AI Conversion Advisor for Magento 2: User Guide

AI Conversion Advisor shows where visitors leave your shop between the product list and the payment, compared with your own store: other products, other lists, other devices, countries, browsers and payment methods. For each leak it lists what in your shop's own data probably explains it, how many orders a month it could bring, and what to do. Every action gets an ID and a date, and the module measures afterwards whether it worked. Nothing changes in your shop until an admin applies it.

The paid add-on AI Conversion Advisor Pro ranks the recommendations by money and adds forecasts, two-dimension views, AI drafts of product information, search matching by meaning, hidden gems, comparison settings and Analyse Now (see section 11).

• Changelog: Release notes · Questions: FAQ

1. Requirements

2. Installation

With Composer. Use the Composer keys from your account on commercegears.com (see https://commercegears.com/composer-access/):

composer config repositories.commercegears composer https://repo.commercegears.com
composer config --auth http-basic.repo.commercegears.com PUBLIC_KEY PRIVATE_KEY
composer require commercegears/module-ai-advisor-conversion
bin/magento setup:upgrade
bin/magento setup:di:compile            # production mode only
bin/magento setup:static-content:deploy # production mode only
bin/magento cache:flush

Connect Google (once for all AI Advisors; skip this if AI SEO Advisor or AI Ads Advisor is connected already): create a read-only Google service account, give it the Viewer role on your GA4 property (GA4 Admin > Property access management), and paste its JSON key in Stores > Configuration > CommerceGears > AI Advisors: Shared Settings. Click Test Connection, switch the scope to your website and choose the GA4 Property. The AI Advisor Core user guide shows the steps with screenshots. The key is stored encrypted and never shown again.

Then open Stores > Configuration > CommerceGears > AI Conversion Advisor, switch the scope to your website and set General > Enabled = Yes.

3. GA4 events and the Setup Check

The module reads GA4's recommended ecommerce events. CommerceGears GA4 & GTM sends all of them on Luma and Hyvä; any other set-up that sends the same events works too.

Own event names: if your GA4 set-up uses its own names for a step (e.g. checkout_step_2 for the shipping step), map them in Settings > Data > Event Names, one per line: checkout_step_2 = add_shipping_info.

Setup Check (CommerceGears > AI Conversion Advisor > Setup Check) lists what was checked, the result and what to do, sorted into Problems, To check, Loading, OK and Notes. It is checked every night from the data loaded so far (nothing is sent to Google when you open the page); Check Now runs it again.

• Set-up: the advisor is switched on for the website; a GA4 property is chosen.
• Data: days loaded per data set and the progress of the first load; GA4 quota used today.
• GA4 events: each funnel event of the last 28 days with its count, or "Not sent", and the screens a missing event switches off. Implausible counts are flagged, for example more begin_checkout than add_to_cart. When events are missing, a line says how to send them.
• Data quality: the share of GA4 item ids that are products of this store (by SKU or product ID, Item ID in GA4) with the 10 most viewed ids that are not, purchases with their products, the share of product lists with a name, the share of shipping steps with a shipping method, GA4's currency and time zone against the website's, days where GA4 hid small numbers or grouped rows into "(other)", visits of other hostnames that were left out, and whether GA4 can report every data set for your property. Products not sold online lists product attributes that look like "enquiry only" or "price on request": yes/no attributes whose products sold nothing online in the last year although they had a noticeable share of the product page views (e.g. disallow_addtocart = 1). Add them to Not Sold Online When (AI Advisors: Shared Settings) if they are, so their pages are judged by enquiries, not by add-to-carts.
• Tracking health: GA4 purchases against Magento orders over the last 28 days: "GA4 records 71% of your orders". Under 50 %, read every expected-orders number with that share in mind and fix the tracking first. More purchases than orders usually means the purchase event fires twice (reloads of the success page).

The same rows are printed by bin/magento commercegears:conversion:setup-check, which ends with exit code 1 when the advisor is blocked (switched off, no GA4 property), so it can run in a deployment check.

