The Profile Problem Behind Your Failing AI Experiment
You bought the AI. The results never came. Before you blame the model or kill your AI KPI, uncover the Profile Problem undermining personalization, targeting, and growth.
You bought the AI. You connected it to your stack. You briefed the team, set the KPIs, told the board this was the year personalization at scale was finally going to happen.
And then... it didn't. Not really. The worst part is you're stuck trying to justify it.
The recommendations feel off. The targeting isn't converting the way the vendor promised. The model is doing something—you can see it working—but the results aren't there. So you start asking questions. Maybe it's the platform. Maybe it's the data feed. Maybe you need a different tool.
Here's what the research and the buyer community keep pointing to: the tool is almost never the problem.
The profile underneath it is.
What's actually happening
Reading across analyst reports, practitioner forums, and what's surfacing in buyer research right now, one pattern comes up consistently enough that it's worth naming. Call it The Profile Problem.
It goes like this.
A brand invests in AI-powered personalization—recommendations, targeting, next best action, whatever the use case. The AI model itself is solid. But the customer profile feeding it is fragmented across systems, partially owned by a vendor, refreshed in batches rather than in real time, or simply incomplete. The AI does its best with what it has. What it has isn't enough.
It's like judging an email campaign by the copy while ignoring the audience list underneath it.
The investment happened. The intent was right. The result falls short because the foundation underneath it can't support the outcome.
The result is personalization that's generic. Recommendations that miss. Audiences that don't convert.
But here's the part that makes this particularly costly—the AI is still producing output. Recommendations are being served. Audiences are being built. The failure isn't visible.
The hidden cost shows up elsewhere:
- Marketing spends another quarter buying traffic that should have converted.
- Media efficiency falls while acquisition costs rise.
- Teams buy another tool to patch a problem the first tool didn't cause.
The AI isn't failing loudly. It's quietly reducing the return on every growth investment around it.
The numbers back this up.
82% of retailers say maintaining real-time customer data is their biggest personalization challenge—not the AI itself (Mastercard, via Contentful).
89% of retail and CPG companies are using or testing AI, but only 7% have reached fully scaled deployment (McKinsey / Stord, via Elogic).
If AI capability were the constraint, you'd expect that gap to close over time as tools get better and more accessible. Instead, adoption is racing ahead while scaled results remain rare. That suggests the bottleneck isn't access to AI. It's the infrastructure AI depends on.
BlueConic's own research into AI readiness across commerce makes it concrete: 91% of retail leaders feel prepared for AI—but less than half can actually act on customer signals as they happen. Awareness is there. The infrastructure to do something with it isn't. (Is AI Exposing a New Growth Gap in Retail? - BlueConic, 2026)
We see the same pattern in conversations with retail and commerce leaders. AI initiatives rarely struggle because the model is weak. More often, the customer profile feeding that model is fragmented, inaccessible, delayed, or incomplete before the AI ever touches it.
The Profile Problem tends to show up in four recurring ways.
The four ways the Profile Problem plays out
The Profile Problem shows up in a few distinct patterns. If you're in a marketing leadership role, at least one of these will be familiar.
Way #1: The symptom swap
A brand decides their AI recommendation engine isn't working, replaces it with a new channel platform, and discovers 12 months later that the new tool has the same problem. Because the fragmented customer profile that broke the first tool is still there, feeding the second one.
The migration took six to twelve months, cost six figures, and solved nothing. The channel wasn't the issue. The profile was.
Way #2: The black box problem
The AI technically works—well enough to keep—but nobody on the team can instruct it, correct it, or understand why it's making the decisions it makes. It requires constant "care and feeding" without getting meaningfully better.
The model is fine. It just can't do much with incomplete identity data.
Way #3: The engineering trap
The customer data infrastructure exists, but it's owned and operated by engineering.
Marketing can't access it directly, can't build audiences from it, can't act on it in real time. Eventually, the team stops asking—because the answer is always a ticket, a wait, and a partial result.
So the activation platform ends up doing work it wasn't designed for. Attribution tools get bought to fill visibility gaps that a properly accessible customer profile would close. The stack gets more complex, the costs go up, and the root cause stays untouched.
