The most interesting thing about this week's holiday programme from a major US discounter is not the discounts. It is the assistant. The retailer has built an in-app AI helper that answers questions about price, availability and gift ideas, and it sits in front of the assortment rather than beside it (Target’s holiday push …). That placement matters: when the assistant becomes the first screen a shopper sees, the store's merchandising decisions are mediated by whatever the assistant chooses to surface.
An assistant becomes the first screen in the store
Retail apps have spent a decade as loyalty wallets with a search box attached. The new pattern is different. The assistant is conversational, it accepts vague requests, and it returns a short list rather than a wall of results. For a shopper standing in an aisle with a phone in hand, that is faster than scanning shelves, and for the retailer it is a chance to steer the decision before the shopper's eye reaches the shelf label.
The risk is equally clear. An assistant that answers badly is worse than no assistant, because it converts a browsing shopper into a frustrated one at the exact moment of intent. The failure mode is not technical; it is merchandising. If the assistant does not know which items are actually in stock at that store, it will confidently recommend something the shopper cannot buy.
Affordability is the message, not the model
How the assistant answers a price question
When a shopper asks whether something is a good deal, the assistant has to translate a merchandising position into a sentence. That sentence is where brand strategy becomes visible. A retailer positioning on affordability will have the assistant lead with unit price and comparison to last year; a retailer positioning on exclusivity will have it lead with scarcity and bundle value. The model is neutral; the prompt and the data behind it are not.
This is why the assistant is a merchandising artefact as much as an engineering one. Someone has to decide what counts as a good answer, and that decision encodes the retailer's price posture. Getting it wrong does not produce an error message; it produces a slow drift in price perception that shows up in traffic months later.
Creator-born brands fill the home aisle
The assortment test
The same retailer is rebuilding parts of its home assortment around brands that first scaled on short-video platforms, betting that creator-driven demand can be converted into shelf traffic (Target bets on buzzy T…). The logic is sound: those brands arrive with an audience and a content library. The test is whether the audience converts at full price, or only when the item is promoted inside the app.
An assistant changes that test. If the helper is trained to recommend based on availability and margin rather than on audience heat, creator brands may lose the advantage they had in the feed. If it is tuned to surface trending items, the shelf becomes a mirror of the algorithm. Either way, the merchandising team is no longer the only gatekeeper.
Store labour shifts from search to service
Exception handling becomes the job
If the assistant absorbs the search function, associates spend less time answering where-is-it questions and more time on the tasks that require hands: substitutions, exceptions, and the last hundred metres of fulfilment. That is a better use of labour, but only if the store's task model is rebuilt to match. Most stores still measure associates on availability and checkout speed, not on exception handling.
Younger shoppers already arrive through social feeds and AI answers before they reach a store, which means the associate increasingly meets a customer who has done the research and wants confirmation rather than discovery (AI, social media, digi…). Training for that conversation is different from training for product knowledge. It is closer to service recovery than to selling.
Where the assistant stops and the shelf starts
The boundary will be set by data quality, not by ambition. An assistant can only recommend what the inventory system can see, and most inventory systems are accurate to the case, not to the unit. Until shelf-level accuracy improves, the assistant will keep sending shoppers to items that are technically in stock and practically missing, and the shelf will keep contradicting the app.
Retailers experimenting with this pattern should measure two things from day one: how often the assistant's recommendation converts, and how often it is contradicted at the shelf. The first number is flattering. The second is the one that decides whether the assistant becomes the front door or gets buried three taps deep.