4. Data and the first load

• First load: after the module is switched on, cron (every 10 minutes) loads up to 425 days of GA4 history, newest days first, within the advisors' share of the GA4 quota (AI Advisors: Shared Settings, default 25 %). On a large property this takes one to three days; product and page data load last. bin/magento commercegears:advisor:status shows the progress.
• Daily updates: every 6 hours the last 3 days are fetched again and replaced (GA4 finishes its numbers with a delay). Magento's orders and carts are counted at 02:17, the recommendations, results and alerts are worked out at 04:17.
• Orders and carts: orders are counted per day (in the website's time zone) by payment method, shipping method and country, and back 425 days. Carts are counted per day by cart value band and the furthest checkout step reached; Magento deletes old carts (by default after 30 days), so only the days Magento still has can be counted the first night. An order waiting for its payment for more than 24 hours, or cancelled before it was paid, counts as never paid.
• Site search: Magento keeps only a running total per search term, so counting starts the night after installation; GA4's search terms are loaded with the history.
• Exact visitors: GA4 cannot add up visitors of several days (one visitor on two days counts twice). The exact visitors of the last 28 and 90 days are stored once a week from installation on; other periods add up daily visitors, and the screens say so.
• Keeping data: daily product and page rows are kept for Keep Daily Product Data (Months) (default 13) and rolled up into monthly totals that are kept for good; the rest is small and kept.
• What is stored: numbers per day, product, list, page, segment value, payment method, shipping method and cart value band. No customer data (see the FAQ).

5. How comparisons work

• Your own store is the measure. A product page is compared with similar products: products of the same deepest category with at least 50 views, at a price between half and twice its own; if there are fewer than 8, the parent category; else all products with enough views (the card says which). Lists are compared with lists of the same kind, checkout steps of one segment value (e.g. phones) with the rest of the store, landing pages with the same channel's pages of the same type (product, category, CMS, home). There are no "industry benchmarks": none are reliable enough to show.
• Small numbers are pulled towards the peers. Before comparing, a rate is moved towards its peers' rate, the more the fewer views it has, so a page with 20 views and no add-to-cart does not top the list by chance.
• Period: Settings > Comparisons > Period of the Recommendations, Automatic by default: 90 days while the store had fewer than 300 orders in the last 90 days, else 28 days.
• Minimum volumes (in the period):

• Counts first. The headline is a number you can check in your own data: "≈14 more add-to-carts a month if it converted like similar products", and the expected extra orders as "up to ≈2.1 orders a month", because the peers' rate is what similar pages reach, not a promise. Below 0.1 expected orders a month the card says "too few visits to expect an extra sale". The base module ranks every card by expected orders and shows no money; Pro ranks by money (section 11).

6. The screens

Everything is under CommerceGears > AI Conversion Advisor, with tabs at the top: Dashboard, Checkout, Products, Lists, Site Search, Landing Pages, Opportunities, Processes, Forecast vs Result (Pro), Setup Check. Every report has the same filter bar in one row: Website, Period (last 7, 28 or 90 days, last 12 or 16 months, or Custom with From and To), Compare With (previous period, same period last year or no comparison) and, where it applies, a Segment. Click Apply at the end of the row. The line under the filter bar names the period and says until which day GA4 data goes ("GA4 data until ...").

Custom periods of up to 90 days are read live from the daily data (on large stores this may take a while); longer custom periods are read in whole months from the monthly totals, as said under the filter bar.

Dashboard

• Funnel: visits, product views, add to cart, checkout, shipping, payment and purchase, with the rate from one step to the next and its change against the comparison period.
• Conversion Over Time: one chart with metric chips (conversion rate, add-to-cart rate, checkout completion, orders in Magento, GA4 coverage). Flags mark done tasks, applied settings, notes and the other advisors' changes (for example AI SEO Advisor); hover the chart for the numbers of each day.
• Alerts: a checkout step that broke for one browser or device, and a drop in tracking (section 9).
• Top Recommendations, the Latest Processes with their results, and a Setup Check badge that says how many points need a look.

Checkout

• Checkout Steps by segment: the steps cart page to checkout, checkout to shipping step, shipping to payment step and payment to purchase for one dimension at a time (device, country, browser, channel; with Pro also new or returning and two dimensions), each value with its visitors and step rates, the values clearly below the rest of the store marked.
• Payment Methods (Magento orders): orders and never paid orders per method.
• Shipping Methods and Countries (Magento orders): orders and shipping as a share of the order value.
• Carts by Value: carts with products by their value and how many became an order (bands in Settings > Data > Cart Value Bands).
• The recommendations of this area: checkout steps, shipping cost shock, payment failures, big carts abandoned, browser or device breakage and checkout settings.

Products

• Grid of the most viewed products of the period: views, view to cart, similar products' rate, cart to purchase, orders lost a month (up to), likely causes and stock. Filters: category, cause, sort (orders lost, views, lowest view to cart) and "With a recommendation".
• Product detail: views and view to cart over time with the flags of processes, the product next to the median of similar products (price, images, description length, reviews), its recommendations and its processes, and a link to edit the product.

