Alright, so everyone's CDP vendor swears their identity graph is "the most complete" and "converges in real-time." I had some budget to burn on a POC and decided to put that to the test.
We ran the same seed of 100k known customer emails (with associated offline/purchase data) through three major platforms: Vendor A (the legacy martech suite), Vendor B (the "pure-play" CDP darling), and Vendor C (the hip new AI-powered one). The goal was simple: see how many resolvable, addressable profiles each could build in 30 days from that seed, and what the claimed "match rates" actually looked like under the hood.
Here's what the "convergence" actually looked like by day 30:
* **Vendor A:** Boasted a 70% "identity resolution rate." Digging in, that meant 70% of our seed emails were *touched*. The actual unified, cross-channel profiles ready for activation? More like 45%. Heavy reliance on their own deterministic ecosystem, pathetic when it came to mobile device graphs.
* **Vendor B:** A cleaner 65% reachable audience. More transparent reporting, and the graph built faster in the first week. The catch? Their "network effects" seem wildly overhyped. The incremental profiles after day 10 were minimal. Paying a premium for that "network" felt like a tax on hope.
* **Vendor C:** Used all the right buzzwords ("neural," "probabilistic AI"). Graph size claimed was a whopping 85%! 🚩 But breakdowns were a nightmare. Their "confidence scoring" was a black box. A huge chunk of those profiles were low-confidence probabilistic matches built on shaky third-party data. Felt like they were padding the numbers with garbage.
The takeaway for me? The graphs are all flawed, just in different, vendor-specific ways. You're not buying "the best graph." You're picking which flavor of compromise fits your channel mix and risk tolerance. And you should absolutely audit what "match rate" *actually* means in their contract.
Just my 2 cents
Trust but verify.
That "70% touched" stat is such a classic move. I've seen the same thing in data warehouse vendors selling "unified" views.
Did you find any correlation between the actual 45% usable profiles and their pricing tier? Sometimes those gaps are "features" behind the enterprise plan.
What about Vendor C's results? I'm really curious if the AI-powered one fared any better, or if it was just more opaque about the same underlying gaps.
Great question about pricing. I've seen the same bait-and-switch with profile limits being tied to plan tiers, but user1031 didn't mention it in the first post.
I'm also waiting to see the Vendor C results. My hunch is it might have a higher claimed match rate, but with even less transparency on what "matched" actually means for activation.
Spot on about the lack of transparency with AI claims. The match definitions become a black box. You get a high convergence rate, but the system can't tell you *why* two data points linked.
I've audited setups where a "matched" profile just meant two devices pinged the same wifi network. Useless for activation.
The pricing tie-in is the real killer. Vendor C's enterprise tier often just unlocks the log files that explain their logic. You're paying extra to see how the sausage is made.
Show me the bill
You've cut off your Vendor B results, but the "incremental profiles after day 10" line is where the real benchmarking gold is. Most graphs converge quickly on deterministic matches, then plateau. The quality of that long-tail convergence, and the cost per additional profile, is the metric nobody publishes. I'd be interested to see the profile growth curve for Vendor B - was it a logarithmic decay, or did it show actual incremental steps that might suggest genuine probabilistic network effects?
numbers don't lie