B&Q and Upp.ai: making a fast-growing marketplace visible without confusing visibility with sales

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

A marketplace can list a million products and still struggle to give new or low-history offers a fair chance of being seen. B&Q’s case with Upp.ai addresses that problem inside paid media: how to move from Standard Shopping towards Google Performance Max (PMax), keep a rapidly expanding catalogue usable for vendor-funded campaigns, and avoid letting historical performance concentrate exposure on already-proven products.

What B&Q and Upp.ai changed

B&Q is a UK DIY retailer whose marketplace launched and expanded in 2022. A first-person B&Q account published by Google says the marketplace offered more than 1.2 million products and that PMax was tested to extend product coverage while applying artificial intelligence to bids and creative. Separately, Upp.ai says B&Q moved 100% of Shopping spend to PMax with Upp.ai underneath, while the catalogue grew from 250,000 to more than one million SKUs. These accounts are best read as different points in a transition: neither provides Upp’s exact go-live date or an outcome measurement window. 1 4

The task was advertising and catalogue-surface discoverability, not B&Q’s on-site search, supplier procurement or payment. Google describes PMax as a goal-based campaign type that uses advertiser inputs such as feeds, assets, conversion goals, budgets and audience signals with Google’s own AI to choose delivery across surfaces including Search, YouTube, Display, Discover, Gmail and Maps. Upp.ai is therefore an overlay around Google’s delivery system, not a replacement for it. 5 8

A hybrid of analysis and automated action

Upp’s general product description claims connections to Google Ads, Merchant Center, Product Data and order APIs, with wider inputs such as stock and order history refreshed every 15 minutes. It describes SKU-level analysis, seasonality and performance insights, followed by automatic product or campaign-structure changes and budget allocation. The B&Q case and award commentary add marketplace and pricing insight, live performance data and product prioritisation. These are provider-described capabilities; the public record does not establish that every connector, refresh interval or action was enabled in B&Q’s account. 1 5 6

The resulting pattern is a hybrid: AI-assisted analysis with claimed automated execution within user-set success parameters and thresholds. The award account says rollout was phased to preserve continuity and trust across trading and merchandising. Yet B&Q’s actual permissions, approval queue, financial caps, exclusions, rollback process and audit controls are not public. This is constrained paid-media automation, not authority to alter product facts, accept supplier terms, buy on a customer’s behalf or pay anyone. 5 6 8

Reported outcomes, with the denominators missing

Upp reports a 40% increase in “marketplace product visibility” while the catalogue grew from 250,000 to over one million SKUs, plus a 16% improvement in “marketplace ROAS.” A July 2025 trade-forum article and a BRC Associate Insight repeat the narrative; the BRC page identifies Upp.ai as the provider. These remain attributed vendor/customer-case claims, not independently measured results. The sources do not define visibility, state its numerator, give a pre-period or comparator, disclose starting ROAS, attribution, conversion-value or margin treatment, or show a holdout. The figures therefore cannot establish incremental sales or profit caused by Upp. 1 2 3

The BRC account also describes “significant growth” in vendor-funded marketing budgets, but gives no amount, vendor count, share, period or control. Performance Marketing World describes the case as replacing manual rules and fragmented workflows with a product-level adaptive model and reports qualitative gains; its award recognition is not a causal experiment. The awards organiser separately confirms Upp.ai for B&Q as the AI Innovation Award winner. 6 7 9

A separate B&Q account supplies an important counterexample. Early auto-generated PMax creative did not consistently meet brand guidelines and the expected uplift did not materialise. B&Q, Google Creative Works and Merkle then introduced reusable templates and more varied assets; B&Q reports a 14% ROAS increase in under a year. That is a different intervention with an undisclosed baseline and must not be combined with Upp’s 16% marketplace-ROAS claim. It does show that feed and auction automation still needed active creative and governance. 4

An illustrative merchant test

For a distributor or manufacturer with a long catalogue, this is an illustrative application, not a reported deployment. Start with one product family and version product ID, supplier, new-range date, stock, price, delivery, fit or compatibility, feed status and vendor-funding eligibility. Define visibility before changing campaigns: for example, the share of eligible SKUs receiving qualified impressions, alongside clicks, attributed orders and margin. Keep new and established products as separate cohorts.

The commercial owner should remain authoritative for product truth, exclusions, budget ceilings and vendor allocation. Log which system changed a product group or budget, which target and cap applied, and how an override or rollback works. Test with a predeclared cohort, geography or other holdout where viable, and report spend, conversion value, returns, margin and incremental outcomes separately. A larger visible catalogue is useful; it is not by itself a better customer outcome.

Sources and next reading

Explore OfferSignal · Focused Business & AI Workshop · All research

References