Vipps Shoppe: product discovery inside a payment app, not an autonomous buyer

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

A payment app may be a natural place to start shopping, but helping someone find a product is not the same as buying it. Vipps’s Shoppe relaunch draws that boundary clearly: the assistant can ask questions, compare products and explain suggestions, while the shopper checks the merchant’s live offer and completes the purchase there.

What Vipps relaunched

Vipps announced Shoppe as a permanent in-app feature on 8 September 2026. Current Help Centre material limits it to Vipps users aged 18 or over in Norway; the public Google Play listing’s 19 October update shows a current app listing but does not name Shoppe or prove feature exposure on a particular handset. 1 2 14

The assistant is not a general web search engine. Vipps says it matches requests against PriceRunner by Klarna and catalogues shared directly by businesses. Its relaunch announcement describes PriceRunner as comparing products and prices from more than 1,200 Norwegian online shops, but that is a Vipps-supplied figure, not independently verified Norway-specific coverage. Relevant offers may be absent. 1 6

PriceRunner describes itself as a comparison service rather than the retailer; prices are usually refreshed at least daily, often more frequently, and shoppers are told to check the retailer site. Vipps likewise says Shoppe cannot guarantee current product details, price, stock, delivery or company terms, cannot guarantee the best price, and has no historical-price access. A surfaced comparison is not a complete or contractual catalogue. 6 7 12

Assistance without purchase authority

Vipps says Shoppe uses an OpenAI language model via Microsoft on capacity reserved for Vipps MobilePay. It understands requests, asks clarifying questions and suggests products, while warning users to verify commercial facts. Public material does not disclose the model version, retrieval or ranking method, source-refresh service level, commercial influence or reliability evaluation. 3 4

The authority boundary is explicit. Shoppe cannot purchase, pay, contract, change Vipps settings or share payment details with a business. The user goes to the merchant’s website to inspect current terms and completes the purchase there. A payment-app context therefore does not establish cart transfer, delegated payment or machine-authorised checkout. 5

From bounded pilot to optional personalisation

The first version was a real-user but time-limited Black Week pilot: Vipps announced it on 26 November 2025, saying it had gone live on 24 November and was planned for removal on 2 December. It used open PriceRunner and FINN Nybrukt data, including used and refurbished alternatives, with no access to users’ Vipps data and no stored conversations. 9

A later Vipps engineering retrospective describes a four-to-five-person part-time team, an internal Model Context Protocol proof of concept, manual testing, automated self-evaluation, Mixpanel instrumentation and an anonymous AI-judge dashboard. These are vendor-reported process claims, not a published benchmark, independent audit or controlled trial. 10

The permanent service’s disclosed scope is broader: it may use profile data, payment and merchant-purchase history, prior chats, club benefits and chat information to personalise suggestions. Training or improvement uses require separate, withdrawable consent; users can also turn Shoppe off. This supports an evolution from a privacy-minimised experiment to an optionally more personal service, not a claim that the pilot statement was false. 8 11

Signals, not demonstrated ROI

Vipps reports that the pilot exceeded its usage baseline, outperformed similar launches, generated strong merchant click-through, received no customer-centre complaints, attracted merchant outreach and required doubled PTU capacity at peak. It also reports guerrilla tests with no negative reaction to adding AI. 10

Those observations lack the denominators needed for a commercial conclusion: no user count, baseline definition, comparison set, click or impression numerator, merchant count, order conversion, sales, complaint denominator, sample size or test protocol is published. “Millions” describes potential app-scale availability, not millions of Shoppe users. FinansWatch confirmed the pilot only at headline level; no accessible independent results report was retained. 10 13

An illustrative supplier lesson

For a retailer, distributor or manufacturer, the transferable application is illustrative, not a reported Vipps deployment. Make product identifiers, variants, attributes, availability, delivery fields, update timestamps and authoritative landing pages machine-readable. Keep discovery separate from a sale: the merchant’s current offer and terms remain authoritative, and a human remains responsible for the final purchase.

Test one governed product family with real buyer queries, near-duplicates and ambiguous alternatives. Measure correct retrieval, omission and wrong-item rates, click-through with impressions as denominator, and merchant-confirmed orders where available. Do not call a click a sale or a permanent feature an adoption result. The machine still does not buy.

Sources and next reading

• Vipps for iPhone — Official iOS client; Shoppe eligibility remains Norway and age 18-plus, and the link does not prove feature enablement.
• Vipps on Google Play — Official Android listing; no installation or sign-in was performed, and the listing does not name Shoppe.
• Shopping assistant eligibility — Official access-scope documentation, not a separate app entry point.

Explore OfferSignal · Focused Business & AI Workshop · All research

References