergence.

The reason I compare this to a schism is that it’s narratively driven and is not dichotomous in terms of the practicality of the deployments or even the belief sets around them.

My actual bet? In a multi-polar power dynamic, there tends to be a detente on practical dimensions even while there’s sword rattling on matters of strategy and vision.

They both have the scale, although Salesforce is about 10x the revenue and Snowflake is about 4x the growth.I have tremendous faith in the Snowflake product, team, and vision. Regarding Salesforce and other MarTech that orient to being your data platform, I think Benioff’s own words describe the challenges of their approach better than I ever could.

If Salesforce succeeds, I could certainly see some of my original hesitations — that composability represents collective industry navel gazing — coming to fruition.

I think it’s far more likely that Salesforce won’t be the ones to upend this trend, and just as they changed their views on the CDP space a half dozen or so times, they will ultimately walk this back (along with being the world’s largest AI company).