AI Conversion Advisor for Magento 2

It shows where visitors leave your shop between the product list and the payment, compared with your own other products, lists, devices, countries and payment methods, what in your shop's data probably explains it, and how many orders a month that could bring, then measures whether your fix worked.

Magento 2.4.7 to 2.4.9, a GA4 property that receives GA4's recommended ecommerce events, a read-only Google service account (shared with the other AI Advisors) and Magento cron. An AI key in CommerceGears AI Core is only needed for the weekly AI suggestions for searches without results.

No. It sends every event the module reads on Luma and Hyvä, so it is the easiest way, but any set-up that sends GA4's recommended ecommerce events works. If your set-up uses its own event names, map them in the settings. The Setup Check shows which events arrive and which screens a missing one switches off.

GA4 only sees an order when the visitor's browser sends the purchase event. Visitors who reject cookies (with Consent Mode, GA4 then only models them), ad blockers, payment pages that do not return to the shop, orders placed in the admin or by phone and test orders are missing from GA4; a success page reloaded twice can count one order twice. GA4 also counts by its own day and currency settings, so its share of your orders is usually well below 100 %. The Setup Check shows the share for your shop ("GA4 records 71% of your orders") and compares GA4's currency and time zone with your website's; the Tracking health recommendation appears when the share is under 50 %, over 110 %, or drops suddenly, for example when a deploy broke the purchase tag. Payment failures, shipping costs and cart values are read from Magento's own orders and carts, so they are complete.

GA4: visits, product views, add-to-carts, list clicks, checkout steps by device, country, browser and channel, landing pages and GA4's search terms. Magento: orders by payment method, shipping method and country, never paid orders, carts by value and checkout step, Magento's site search, stock, prices, images, descriptions, reviews and checkout settings.

On the first run the module loads the GA4 history GA4 still has, up to 425 days (GA4's default retention is 14 months), and Magento's orders of the last 425 days. Carts can only be counted as far back as Magento still keeps them (30 days by default), and Magento's own site search from the night after installation. From then on every day is kept: daily product and page rows for 13 months by default, monthly totals for good.

GA4 finishes a day's numbers with a delay, so the newest day is usually yesterday. The last 3 days are fetched again every 6 hours; Magento's orders and carts are counted every night at 02:17; the recommendations, results and alerts are worked out every night at 04:17.

It uses only the share of the GA4 quota the AI Advisors may use together (default 25 %, shared with AI SEO Advisor and AI Ads Advisor), so your own GA4 reports keep working. The first load spreads over one to three days on a large property, newest days first; after that only the last 3 days are fetched again.

No. Only totals per day, product, list, page, segment value, payment method, shipping method and cart value band. No customer ids, e-mail addresses, addresses, IP addresses or session ids. Search terms are visitors' own input: terms that look like an e-mail address, phone, card or order number, postcode or URL are dropped before storage, and a term is only shown with 3 or more searches. The module adds no storefront script and no cookies.

No. Every rate is compared with your own shop: a product with similar products (same category, similar price), a list with lists of the same kind, phones with other devices, one country with the others. "Industry benchmarks" are not reliable enough per sector and country to show.

Each comparison has a minimum volume (for a product page, 150 views in the period), and a rate is pulled towards its peers' rate before comparing, the more the fewer views it has. Smaller shops (under 300 orders in 90 days) get a 90-day period automatically. A card also needs a clear gap: for a product page at most 75 % of similar products' rate, for a checkout step at most 80 % of the rest of the store and 3 or more standard errors below.

Usually because the shop is below the minimums. Opportunities and the Dashboard list every kind without cards with the reason, e.g. "The most viewed product had 54 views in 90 days; a comparison needs 150 views and 3 expected add-to-carts at the rate of similar products", and a shop with fewer than about 30 orders a month sees its orders in the period. The tracking check, the checkout setting checks and pages that cannot sell need no comparison and still run.

No. The base module counts: "up to ≈2.1 orders a month if this page converted like similar products", and ranks the cards by those expected orders. Money ranking with your prices and margins is part of Pro.

No, and the card does not claim they are. Causes are what is different from similar products in your own data (fewer images, a shorter description, a higher price, out-of-stock variants), shown as "compared with similar products". They are the first things to check.

No. Tasks are done by you and marked as done. Setting changes (a search synonym group, guest checkout, a required telephone or company field, persistent cart) are shown with the current and the proposed value, applied only when an admin with the permission to apply changes does so, and can be reverted. Prices, payment methods and the minimum order amount are never changed.

For each done task or applied change it compares the 28 days before with the 28 days after a settle week, for the product, list, step, segment or search term it was about, next to the change of the rest of the shop over the same days. The result is "Better" or "Worse" at 10 % or more against the rest of the shop, otherwise "No clear change"; "Too little data" below the minimums and "Measuring" until the 28 days are complete. It is a rule of thumb on your own numbers, not a controlled experiment. A task still to do is never measured.

Once a week (Mondays) one small AI request suggests what the searches without results that the spelling rules could not match mean in your catalogue: other words for the same thing, other languages. Suggestions that use words not in your catalogue are dropped, and you decide what is applied. It goes through AI Core's budget and usage log; Ask AI Now on Site Search runs it at once. Everything else (the comparisons, the causes, the cards) is computed by rules, without AI.

Yes, and each module also works alone. With AI SEO Advisor installed, its GA4 landing page data is used (nothing is loaded twice), and its changes appear as flags on this dashboard's chart; this advisor's changes appear as flags on the charts of AI SEO Advisor and AI Ads Advisor.

Ranking by money with Product Value (your prices and margins), forecasts with Forecast vs Result, checkout steps by device and country or device and browser, new or returning visitors, products by device, AI drafts of missing product information, search matching by meaning, hidden gems, the comparison settings and Analyse Now.

Without Pro, its features carry a Pro badge where they would be. Opening one shows what it does and the screen itself with made-up sample data ("Sample product A") under a sample data banner; its buttons explain instead of acting. Previews never show or count your own data. With Pro installed the real screens take their place.

Analyse Now is a Pro feature (without Pro it is shown with a sample result). It sends a compact summary of the figures the advisor has already worked out to the AI and returns the findings most worth your time, each citing those figures; a finding can become a task or a setting proposal, nothing is applied by itself. Before a run you see its cost range ("about" and "at most") and confirm it. The most a run may cost is 5 USD by default (at most 7), runs per day are 1 per website by default, and AI Core's monthly budget applies; the model and effort are settings of AI Advisor Core (*AI Advisors: Shared Settings > Analyse Now*). The limit decides how deep a run goes, never whether you get a result. As an example, a run on 28 days of a demo shop's figures with Claude Sonnet 5.5 cost about 0.06 USD. Only admins with the permission Analyse Now (AI Costs) can start one.

Yes. The module works only in the Magento admin and adds nothing to the storefront, so the theme does not matter. The checkout steps are read from GA4 whatever the theme.

Yes. Each website has its own GA4 property, settings, data and recommendations. The store views of a website share its GA4 property.

No. It finds where and for whom the shop loses visitors and measures your changes over time. A/B tests need traffic splitting and volumes most shops do not have, and heatmaps need a storefront script and personal data; both are out of scope on purpose.

Magento Open Source, Mage-OS or Adobe Commerce 2.4.7 to 2.4.9 with PHP 8.2 to 8.5, including Adobe Commerce content staging. Magento's cron must run.