Top findings: what Commander found and how it ranks it
Top findings is the ranked feed of what Commander detected in your store data. It covers products, customers, customer experience (CX) and ads. Open it from Insights > Top findings. Each card carries a big number with its place in the list as you see it, so it follows your sort and filters.
Problem and intended user
The ecommerce manager (the person who runs the store) has more signals than time. Sales, returns, reviews and ad results sit in different places. The media buyer (the person who runs your paid ads) has the same issue across ad platforms. Without a ranking, a quiet but costly issue can wait.
Current workflow
Many merchants check Shopify admin reports, GA4, Meta Ads Manager, Google Ads and their review app one after another, then rank the numbers in a spreadsheet by judgment. Some hand this work to an agency.
Prerequisites
Findings come from your imported store data and from the sources you connect. Each card shows its sources as icons in its top right corner. GA4 adds findings about traffic, conversion and revenue changes. It also lets Commander check whether GA4 records your orders. A GA4 change that is not yet clear enough to act on shows with an Early signal badge. A store with little traffic or few orders gets fewer GA4 findings. Without GA4 you still see findings from your other sources. See Google Analytics 4.
The stock, returns and product sales checks need your store connected (Shopify, or a data feed set up by Omniconvert) and your order history imported. The product sales check also needs order history that reaches back more than a year. The new customers check needs order history that covers the same 90 days a year earlier in full. The repeat orders check needs several months of order history.
The win-back and bundle checks read the results of your latest Growth Audit, so only a store with a recent Growth Audit gets them. The win-back check also needs more than a year of order history and enough customers. The ad spend check needs Meta Ads, Google Ads or both connected, with recent ad data.
The findings from your audits need a published Creative Audit or Media Audit report. A store without one gets none.
The NPS findings read the NPS survey answers that Commander receives for your store. See NPS.
For revenue estimates on ad findings, connect your ad platforms. See Meta Ads and Google Ads.
What the revenue estimate needs
Ad findings need three inputs before they show a revenue range:
- Ad spend from a connected ad platform.
- Orders attributed to the ad creative. Your landing page links must carry
utm_contenttagged with the creative ID. - Order profit. Your product costs must reach Commander.
Without all three, the finding shows no revenue range.
How Commander works
Commander checks your imported store and ad data against fixed rules and statistics, then ranks the results. An AI model writes or classifies some findings, for example those drawn from reviews, customer feedback or ad analysis.
A finding is one signal that earned a recommendation, for example a product that sells but is returned often. Nothing in the feed changes your store, your ads or your customers, so no approval step applies.
Findings about ad budget and customer value
Commander compares what your campaigns, concepts, offers and advertised products cost with what their customers are expected to be worth. Fixed rules do this, with no AI. A finding of this kind names the campaign, concept, offer or product, and its title states the verdict. Examples:
- A campaign or concept that buys expensive customers who do not come back.
- One that looks weak on the ad platform's numbers, but whose customers pay it back, so you should not cut it.
- One that loses money, with the cause named when Commander can tell.
- One that brings new customers cheaply and can take more budget. When it is too early to tell whether they are worth it, the title says so and the action is to keep the budget where it is.
- A campaign that mostly sells to returning customers, so it is retargeting, not acquisition.
- A rise in the cost per new customer (the nCAC KPI card), split into budget moves and a campaign doing worse.
- Growth bought at a loss: new customers won at a loss on the first order. The title says whether they are expected to pay it back.
- An offer, or a product with much ad spend, whose customers are expected to be worth less over time.
- A campaign, concept, angle or offer whose new customers buy again less often, or less often start with the products that bring buyers back, than customers from the rest of your account. It shows only when no other finding already covers that campaign, concept, angle or offer. It ranks low, at P3 · Low, and has no money figure.
- A campaign or concept whose ads promote products that most of its new customers do not buy first, with the product they buy most instead. It shows only when no finding about that campaign's losses, or about the advertised product, already says so. It ranks low, at P3 · Low, and has no money figure.