Lists

• Category pages and other product lists (search results, related products, up-sells, cross-sells) with products seen, clicked, click rate and the median of lists of the same kind.
• Lists to categories: CommerceGears GA4 & GTM sends category pages as category_<id>, which are matched automatically; other names are matched by category name, and the rest can be matched by hand on this screen (category id, Save).
• Products seen in lists without a name are not compared; the Setup Check says why.

Site Search

• Search terms with their searches (Magento and GA4), results, the suggestion and whether the term is in a synonym group. Only terms searched 3 or more times are shown; "Only without results" filters the list.
• For a term without results the suggestion is another spelling of your catalogue's words, the weekly AI suggestion, or "not in your catalogue". A suggested search synonym group can be proposed, applied and reverted (section 7).
• Ask AI Now sends the searches without results and a sample of your catalogue to the AI at once instead of waiting for Monday (permission AI Suggestions).

Landing Pages

• Landing Pages by Channel: page, type, channel, sessions, engaged sessions, add-to-carts per session and purchases; a channel filter. Sessions that landed on pages that are not a known product, category or CMS page are counted in one line.
• Pages with Visits That Cannot Sell: product pages that get visits while the product is out of stock, disabled, not visible or not in the website.
• With AI SEO Advisor installed, the landing pages come from its GA4 data (nothing is loaded twice).

Opportunities

Every recommendation as a card: why it was flagged, the numbers, the likely causes compared with similar products, what to do and the actions Propose Setting (where a setting can be changed), Mark as Done, Add as Task and Dismiss; Pro adds Draft Product Information on product cards with a short description among the causes. Cards are ranked by expected orders a month (Pro: by money), at most 50 of a kind; the navigation filters by kind. A card whose target already has an open process, or one done in the last 35 days, shows "Waiting for results". Section 8 lists the kinds.

Processes

All processes with ID, kind, status, the day it was done or applied, its result and who made it; filters by status (To do / proposed, Done / applied, Reverted, Dismissed, Notes) and kind. Add a Note puts a dated note on the charts (a campaign, a price change, a new payment method). Open a process to see what and why, its changes, result and log, and to mark it as done, apply, revert or dismiss it.

Setup Check

Section 3.

Pro places

Forecast vs Result (tab and menu item) and AI Analysis Reports (menu item; dashboard button "Analyse Now") are Pro. Without Pro they show a preview with sample data (section 11).

7. Processes: tasks, setting changes, Mark as Done, revert

A process is one action with an ID such as CRO-20261105-001 and a date. Once done or applied, it appears as a flag on the charts.

• Task: something you do outside the module (add images, restock, test the checkout on an iPhone, check the payment provider's dashboard). Mark as Done records today and measures the effect from now; Add as Task keeps it as "To do" until you mark it as done. A task to do is never measured.
• Setting change: a change the module can make and undo for you: a search synonym group (Site Search) and the checkout settings of section 8 (guest checkout, required telephone or company field, persistent cart; at website scope). Propose Setting shows the current value and the proposed one; an admin with the permission Approve, Apply and Revert Changes applies it (the old value is kept), and Revert restores the old value, or the inherited one when the scope had none. A revert is refused when the setting was changed elsewhere since it was applied (the message names the value it has now), so a later change is not overwritten by accident; Revert Anyway then puts the old value back over it.
• Note: a dated note on the charts, without a result.
• AI draft (Pro): a draft of missing product information, reviewed, applied and reverted the same way.

Statuses: tasks To do, Done, Dismissed; setting changes and drafts Proposed, Applied, Partly applied, Reverted, Dismissed. Every step is logged with the admin user and time.

Results

For each done or applied process the module compares the figure it was about (the product's view to cart rate, the list's click rate, the step rate of the segment, the searches without results of the term) in the 28 days before with the 28 days after a settle week, next to the change of the rest of the store over the same days. The result is Better or Worse at 10 % or more against the rest of the store (searches without results: 50 %), otherwise No clear change; Too little data below the minimums, Measuring with the days to go while the 28 days are not complete. Small numbers give large percentages: read the result with the numbers.

8. The recommendations

Cards at the same checkout step share that step's budget: together they never claim more visitors than leave the step (times the recovery factor); the strongest card names the others. Their figures add up only when the budget held them back or when they are different visitors (desktop and phones); otherwise only the largest counts. When all open cards together would bring more than 30 % of your successful orders a month, the Dashboard calls their figures an upper bound and shows no total.

