Target's wish lists: 45% more demand, then AI on top

September 2026: a school list worth 45%

In September 2026 Target described how its site works during the back-to-school season. At the centre sits the wish list, together with a panel that suggests the next move. The order of operations deserves attention: first the measurement of the behaviour, then the model that supports it.

The company stated that customers who build a list generate roughly 45% more demand in the category, a figure reported by Retail Dive on 3 September 2026^[1].

That number has a readable denominator: customers with a list, set against the other customers in the same category. Here lies the difference between a metric and a promotional line.

The back-to-school season carries a lot of weight in the American retail calendar, with spending that this year could reach record levels according to the same outlet. On such a broad base, a lever that moves demand by 45% becomes a line in the accounts.

The original idea: the lever first, the model after

Many companies pick the model first, then look for a place to put it. Target took the opposite road: it isolated a behaviour that already produced demand, then asked AI to make it easier. The job handed to the technology stays narrow, therefore measurable.

Target's senior vice president for technology, Brad Thompson, described the mechanism as a gentle nudge. The system proposes products while the person fills in the list, so the customer's memory receives concrete help.

In the words gathered by Retail Dive, the suggestions come from three signals:

• what other customers buy in the same category
• the past purchases of the person filling in the list
• the supplies that specific school and its teachers ask for

The third signal is the most interesting one for builders: it brings local context into the catalogue. The kit list of one precise school's classroom is worth more than a national average. Thompson says this combination «is making a big difference», a formula that stays qualitative.

The panel that points to the next move

On Target's site there lives an element called the «next best action widget». It offers the customer one precise action, chosen on the basis of what that person is doing right then.

The panel stays on the page all year, with content that changes with the season. During the back-to-school period the recommended action becomes building the list, Thompson explained.

«We use AI to sift through all the signals of that customer's browsing session and recommend the next best action», the executive said. The same statement appears in the version of the story published by CX Dive^[2].

Here lies the second design choice: the surface already existed. AI fills a proven container, so the risk of breaking the experience stays low. Anyone launching a new section to host a model pays twice, in development and in habits that have to be rebuilt.

June 2026: the assortment arrives before the software

Three months ahead of the season, in June 2026, Target presented its back-to-school package. Inside there was a discount on the Target Circle 360 membership, school uniforms from an own brand and a limited-run collaboration with LoveShackFancy. The calendar says a lot: the offer stood ready before the traffic peaks.

AI arrives on top of this offer, so it has something to recommend. A recommendation engine on a weak catalogue produces weak recommendations. The wish list, in this scheme, becomes the folder where assortment and model meet.

The official announcement about the season's digital tools lives in the company's newsroom^[3].

For anyone leading a product the lesson is one of sequence. First the choice of assortment, then the measurement of the behaviour, at the end the model that accelerates that behaviour.

Stated result and verified result

Now the rough point, the one that makes the story useful. The 45% comes from the company, picked up by two outlets, with a reading that stays a correlation.

Customers who build a list are probably the most motivated in the category. Part of that extra demand existed before AI, inside the head of whoever opens the list. An A/B test on the product suggestions would settle the question in a week.

What is missing is what makes a case fully proven: a published control group, a before and after across the season, a stated time horizon. Target speaks of an effect measured on the category, therefore of a number with a denominator, yet it leaves the design of the measurement out.

For a reader this gap works as a note of method. A company figure with a denominator stays far more informative than a «significant improvement», so it deserves respect; full proof arrives with the comparison between groups. Asking for the design of the experiment is the job of whoever evaluates.

What changes for decision makers

The story speaks to four different tables, with a different message for each one.

• SME founders: the replicable playbook is the measurement of your best customers' behaviour
• CTOs: the solution in production is a suggestion inside a flow that already exists
• Boards: the wish list becomes a measurable e-commerce asset
• Team leads: the contextual panel travels to any page already live

For an SME the cost of the AI piece stays modest, so the real barrier is the discipline of measurement. An orderly catalogue and a metric with a denominator come first, before any model.

For a board the figure moves the bar on one precise point: collecting intent is worth as much as immediate conversion. The interesting comparison concerns the other seasonal categories, from the holidays to the change of season.

For anyone leading a team the replicable move is the contextual panel. A surface that already exists, content that changes with the moment of the year, one proposal at a time. The strength of that choice lies in the subtraction.

What you can take away from this story

The value of the case lies in the sequence, even before the technology. Target found a behavioural lever with a number attached to it, then put AI at the service of that lever. The reverse order, with the model looking for a problem, produces the pilots that stay stuck.

Three steps sum up the method:

1. measure a behaviour of your best customers, with a denominator
2. make that behaviour easier with an automatic suggestion
3. repeat the measurement against a control group

The step still missing, in the public account, concerns independent proof. Anyone replicating the scheme should build the control group from day one, so the number holds up in front of a hostile reviewer. The cost of that rigour stays low when the decision arrives before the launch.

There remains the question that holds for any organisation. Which behaviour of your best customers have you already measured with a denominator, ready to be made easier by a model? The answer shows where to put the next euro of AI, far better than a list of use cases.