Way #4: The disjointed ecosystem
Multiple AI tools, each optimizing its own slice, none of them sharing context. Personalization on the website doesn't know what happened in email. The loyalty platform doesn't connect to the targeting layer. Every tool is working—just not together.
What looks like an AI orchestration problem is really a Profile Problem by another name.
AI is creating a new divide in commerce
What's shifted recently is buyer awareness.
For a long time, marketing leaders knew their AI wasn't performing but couldn't articulate why. Increasingly, the diagnosis is becoming clearer—and it's redrawn the competitive landscape in a way that isn't yet widely talked about.
The divide forming across commerce isn't between brands that have AI and brands that don't. Almost everyone has AI now. The divide is between brands that own a usable, real-time customer profile and brands that don't.
The first group compounds every interaction. Each signal enriches the profile. Each decision gets sharper. Each investment becomes more effective than the last. The second group keeps adding AI to the same fragmented data and wondering why the results don't match the pitch deck.
The brands getting real returns from AI personalization—McKinsey documents 10–15% revenue uplift for retailers who implement it properly (McKinsey, via Elogic), with 70% of retailers who invested in personalization seeing ROI of at least 400% (Sailthru/Marigold, via Contentful)—aren't using fundamentally different AI tools. They're making decisions from better profiles.
And crucially, profiles they control.
Every interaction, preference, purchase, and signal becomes an asset that strengthens future decisions rather than enriching a vendor's black box.
Which raises an uncomfortable possibility: many AI initiatives are being evaluated at the model layer when the real constraint sits underneath it.
If that's true, some AI investments were never really AI evaluations at all. They were profile evaluations in disguise.
Most AI conversations start in the wrong place
The instinct when AI isn't working is to ask:
What model should we use?
What recommendation engine should we buy?
What agent should we deploy?
Those are reasonable questions. But they're the second questions, not the first.
The first question is: What customer profile is that system making decisions from?
Because every AI system eventually hits the same ceiling. It can't learn what it can't see. It can't personalize against data it doesn't have. It can't create value from fragmented customer context. Act one is building the customer profile that makes your existing AI investments start returning: clean, unified, portable, real-time, and brand-owned. Act two is what compounds on top of that foundation: personalization that adapts in real time, activation that doesn't require a ticket to engineering, AI that gets smarter with every interaction because the profile it's working from is actually yours.
Most brands are deep into act two conversations while act one is still unfinished.
That's where the growth is leaking. The risk isn't that AI stops working. The risk is that teams keep evaluating models, agents, and platforms while the underlying constraint remains untouched. Another pilot launches. Another vendor gets added. Another quarter passes.
The good news is that act one doesn't require starting from scratch. The data mostly exists. The AI investment is already made. What's usually missing is the connective layer that makes the profile clean, accessible, and brand-owned.
The question worth asking
Before your next AI investment—before the next platform evaluation, the next pilot, the next board update on personalization progress—one question is worth sitting with:
What's actually feeding my AI right now, and how confident am I in that data?
Most teams find that question harder to answer than they expected.
AI has become remarkably good at making decisions.
Most AI failures aren't intelligence failures.
You have a Profile Problem.
Sources
- Is AI Exposing a New Growth Gap in Retail? - BlueConic, 2026
- 39 Ecommerce Personalisation Statistics - Contentful, 2025 (Mastercard real-time data stat; Sailthru/Marigold ROI stat)
- AI in Ecommerce Statistics 2026 - Elogic Commerce (McKinsey adoption/deployment gap; McKinsey revenue uplift)
Alex Sarson is the Revenue Enablement Manager at BlueConic, where he works across sales methodology, commercial execution, and go-to-market strategy. He spent over a decade in martech - moving through client services, onboarding, and solution consulting - which means he's been in enough rooms with enough customers to know the difference between technology that sounds good and technology that actually solves something. That experience shapes how he thinks about AI and personalisation. Not from the outside in, but from the ground up - starting with real problems, real people, and whether the solution actually holds up when it matters. He made the move from solution consulting into enablement because one deal landing well was never enough. The question that drives him now is whether it lands for everyone, every time. He has little patience for complexity that doesn't need to exist - in martech or anywhere else. An American football fan living in England and a self-confessed beer snob, he's learned to back unfashionable picks and defend them convincingly. It turns out that's useful in tech too.
