Order a free copy of the Positionless Marketing book
Claim your copyRead time 7 minutes
Personalize anything. Everywhere.
Tailor every page, message, and moment with AI-driven personalization.
Why it matters:
In this post, marketers can learn that many marketing teams believe their personalization is working because their metrics say so. This post challenges that assumption directly, showing how programs built on personas, business rules, and even machine learning can be confidently wrong, and how the metrics often agree with the mistake. Marketers who read it will recognize the patterns in their own programs and leave with a clear framework for fixing them.
Key takeaways:
Rhylan Johnson, product lead for Optimove Personalize, does not own a cat.
He opened his Optimove Connect 2026 session by explaining that he bought a bag of cat food about a month ago. He was looking after someone else’s. One purchase. That was all the data the retailer had on Rhylan.
The next week, the site recommended cat toys. The week after that, urine pads. The week after that, joint oil. By the third week, the retailer had constructed an entire life for an imaginary geriatric cat. It was all inferred from a single bag of grain-free food that Rhylan bought.
The interesting thing, he said, was not that the retailer was wrong. It was that the retailer was confidently wrong.
That distinction is the core of everything that follows.
Most personalization programs do not fail because they lack data. They fail because they treat incomplete data as complete. They make inferences they have not earned, and then they act on those inferences with full conviction.
A persona is the most familiar version of this. We have all been in the meeting where Sally was introduced. Sally is 40. She has two kids, two dogs, and a fondness for 80s pop. Everyone nods. Everyone knows, privately, that Sally does not exist.
What we hope (and this is the honest version of how personas actually work) is that the blast radius of the email is wide enough that people like Sally receive it.
That is a workable strategy. It is also a confidently wrong one. And roughly 70% of businesses never move past it.
The businesses that do move past personas tend to leap. They skip to business rules, or straight to machine learning, chasing whatever sounds most advanced without proving anything along the way.
The result is familiar: more sophisticated personalization that is, on a different axis, confidently wrong about more people, faster.
This is the part that makes confidently wrong personalization so hard to catch: the numbers often look fine.
When a system measures its own outputs (click rates on the recommendations it served, conversions among the segments it targeted) it is measuring inside the error, not around it.
The cat food retailer’s recommendation engine was almost certainly showing solid engagement metrics on cat product clicks.
The Netflix recommendation prize is the clearest large-scale version of this. A decade ago, Netflix offered a million dollars for the best recommendation algorithm. Someone won. Netflix did not implement most of the winning approach.
Why not? The recommendations were too good. Users saw the same titles every time they logged in. They stopped discovering. Viewing time stayed flat. Login frequency quietly dropped. The algorithm had optimized perfectly for predicted preferences and missed something more important: the human need for novelty, for the unexpected, for a catalog that feels alive.
The metrics looked fine. The system was confidently wrong. And the metrics agreed.
The uncomfortable truth is that confidence without evidence is not a beginner’s mistake. It shows up at every level of personalization maturity.
At the persona stage, the assumption is that a fictional archetype maps onto real humans well enough to be useful. Sometimes it does. More often, it produces campaigns optimized for someone who does not exist.
At the business rules stage, the assumption is that if-then logic captures customer intent. Johnson calls this the “bus factor” risk: business logic so complex that only one or two people on the team can maintain it. When those people leave, the rules run on, confidently, based on assumptions nobody can fully explain anymore.
At the machine learning stage, the assumption is that a model trained on historical behavior knows what a customer wants next. Sometimes it does. And sometimes it builds an imaginary geriatric cat, faster and at greater scale than any persona could.
The tool changes. The underlying failure mode does not. Making inferences before the evidence has been earned is confidently wrong personalization, no matter what technical sophistication of the system producing it.
Elana Yentis, customer success lead for Optimove Personalize, put it plainly during the session: the fix is not jumping to the top. It is earning each rung.
The Personalization Ladder that Johnson and Yentis introduced at Connect 2026 is built on this logic.
It has four rungs: popular content, reactive personalization, predictive personalization, cross-channel personalization. The sequence is not optional. Each rung earns the right to climb to the next. Skip one, and the personalization above it rests on assumptions that have not been tested.
The first rung is also the most underused: show every user what is actually popular right now, based on real on-site behavior. No inferences about the individual. No invented personas. Just the honest truth of what is working, served to everyone. It feels too simple. It is not. It is the only starting point that earns its confidence from evidence rather than assumption.
From there, reactive personalization earns the right to get specific about the individual, responding to what they actually did, not what was assumed about them. Predictive personalization earns the right to anticipate, using behavior from similar users to surface what this one will likely want. Cross-channel personalization earns the right to apply that intelligence everywhere the customer touches the brand.
At every rung, confidence is the output of evidence. Not the input.
Optimove Personalize is the product Johnson and Yentis presented at Connect 2026. It is part of the Optimove Positionless Marketing Platform, and it is built around exactly this framework.
It gives marketing teams the tools to move through the Personalization Ladder at their own pace. It starts with proven popular content, building up to reactive and predictive personalization, and eventually delivering consistent, relevant recommendations across web, banner, search, email, SMS, and push. Each stage is connected to the same foundation, so the intelligence built at rung one strengthens everything above it.
It gives teams visibility into where they actually are on that ladder, rather than where they assume they are. That distinction matters. Most teams believe their personalization is working. The cat food retailer believed that too.
Personalization does not fail for lack of data or ambition. It fails when programs make inferences they have not earned and measure success in ways that confirm the error rather than catch it.
Confidence in personalization is not a starting point. It is something a program builds, one earned rung at a time.
To learn more about how Optimove Personalize helps teams move from assumption-driven to evidence-driven personalization, request a demo.
Generic experiences cost revenue.
Writers in the Optimove Team include marketing, R&D, product, data science, customer success, and technology experts who were instrumental in the creation of Positionless Marketing, a movement enabling marketers to do anything, and be everything.
Optimove’s leaders’ diverse expertise and real-world experience provide expert commentary and insight into proven and leading-edge marketing practices and trends.
What does “confidently wrong” personalization mean?
It means a personalization system that delivers recommendations or experiences with conviction based on personas, inferred attributes, or algorithmic logic that are actually wrong about the individual customer. The system isn’t uncertain. It’s certainly incorrect.
Why do even sophisticated personalization programs get it wrong?
Because sophistication and accuracy are not the same thing. A machine learning model can be more efficiently wrong than a simple persona. The root cause is the same: assumptions made before the evidence has been earned.
How can metrics agree with a mistake?
When a personalization system is measured against its own outputs, such as click rates on the recommendations it served or conversions among users it targeted, it can show strong numbers while missing the broader picture.
What is the first step toward fixing confidently wrong personalization?
Start with what is actually provable. Show every user content that is genuinely popular right now, based on real on-site behavior. No assumptions about the individual. Just the truth of what is working. Prove that it works, then build from there.
What is Optimove Personalize?
Optimove Personalize is a product within the Optimove Positionless Marketing Platform that enables marketers to deliver evidence-based personalization across web, app, email, and other channels. It moves through proven stages of maturity rather than leaping to unearned sophistication.
