Feedoptimise adds statistical significance engine to product feed A/B testing

Feedoptimise has upgraded its A/B Testing Suite with a decision engine that uses retailers’ own item-level performance data to judge whether a product feed change truly won. The release is designed to cut false positives, show results by SKU and require tests to clear data, timing and guardrail checks before a winner is applied.

Why it matters: - Feed teams often act on small performance gaps that may be noise, not signal. - The new decision engine is built to reduce bad rollouts by requiring statistical significance before a feed change is declared a winner. - The suite also helps retailers and agencies test AI-generated content before they scale it across a catalog.

What happened: - Feedoptimise released a major upgrade to its A/B Testing Suite for product feed testing. - The platform now connects experiments to a retailer’s own item-level reporting and returns verdicts per report and per SKU. - The upgrade is available now to Feedoptimise customers. - Retailers and agencies can start a free trial or book a demo at product A/B testing.

The details: - The suite can run duplicate and rotated experiments on any product feed field, including titles, descriptions, images, categories and product attributes. - It reads item-level data from Google Ads, Google Analytics, Google Merchant Center, Facebook, Shopify, WooCommerce, Magento, Centra or a custom URL-based report. - Users map the report columns for impressions, clicks, sales, cost and revenue, and the decision engine evaluates the test from that data. - The system does not require separate tracking setup or manual spreadsheet comparisons. - Feedoptimise reruns the math against daily data thousands of times to estimate uncertainty and returns a range instead of a single point estimate. - A result is a win only when the full range stays above zero. - If the range crosses zero, the suite marks the test inconclusive even if the headline lift looks positive. - A timeline shows how the range changes each day, which helps teams avoid stopping early on one strong day. - The default test thresholds require at least 14 days of runtime. - The default thresholds also require minimum impressions, clicks and sales for each version separately, not combined. - The improvement must also clear a minimum lift threshold and remain stable over time. - Users can change every threshold to fit their catalog and traffic. - Feedoptimise holds back the last few days of data so late-arriving sales still count. - The suite tracks guardrail metrics alongside the primary metric, including return on ad spend, conversion rate and cost per click. - If a title increases clicks but reduces sales, the system can return a Guardrail failed verdict. - The Product decisions tab scores products individually and shows which threshold determined each status. - The suite returns nine verdicts, including win, loss, waiting for exposure, waiting for conversion lag, insufficient data and guardrail failed. - Most tests do not produce a clear winner, and the platform reports that outcome rather than forcing one. - Winning values are saved to a static sheet outside the live feed, and users review a preview before applying the rollout. - Rollouts require separate approval, and users can review or undo them later. - Feedoptimise customers already use AI inside the platform to write titles and descriptions, fill missing attributes and edit product images. - The same system lets users test AI-generated enrichment against current published content before expanding it across the catalog.

Between the lines: - The release pushes feed testing closer to a decision system, not just an analytics layer. - That matters because catalog changes can look good in aggregate while helping some SKUs and hurting others. - The per-SKU verdicts and guardrails suggest Feedoptimise is aiming at safer, more granular merchandising decisions. - The AI enrichment workflow also addresses a common problem in generative tools: content is easy to create, but proving commercial lift is harder.

What's next: - Feedoptimise is positioning the upgraded suite as a default workflow for testing, approving and rolling out feed changes. - Retailers and agencies can use the free trial or demo to evaluate the system on their own reporting stack. - As more teams test AI-generated feed content, the same decision engine could become a standard check before catalog-wide deployment. - More information is available on Feedoptimise.