Two findings of this kind report good news and show a Good news badge: your profit after ad spend rose, or your cost per new customer fell. Commander holds one back when fewer of your orders can be matched to your ads than in the period before. It also holds one back when the change comes from lower ad spend and your first orders fell. A fall in the cost per new customer must also hold on more than one day before it shows. Good news ranks low, at P3 · Low, and has no money figure. When one run of Commander can create only some of its findings, it keeps every problem before any good news. In a list, good news can still sit above a problem that ranks P3 · Low with a lower rating. It goes when the change no longer holds.
A finding about a decision, such as raising a budget, often has no money figure. On the Media buyer and Creative strategist tabs of Home, only the findings with a money estimate or a direct budget action appear, and only the top few. This feed shows them all. See Yesterday and findings.
Findings about customers, products and orders
Commander also compares groups in your order, customer and review data. It reports a difference only when it is clear and large enough to matter. Fixed rules and statistics do this, with no AI. AI writes none of the title, the text or the numbers.
A finding of this kind states the difference in plain words and says what was compared. Its text notes that the difference is an association, not a proven cause. Examples:
- A group of customers, defined by a product, a country, a payment method or a similar trait, that comes back more or less often than customers with a similar order history.
- Customers whose first order carried a deep discount, who reorder more or less often than other new customers.
- New customers from the latest year who reorder or spend more or less than those from the year before.
- Customers who buy from several categories and spend more than those who order as often from one.
- Many live products that sold nothing in the period.
- Many orders that include products with no brand, or no top-level category. Commander reports brand and category separately, so you may see one finding, the other, or both.
- One product, together with its offer versions, that is in a large share of your orders. Commander names the product.
- A large share of orders that were cancelled or returned. The title names a payment method or courier that holds much of that share.
- Your average star rating moving up or down between the two halves of the last 12 months. Each product is compared with itself, so a change in which products get reviewed does not count as a move. Commander reads the reviews it stores from your review platform, and the text notes that ratings also move with holidays, sales and staffing.
With Google Analytics 4 connected, Commander also compares groups of your website visits in the same way, over the last 91 days:
- Visits on one device or from one country that end in a purchase less often than the visits they are compared with: the other main device, or your largest market. A browser on one device is compared with the device's other browsers.
- Visits from one channel, compared like with like: a paid channel with your other paid channels together, and Direct or Organic Search with your other unpaid channels together. Other unpaid channels, such as Email or Referral, are only part of that comparison and get no finding of their own.
- Paid traffic, compared like with like. Visits from one paid source and medium (for example an ad platform) are compared with your other paid sources together, and all your paid visits with all your unpaid visits. Commander reads each source and medium itself to tell paid from unpaid, and leaves out campaigns whose name says they are brand campaigns ("brand", "branded" or "brnd" anywhere in the name, unless the name says "non-brand", "not brand" or "unbranded"). When a paid source buys at under half the usual rate of your other paid sources, the finding says it may be sending low-quality traffic, shows no money and suggests checking the traffic before you spend more on it.
- When a tenth or more of your paid traffic sits in an unpaid channel in Google Analytics (for example ad clicks counted as Referral), Commander shows no finding about a paid channel, and relies on the paid source comparisons above. Direct and Organic Search are still compared.
- One busy entry page that turns visits into purchases clearly better than your other busy entry pages. The text says which pages count as busy. It is a pattern to learn from on your weaker pages.
- Checkout starts that end in a purchase less often than in your store's best month and device of the year before.
- Shoppers on your main device who add a product to the cart less often than shoppers on another device. These count people who viewed a product, once per day.
These findings count visits, checkout starts and purchases as GA4 records them, and they name no cause. A finding about visits or checkout starts that do worse ends with a projection: "If visits from this channel bought 10% more often than today, that would bring about" an amount "more a month", in your store's currency. The amount is the extra, not a total: 10% of what the group brings in now, at the revenue per purchase that GA4 records: the group's own for a device or a channel, your store's for a country, a browser, a source or your checkout starts. It is the same 10% for every finding, so it does not show how large the gap is; the text does. It appears only when GA4 records revenue in your store's currency. The entry page finding and the add-to-cart finding show a dash in Impact. Their (i) tooltip says why they have no amount.
