Last week, I was hunting for new trail running shoes. Not just "any shoes," but the kind that survive wet roots, three hours of downhill, and my knee, which starts negotiating for early retirement around the 12-kilometer mark. I opened BergFreunde.eu (a Decathlon story in a different package), where I've dropped enough cash over the last three years that they could at least send me a handwritten postcard. And sure enough - the next morning, a "personalized" email awaited me. Subject: "Igor, Today Only!" Content: a sale on kids' swimsuits, 10% off weights, and some weird vitamin pack that looked like someone ordered it at 2:15 AM.
I sat there, staring at the screen, thinking: seriously? In their systems, I'm a "male 35-44," someone who bought shoes, a backpack, and poles. And their best attempt at "personalization" is to try and sell me weights. (As if someone slapped a "male" sticker on my forehead and then dumped whatever was left in the warehouse.) That same day, I opened ChatGPT and typed three sentences: where I hike, what the terrain is like, how much I weigh, and that I dislike overly stiff soles. In 10 seconds, I got a better advisory experience than any retailer had offered me in the last decade - and all without my birth date, without a "loyalty card," and without 37 cookies.
Looking back, it wasn't that they missed the product that annoyed me. It was that they missed the moment. I wasn't a "segment." I was a human with a purpose: to buy trail running shoes, now.
Current state of affairs is absurd: retailers are sitting on a mountain of data, yet they still communicate like a coffee machine with only three buttons. Personalization is sold as magic, but in practice, it's often expensive segmentation with a little lipstick - and yes, it kills sales because it ignores the user's immediate intent.
In our practice, I see the same pattern again and again. Personalization in retail means: "who are you," "what did you buy," "when did you click," "which segment do we put you in." Then, from this, magical logic emerges: "male 35-44, bought shoes, let's send him socks." And if he doesn't buy socks, we move him to the "unresponsive" segment and send him an 8% discount in 14 days. As if the problem is the discount, not that you're selling him the wrong thing at the wrong time.
Even funnier: a huge amount of "personalization" is just cosmetic. First name in the subject line, dynamic banner, recommendation engine, three rules in marketing automation. These are all variations of pre-prepared scenarios. This isn't a response. It's a costume.
If you have 12 segments and 40 "if-then" rules, that doesn't mean you understand the customer. It just means you've built a labyrinth where even your own team gets lost.
Let's put it this way: most retail personalization is optimized for CTR. A click. A micro-victory. A nice graph on Monday morning. But CTR is often a measure of "how well we interrupted someone," not "how well we helped them."
ChatGPT, however, optimizes for something entirely different: the relevance of the next sentence. And the user feels it immediately. When someone helps you, you stay in the conversation. When someone chases you with discounts, you run away.
Let's look at a concrete example from a retail team we worked with last year: they had a series of "personalized" emails that boosted CTR by 23%, but sales remained practically the same. Why? Because people clicked out of curiosity, then landed on a page with three unrelated recommendations and a generic "Top picks for you" banner. Two clicks later, they were back on Google. That's not a sales funnel. That's a merry-go-round.
Here's where it gets complicated: once you teach an organization that a click is the goal, you'll get clicks. Even on kids' swimsuits.
This surprised me: ChatGPT often knows almost nothing about you. Sometimes it has five lines of context that you typed in the last 30 seconds. A retailer, on the other hand, has 200 attributes: age, gender, purchase history, last 90 days of clicks, categories, cart value, "churn probability," plus some internal tag like "VIP_3."
And then the paradox happens: the retailer knows 200 things about you, but sends you a dumb recommendation. ChatGPT knows 5 things about you, and nails the essence.
The game-changer is that ChatGPT doesn't "personalize the offer." It personalizes understanding. It reads the context on the fly, understands language, infers intent, and adapts the response in real-time. It doesn't look for which box to put you in, but what you want to achieve right now.
Fact is: the problem for most companies isn't a lack of data. The problem is a poor translation of data into a relevant response. Data is like a warehouse of parts. If you don't have a mechanic, you just have a pile of scrap metal.
At FrodX, we often see that the data is "serious," but the usage is childish. CDP, CRM, DWH, events, identities, consent… all correct. But then the marketing team turns it into: "if he bought X, show him Y." And that's it.
The problem is that buying X in the past isn't the same as the intent today. I might buy trail running shoes in March, but in April, I'm looking for a waterproof jacket because I've been soaked to the bone twice. If they're still trying to sell you socks in April, they've proven they're not listening. And once a customer feels you're not listening, you're done. Not immediately, but surely.
Here's the trick: intent is a signal, an attribute is history. Retail systems are obsessed with history because it's organized and neatly stored. Intent, however, is messy, linguistic, sometimes contradictory. But that's precisely why it's valuable.
And yes, this means "personalization" isn't a marketing project. It's a product project. If your personalization is just a campaign, the result will also be just campaign-level: a little noise, a few clicks, a bit of user fatigue.
Check if your personalization is based on attributes or intent - if it's the former, you might as well turn off your campaigns. Better nothing than expensive coercion with wrong suggestions.
Now for the part most people skip because it's uncomfortable. If ChatGPT (or someone else) also gets the logistics layer - inventory, delivery, returns - many e-commerce providers will become mere "warehouse managers." AI will be the advisor, the store will be fulfillment.
Why? Because trust in advice is already on the side of conversation today, not on the side of banners. The user prefers to ask: "What should I buy for a 3-day trek with a 1,200m ascent?" rather than clicking through filters that someone last updated in 2021. And if the conversational interface can explain "why," the battle is almost won.
This isn't science fiction. This is just the next step. And once someone packages it into a simple buying flow, your "recommendation engine" will look like an elementary school calculator.
A dialogue isn't a chatbot that, after three questions, says: "Sorry, I don't understand that." A dialogue is a system that can assemble a meaningful next step from your existing data and current signal.
In practice, this means three concrete things. First: on your website, you need to be able to recognize whether a person is exploring or buying - the difference is huge. Second: your communication must have a memory of the conversation, not just the segment ("Last time you said you don't like stiff soles" is a different league than "hello, Igor"). Third: you need to measure "progress in decision-making," not just clicks; with one team, instead of CTR, we started tracking how many people reach a comparison of two models and how many of them buy from there - and that showed more in 6 weeks than a year of CTR graphs.
In short: personalization that works is no longer about "which banner," but "which question" and "which answer."
If you're a marketing, sales, or e-commerce director, answer these five questions. No sugarcoating.
1) Can you explain today, based on what your system infers what the user wants at this moment - not what they bought last year?
2) How many of your "personalized" elements are actually just variations of pre-prepared scenarios (name in subject, dynamic banner, recommendations)?
3) Do you measure the success of personalization with CTR or with a metric closer to the decision (e.g., adding to cart, comparison, returns, repeat purchase)?
4) When a user changes intent (e.g., from "running" to "hiking"), how long does it take your system to grasp that - 5 seconds or 5 campaigns?
5) How much data do you collect "just in case," because you don't have a clear idea how you'll translate it into dialogue?
If any of these questions give you a stomach ache, that's a good sign. At least you know where you stand.
You don't need another attribute. You don't need another table. You don't need another "segment with lipstick." You need the ability to create a relevant response from what you already have, at the right moment.
ChatGPT doesn't win because it's miraculously smarter or because it has more data about you. It wins because it's present in the moment and can construct meaning from an incomplete signal. Retailers, on the other hand, often do the opposite: they construct nonsense from complete data.
Stop collecting the 201st data point about a customer. Instead, contact me to see how we can turn your existing data into a dialogue that actually sells.
Email me at igor.pauletic@frodx.com.