What Comes After the Composable CDP?

For a bunch of smart people, the technology community can be awfully slow learners. One mistake they keep repeating is believing that the latest technology is a “silver bullet” that solves all problems. This despite years of experience finding that (a) no technology solves all problems and (b) every new technology is eventually replaced by something else.

The Customer Data Platform industry is as prone to these errors as everyone else. The current silver bullet is composability, which is touted as the ultimate solution to problems of CDP cost and complexity. Just hearing the claim should lead anyone to realize it’s not true: composability solves some problems in some situations (i.e., unnecessary data movement at firms with sophisticated technical resources) but it won’t help companies that lack the necessary data infrastructure or staff resources. I suspect that even the most ardent composable enthusiasts would agree with this more nuanced view if you pinned them to a wall.

It's a step in the right direction to move past “composability is great” to discussing “when composability is the best solution.” But it’s also important to address the second flaw in the “silver bullet” fantasy, which is to assume that the best current technology is the best future technology. In other words, it’s worth asking, what comes next?

Remember, composability isn’t a solution: it’s an architecture. The fundamental argument for composable CDPs is that it’s better to have an architecture that uses the enterprise data warehouse as the primary store for customer data, rather than an architecture where customer data is assembled, stored and accessed in a separate CDP database. So the question about what’s next is really a question of what other architectures are possible.

I can think of two candidates.

• Read customer data directly from the source systems, without assembling it in any persistent database (i.e., neither a data warehouse nor a CDP). This vision of real-time, on-demand customer profile assembly is what many people thought a CDP should be back in the early days of the industry. At that time, ten to fifteen years ago, the existing technology simply couldn’t make it work. Many source systems didn’t allow real-time access, internal networks couldn’t handle the volume, and the processing costs to assemble profiles in real-time were too high. The problem pre-dates CDPs, of course: the need to make data accessible by pulling it from source systems into a separate database is why data warehouses were first created in the 1980s. There has certainly been progress in recent years, although I can’t say that the necessary technology is fully available. Still, the scope of what can be done in real time continues to expand, which diminishes the share of data that must be assembled in a warehouse. This, in turn, diminishes the relevance of the warehouse-centric composable architecture.

• Create customer digital twins. I’m highly skeptical that AI can accurately mimic customer behavior: this seems most likely to reinforce mediocrity by failing to predict unexpected behaviors (the most interesting kind) or reactions to novel situations (which includes the most important  innovations). But I can still imagine creating a digital twin for each customer, updated in real time with data captured across all company systems. The twins would be self-contained objects (or maybe agents) that can be queried any time a company system wants to find the best action for a particular customer in a given situation. While their role might resemble the customer profiles assembled in today’s data warehouses, I’m sure the technology will be quite different. I can't predict the specifics of that technology but am sure that clever people somewhere are already working on it. My only point here is that, again, the architecture won’t look like the warehouse-based composable CDP.

There are surely other possibilities that haven’t occurred to me. In fact, I’d offer better than 50/50 odds that the next CDP silver bullet will be something I haven’t listed. Placing the right bet matters a lot to investors and developers but not so much to users, who can wait to see what becomes available. 

What does matter to users is identifying their requirements.  This is something that debates about architecture only distract them from doing. So my fundamental recommendation is that CDP users think less about chasing the silver bullet and more about building the golden record – that is, how to create the accurate, unified, accessible customer data that a CDP using any architecture is intended to provide.