Ask, Don't Infer
Behavioral personalization mistakes gifts for preferences and one-off purchases for interests. How to spot it on your store, and the one question that fixes most of it.
Most ecommerce personalization guesses what a shopper wants from what she clicked. Often it guesses wrong, and shows her the wrong guess under a heading that says Picked for you.
How it goes wrong
Recommendation engines usually assume two things:
- The buyer is the user. But many purchases are gifts, or are bought for someone else in the household. Buy a birthday present for a nephew and your profile now says you like what he likes.
- Every click is a preference. But people browse to help a friend, compare things they’ve already ruled out, or look out of curiosity. A click records attention, not desire.
There’s also the one-off problem. Buy a fridge, and the store recommends fridges for weeks. The purchase meant she was finished with fridges. The system read it as the start of an interest.
Why a wrong guess costs you
A generic section, such as bestsellers or new arrivals, doesn’t claim to know her. It’s neutral.
A personalized section does make that claim. So when it’s wrong, it’s visible proof that the store doesn’t understand her. And once she’s seen Picked for you full of the wrong things, she stops trusting that section, even when it later gets something right.
Not creepy, just wrong
This is different from showing a shopper too much of what you know: “you’ve viewed this four times.” That’s the creepy failure, and the fix is to resume, never report: use what you know to pick up where she left off, without reciting it back to her.
This is the clumsy failure: the conclusion is simply wrong. It’s fixed by asking more.
A hard line: life events
Inference can guess at more than taste. Browsing baby clothes can lead a system to assume someone is expecting. Browsing sympathy cards, that she’s had a loss.
Don’t act on inferences like these. Even when they’re right, they’re intrusive. When they’re wrong, they can be upsetting. Someone’s circumstances are theirs to tell you.
Check your own store
Buy one gift on a test account. Something clearly for someone else. Then watch what the store decides you are over the next few weeks.
Look at what follows a big one-off purchase. If buying a sofa triggers a month of sofa recommendations, completion is being read as interest.
Compare clicks on personalized and generic sections. If “Picked for you” underperforms “Bestsellers,” the personalization may be costing you.
Check whether shoppers can correct it. If there’s no way to say “this was a gift” or “not for me,” a wrong guess is permanent.
What stays
This is not an argument against recommendations. Done well, they save a shopper time, and a store with a large catalog needs them.
The argument is against inference that never gets checked: a guess that is made once, shown as knowledge, and can’t be corrected by the person it’s about. Keep the recommendations. Add the correction.
The fix: ask
One question beats a hundred inferred signals.
At checkout: an optional “Is this a gift?” Keep gift purchases out of the buyer’s own profile. If she’s willing to say who it’s for, you can be useful to her again next year.
On recommendations: a simple “Not for me” option, so a shopper can correct a wrong guess.
In conversation: “Who are you shopping for?” resolves in seconds what an algorithm spends weeks getting wrong.
Prefer what she tells you over what she clicks. A stated preference is what she meant. An inferred one is what the system supposed. When the two disagree, the stated one wins.
The answers are worth keeping. What shoppers say when asked is the same kind of knowledge that usually lives in one person’s head, and it belongs with the rest of what the founder knows.
What to measure
- Click-through on personalized sections against generic ones.
- Share of orders flagged as gifts once you start asking. Count your own. Most stores have never looked.
- Recommendation dismissals (“not for me”): a direct signal of where inference is failing.
Where Immerss fits
Immerss puts an AI sales agent on your store that answers shoppers from your catalog, at any hour. It does follow what a shopper is looking at during her visit, which is the same kind of signal this article is about. The difference is that it’s a conversation. The moment she types it’s for my nephew, he’s seven, her own words lead, and what it recommends comes out of what you stock for him. A recommendation row gives her nowhere to say that. A conversation does.
When she wants to see something before deciding, one tap brings a real person onto live video.
The gift checkbox at checkout and the “not for me” control live in your store platform and your recommendation app, not in Immerss. Immerss installs on Shopify, WooCommerce, Salesforce or any store by script, and it’s free to start. See how other stores use it in our customer stories, or get started.
You’re inferring from a hundred clicks what one question would have told you.
Questions
Why do product recommendations go wrong after a gift purchase?
Because most recommendation engines assume the buyer is the user. A gift is bought for someone else, so it adds another person's taste to the buyer's profile, and the store goes on recommending for that person until something corrects it.
Is a wrong recommendation worse than no personalization?
Often, yes. A generic section such as bestsellers or new arrivals claims nothing about the shopper. A section headed Picked for you does, so when it is wrong it is visible proof that the store doesn't understand her, and she stops trusting that section even when it later gets something right.
How can I check whether my store's personalization is misfiring?
Buy one obvious gift on a test account and watch what the store decides you are over the next few weeks. Look at what follows a large one-off purchase, compare clicks on personalized sections with generic ones, and check whether a shopper has any way to say that a purchase was a gift or that a suggestion is not for her.
Should a store act on life events it infers from browsing?
No. Guessing that someone is expecting a child or has had a loss from what she browsed is intrusive when it is right and can be upsetting when it is wrong. A shopper's circumstances are hers to tell you.