Checkout settings need four earlier stored 28-day periods before they have a target, and they are measured in your main market (the country with the most purchases): if the step is as good there as in its best 28 days, the drop comes from other visitors, not from the setting. A product out of stock without its own sales history is valued at the purchase rate of products in its price range. A buy now, pay later payment method is compared with your other such methods, or with an expected decline rate (35 %).

A card you started a process from, or dismissed, stays when tonight's check no longer finds it: it shows "Last estimate, no longer found" and why. Applying the process or marking it as done then asks you to confirm.

Each card names its numbers and its comparison. A card is never a claim that the cause is proven: the causes are shown "compared with similar products", as evidence to check.

When a kind has no cards, Opportunities says why (as its empty state, or under "Why some kinds have no cards"), and so does the Dashboard under Top Recommendations: the shop's best figure against what a comparison needs ("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"), what was checked when the volume was enough ("Products compared with similar products: 120; none clearly weaker"), or why a check does not apply (no shipping step). A shop with fewer than about 30 orders a month also sees its orders in the period: cards appear as traffic grows, while tracking, setting and pages-that-cannot-sell checks run every night. With AI Conversion Advisor Pro, minimums you changed in its comparison settings are marked "your setting"; lower minimums give more cards, but more of them are chance.

9. Alerts

• Browser or device breakage: a checkout step falls by half or more for one browser or device on each of the last 2 days while the others stay within 20 %, for example after a JavaScript error in Safari. The alert names the step, the browser or device and the rates.
• Tracking drop: GA4 suddenly records far fewer of your orders than before.

Alerts show on the dashboard and as an admin notification; with Settings > Alerts > E-mail Recipients set, they are also e-mailed. The same alert is sent once a week at most, and each can be switched off. GA4 data of the same day is incomplete, so alerts work on whole days.

10. Settings, permissions and command line

Stores > Configuration > CommerceGears > AI Conversion Advisor (website scope unless noted):

Shared settings in AI Advisors: Shared Settings: the Google connection and GA4 property, Share of the GA4 Quota for the Advisors (%), GA4: Also Count These Hostnames (for a test copy that should count the live shop's visits), Instructions for the AI, and "What Should Count When Ranking?" (money settings used by Pro).

Permissions (System > Permissions > User Roles > Role Resources > CommerceGears > AI Conversion Advisor): View Conversion Data, Create Tasks and Drafts, Approve, Apply and Revert Changes, AI Suggestions, Settings.

Command line:

Cron: commercegears_aiadv_fetch every 10 minutes (AI Advisor Core), commercegears_aiconv_store_daily 02:17, commercegears_aiconv_nightly 04:17, commercegears_aiconv_weekly Monday 05:47 (AI search suggestions).

11. AI Conversion Advisor Pro

Without Pro, every Pro feature is visible where it would be, with a Pro badge: the Forecast vs Result tab and menu item, AI Analysis Reports (menu item; dashboard button "Analyse Now"), the money ranking on Opportunities and the dashboard, Product Value on Products and the product detail, the forecast on a process, the two-dimension and new or returning segments on Checkout and products by device, Draft Product Information on product cards, matching by meaning on Site Search, the hidden gems kind and the Pro settings (read-only). 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 your own data. With Pro installed and licensed, the real screens replace the previews. See the Pro user guide for what each feature does.

What Pro would add. The dashboard has one panel that puts your own numbers next to what Pro would show, for example "Your checkout: 1,240 visitors started it in the last 28 days, and 412 of them left at the shipping step. Pro would show what that costs in orders and money a month, and which device and country together cause it." These are counts only; Pro's money figures and forecasts are worked out only once Pro is installed. Hide for 30 Days hides the panel for you only. Compare Base and Pro shows the full table, the same screens now and with Pro (example data of a made-up shop, marked "Example") and a few illustrative examples. Your developer or agency can switch this information off, or point its links to their own page, under Stores > Configuration > CommerceGears > AI Advisors: Shared Settings > Pro Information.

12. Limitations

• Needs GA4 ecommerce events; without them only the Setup Check is shown.
• Step rates are "visitors who reached the step" (users per event), not strict sequences.
• GA4 records fewer purchases than Magento has orders, may hide small numbers (thresholding) and groups rare rows into "(other)" on large properties; the Setup Check shows when and how much.
• Comparisons need volume; small shops get fewer cards, never guesses, and the screens say which minimum is missing.
• No A/B tests, heatmaps, session recordings or per-visitor predictions, and nothing is changed without an admin applying it.
• One GA4 property per website.