A recommendation is useful when it respects the shopper’s requirements, explains relevant differences and lets them verify information before ordering. Evaluating an AI shopping assistant should include selection and purchase alongside mismatches, unanswered questions and the subsequent experience.
Showing three products after a question is straightforward. The work lies in why those products fit and what still needs checking. A plausible list can contain options that violate an important requirement.
What does “suitable” mean for this shopper?
Some conditions are firm: maximum size, compatibility with an accessory or a stated budget limit. Others are preferences, such as color, finish or style. We clarify these without turning selection into an endless questionnaire.
In a fictional example, someone has 115 cm of space for a desk. A 120 cm model does not fit merely because the price is attractive. Two models that do fit can be compared by depth, storage and a verifiable delivery estimate.
Missing information remains unconfirmed. Desktop dimensions alone do not establish whether two monitor stands and the person’s other equipment will fit.
How should the recommendation be explained?
We would show the matching criterion, differences between options and relevant limitations. For example: “This model fits your stated width. Its surface is shallower than the alternative. Space for both stands still needs checking against their dimensions.”
That explanation lets the shopper correct the requirement. Perhaps the width limit was approximate, or one screen will use a monitor arm. A recommendation can evolve with the information while preserving changes in context.
Who decides the order of products?
The store needs explicit selection and presentation rules. Availability, fit, price and commercial priorities may play different roles. A commercial preference should not create a false explanation of compatibility.
We would check whether the same request produces coherent options and whether the shopper can understand why each one was included. Persuasive wording alone is not the measure of quality.
What do we measure before purchase?
| Signal | What it may show | What it does not establish alone |
|---|---|---|
| Opening a recommended product | Interest in an option | Product suitability |
| Adding it to the cart | Movement toward an order | A completed purchase |
| Correcting the recommendation | A clarification need or an error | That every recommendation is poor |
| Moving to support | A need for further assistance | Automatic failure of the conversation |
A support handover may be the correct outcome for a question the system cannot resolve. We would check that context is preserved and the person receives the required help.
What happens after the order?
We examine cancellations, returns and recorded reasons where the information can be collected appropriately. If a product is returned because it does not fit, we review the recommendation and data used. We do not assume the assistant caused every return.
Team effort matters too. A quick answer that creates repeated corrections may shift work to support. Evaluation should include that time and error severity, alongside completed conversations.
How can we compare experiences with and without the assistant?
Visitors who choose conversation may differ from those who browse independently. A higher conversion rate in the first group does not automatically establish the assistant’s effect. We select an appropriate comparison and account for traffic source, availability and offer changes.
With low volume, we can start by reviewing concrete tasks and conversations without claiming definitive commercial impact. With sufficient volume, a properly designed experiment may provide better evidence. The method should be chosen before a result is presented as proof.
NOT MORE. can connect this experience to product information, interface design and agreed measurement. The recommendation should be understandable, verifiable and usable.
What should a shopper be able to check before accepting your store’s recommendation?