The personalization vendors that quietly went dark in 2023 are back. Same logo, new pitch deck, one new word stapled to the front: agentic. The category that promised one-to-one experiences and delivered IF/THEN spaghetti is being resold to the exact same buyers who got burned the first time. Most of it is the same broken assumption in a better wrapper.
Sinequa put a number on the rot in a June report worth reading. 84 percent of enterprise leaders say they encounter rebranded products marketed as agents during evaluation. 87.5 percent say that experience has damaged their trust in AI broadly. 51.3 percent claim agents are in production today, but only 10 percent have deployed a true multi-agent system. The remaining 70.7 percent are running a sophisticated chatbot and calling it autonomy.
I sit in the forward-deployed seat inside enterprise retail, the place where the slideware meets the data warehouse. From that seat, the recycled-vendor wave is easy to spot and expensive to ignore.
The 2021 rules engine in an agentic costume
Here is the operator-grade tell, and Sinequa names it cleanly: true agency requires a system to independently decide how to pursue a goal, select its own tools, and adapt to the result. A rules engine does none of that. It waits for a merchandiser to author the segment, write the condition, and define the next-best-action. The marketer is the intelligence. The software is a switchboard.
Wrapping that switchboard in a conversational interface does not change what it is. It changes what it looks like in the demo. You ask it a question, it returns a templated answer, and the room nods. Behind the curtain it is still firing the same decision tree your team wrote in 2021, now narrated in fluent English. That is not a new capability. That is a costume.
The reason this matters for personalization specifically: personalization was always the hardest place to fake autonomy, because the failure is visible to the customer. A bad segment shows up as the wrong product on the homepage. A real agent should catch that, reason about why, and reroute. A repackaged rules engine just keeps serving the wrong thing, faster.
The broken assumption was never the model
Every personalization vendor that died had the same autopsy. It was not the algorithm. It was the substrate underneath it. Customer identity fragmented across six systems that never reconciled. Inventory truth that lagged a day behind reality. Consent and channel state that lived in three places and agreed in none.
You cannot agent your way out of a broken substrate. You make it worse. An autonomous system acting on fragmented identity and stale inventory will personalize at machine speed, with no human in the loop to catch the error, and it will be confidently wrong thousands of times before anyone notices the conversion dip. The model is not the problem. The model never was. The data foundation is the problem, and it is unglamorous, so the wrapper sells better than the fix.
This is exactly why Sinequa lands on governance, knowledge readiness, and operational trust as the real barriers, not technology. The retailers producing results are moving carefully, mapping where the system breaks before they let it act. The ones buying the costume are moving fast toward a faster failure.
The Cost of Doing Nothing compounds quarter over quarter
Here is the math that should sit on the CDO’s desk. The Cost of Doing Nothing is not a flat line. It compounds.
Quarter one, you buy the agentic-labeled rules engine and skip the identity-resolution work. It looks fine. Quarter two, your competitor who fixed the substrate first now has an agent that genuinely acts: it reconciles a customer across channels, reads live inventory, and reroutes the experience without a human writing a rule. Quarter three, that gap is not a feature gap. It is a learning gap, because their system is improving on real signal and yours is replaying a frozen decision tree. By quarter four, the distance between an engine that adapts and one that pretends to is structural, and you cannot buy it back at the speed they earned it.
The diligence question is simple and it filters the field fast. Show me a decision the system made that no human pre-authorized, and show me the moment it changed its mind. If the vendor cannot produce that on your own data, you are looking at the costume.
Personalization is back. The buyers are the same, the substrate problems are the same, and the vendors are betting you forgot. Fix the foundation first, then hand it to something that can actually act. The retailers who do that will not be running the most agents next year. They will be running the few that are real, and that will be the only number that matters.