An AI shopping quiz should be able to say the shop has nothing for you

In its October 23 press release, Octane AI says Phoenix can build quizzes, explain product matches, check recommendation logic and suggest experiments. Its help documentation describes a builder that works with the store's Shopify catalog and lets merchants refine the result through conversation or manual editing. This is a product launch, not an independent test of how well those features perform.

The detailed Find issues page is more useful than the broad promise that a quiz checks itself. It names unreachable pages, loops, deleted answers still referenced by rules, formula errors and products that are archived, missing or out of stock. It also says the check examines every answer combination, or a sample when there are too many. A pass should not be read as proof that every possible journey was tested.

For a shop that changes stock regularly, catching an old recommendation is an ordinary, worthwhile convenience. Nobody wants a new AI tool to become another place where a sold-out item has to be remembered and removed by hand.

Phoenix describes a dead end as a shopper finishing before any results page. That is a structural problem. A deliberate results page saying that no current product meets the shopper's requirements would serve a different purpose: the customer gets an answer, even without getting a product.

Take the imagined lamp shop. Cordless power might be essential because the lamp is going on a table far from an outlet. A preference for a green shade might be negotiable. Treat both answers as points in one ranking and a sufficiently attractive green corded lamp could win. The merchant needs to distinguish requirements that rule a product out from preferences that help rank the products left.

The sources reviewed here do not establish how Phoenix handles that exact case by default. This is a test to bring to the builder, not a report that Phoenix has failed it. Ask for a case where the catalog cannot satisfy the requirements, then preview what the shopper actually sees.

Before launch, choose a requirement your current catalog cannot meet. Complete the quiz with that answer. Then try a combination where the only suitable item is unavailable. Check whether the result explains the missing requirement or silently substitutes something else.

An alternative can still be helpful. It should say what changed: this lamp needs a socket; this other version exceeds your stated budget. Let the customer choose whether that compromise is acceptable rather than hiding it inside a confident match explanation.

Run the built-in checks as well. Octane AI's issues page says merchants can choose Publish anyway despite findings, so the review is not an automatic guarantee. Recheck after changing questions, recommendation rules or the catalog. Keep the first quiz small enough that someone who knows the products can read its results.

A no-match result needn't abandon the customer. It can explain which requirement the shop cannot meet, offer a genuinely relevant stock notification or let the person leave without joining a mailing list. The next step should fit the reason there was no match. An email address will not make a corded lamp cordless.

For a merchant, fewer immediate orders from that branch can look disappointing. Compare that with what happens after unsuitable recommendations: clarification emails, cancelled orders or returns. Those are outcomes to examine in your own store, not benefits this launch has already proved.

A good shop assistant sometimes says, ‘We don't carry that.’ Putting the conversation on a screen should not make the shop less willing to say it.