Allegro AI Assistant: conversational discovery before autonomous shopping

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A marketplace buyer may know the recipient, problem or desired outcome without knowing the product name. Conventional search makes that buyer translate a need into catalogue language. Allegro’s response in Poland is a conversational discovery layer in its mobile app: the buyer describes an intent, and the assistant proposes relevant options, supports follow-up questions and comparisons, and explains why an option may fit.

What Allegro deployed

Allegro began a selected-user mobile-app test in November 2025 and announced the assistant for all mobile-app users on 7 May 2026. Public evidence supports intent interpretation, product matching, comparison and recommendation explanations. Allegro says the implementation combines its own infrastructure with Google’s latest solutions and large language models, but it does not name the model or disclose its catalogue retrieval, ranking, product-data schema, price or stock freshness controls, sponsored-offer treatment, evaluation protocol or quality rate. 1 2 3

A Google shopping assistant mentioned for desktop and mobile web in Allegro’s first-quarter update was a separate pilot. It should not be treated as evidence about the in-app assistant’s scope. 4

Discovery is not autonomous commerce

The evidence draws a clear authority boundary. In May, Allegro described one-seller order completion, easier basket-building and finding the best offer as future developments, rather than live capabilities. A first-person account from a stated test participant says the shopper still selects the product, adds it to the cart and pays. That article is disclosed paid placement by Allegro, so it is useful interface evidence, not independent performance validation. No retained source establishes autonomous order placement, payment, seller contact, price negotiation or request-for-quotation handling. 2 6

This makes the case useful precisely because it is narrower than “AI shopping.” It is deployed AI-assisted discovery with human-controlled transaction steps.

What the numbers do—and do not—show

Allegro’s 7 May release says more than 600,000 people made purchases with the Assistant during testing, then describes the same number as active users in the test phase. It gives no test window, eligible-user denominator, unique-user rule or purchase-attribution method. The figure is therefore an ambiguous company-reported test measure, not a conversion rate or proof of incremental orders. 2

On 17 August, Allegro reported 500,000 monthly active users and repeated around half a million monthly in its September second-quarter communication. The figures are not comparable with the earlier test measure and do not demonstrate a decline. Neither disclosure provides retention, query volume, product clicks, orders, gross merchandise value, customer satisfaction, returns, accuracy or controlled-experiment results. Usage is not, by itself, a quality, revenue or return-on-investment result. 3 5

Keep the seller assistant separate

Allegro also has a Sales Center AI Assistant for merchants. Official help documentation says that separate tool uses seller sales data and Allegro’s seller guidance, Academy and terms; updates its knowledge daily; cites sources; cannot change the account itself; stores conversations for 18 months; and may make mistakes. Allegro later reported that it resolved 95% of sales-quality issues without human assistance, but the disclosure gives no unit, denominator, period or audit. That merchant metric must not be transferred to the consumer assistant. 5 7 8

An illustrative business application

For an illustrative distributor or manufacturer selling through a marketplace, the transferable lesson is to make product discovery explainable before granting transactional authority. Provide current offer attributes, compatibility or fit, price, stock, delivery and merchant terms, with an inspectable reason for each recommendation. Keep the authoritative owner responsible for technical suitability and commercial facts; a conversational interface should not decide that an accessory is compatible or that a price is approved merely because the wording sounds plausible.

Start with a versioned query set drawn from real buyer language, including ambiguous needs that should trigger clarification. Measure eligible users, unique assistant users, sessions, recommendation views, product clicks, cart additions, buyer-confirmed orders and returns as separate events. Test compatibility correctness, stale price or availability, reason usefulness, user overrides and incremental conversion against a control. Treat best-offer comparison, basket construction, checkout, payment and merchant action as separate authority milestones. Allegro’s public record supports the first discovery step, not the later ones.

Sources and next reading

• Allegro official mobile-app page — consumer access channel; device, app version, region and account state may govern access.
• Allegro Sales Center — separate merchant tool requiring an eligible seller account and sign-in.

Explore OfferSignal · Focused Business & AI Workshop · All research

References