Some of these findings show money. When a group does worse than the customers or orders it was compared with, the text ends with Up to an amount a month: what that group would have brought in each month had it done as well. This applies to customers who come back less often than expected, new customers who reorder or spend less, and cancelled or returned orders. For cancelled and returned orders, the amount is the value of those orders, and not all of it is lost. Impact shows the same amount, and its (i) tooltip shows how it was worked out. The amount is an upper bound: it assumes the whole difference could be closed and that the difference is the cause. That is why Commander ranks such a finding below a measured loss of the same size. The amount appears only when your order data is in your store's currency. Commander shows no amount for cancelled and returned orders that are recorded at much less than a normal order. When several of these findings describe the same customers or orders, Total Revenue Impact counts only the largest of them.
Every other finding of this kind shows a dash in Impact, for example a group that does better, or many products that sold nothing. Its (i) tooltip says why it has no amount.
Commander opens such a finding only after it sees the same difference again on newer data, and closes it when the difference no longer holds. If you dismiss a finding or mark it Resolved, it comes back only when the difference is much larger. Commander also limits how many of these findings stay open at once, and keeps the strongest. It checks your order data on its own schedule, not on every data update. A store with little order history or few reviews, or data that Commander cannot yet rely on, can have none.
Findings from store checks
Commander also runs fixed checks on the data of every active store, on its own schedule. Fixed rules do this, with no AI, and AI computes none of the numbers. AI writes none of the titles or text either, with one exception: some GA4 findings about a change in visits, engagement, conversion or revenue, described below.
Today there are nine checks.
The GA4 check compares the purchases GA4 records with your Shopify orders. When GA4 records far fewer purchases than you had orders, you see GA4 is not recording most of your orders. When GA4 records most of your orders overall but recorded none for many days in a row lately, you see GA4 stopped recording your orders for a week or more. The text says how many purchases GA4 recorded for every 100 orders in the last 28 days. Impact shows a dash, because a tracking fault has no direct revenue figure.
The GA4 purchase-tracking check looks at the last few days of your GA4 data. When your store kept its usual orders and GA4 kept its usual visits, but GA4 recorded almost no purchases, you see GA4 may have stopped recording purchases. The card first shows on the morning after the third day with almost no purchases. The text gives how many purchases GA4 recorded and about how many it recorded a day before. Impact shows a dash, because a tracking fault has no direct revenue figure. To act on it, place a test order and check in GA4 that the purchase event still fires. While this finding shows, Commander makes no GA4 conversion or revenue findings, because their numbers cannot be trusted. It closes by itself once GA4 has recorded purchases again on two checks in a row. Conversion and revenue findings stay paused for a while after that: a GA4 finding compares two weeks, and they return once the days without purchases have left both weeks. Without that pause they would show a drop that did not happen, then a rise. If you mark this finding resolved, Commander waits a few days before it can raise it again, so GA4 has time to record purchases after your fix.
A GA4 finding about a change in visits, engagement, conversion or revenue compares the last 7 days with the 7 days before. Fixed rules decide whether the finding shows and compute its numbers. A change that last year's same weeks already explain makes no finding. When the change is clear, you see a full finding. AI writes the title and text of the clearest of these findings each day; the others get short fixed text. A finding about a rate, such as bounce rate, always has a fixed title that gives the rate in both weeks. Commander checks every number in the AI text against the finding's own numbers. When a number does not match, the finding gets the short fixed title and text instead. When the change is real but not yet clear, you see the Early signal badge, the finding ranks low, and its text is short and fixed. The badge goes when the change firms up. Commander looks at the data again every day. One GA4 finding of each kind shows at a time, the strongest first.
The stock check compares the stock Shopify reports for a product with what the product sold over the last 4 weeks. When that stock will not last the next 30 days at the same pace, the title names the product and says it may run out of stock soon. When no stock is left, it says the product is out of stock. The text gives the stock on hand, the sales per week and about how many days the stock lasts. Impact shows the sales you could miss over the next 30 days at that pace. Commander checks only products that sell regularly, and it makes no finding when the sales at stake are very small.
The returns check looks for a product that comes back far more often than your other products. Commander checks a product only when it has enough order lines, and a statistical test makes sure the difference is not chance. Commander makes no finding when the estimated amount is very small. The title names the product and says it is returned far more often than your other products. The text compares its return rate with your store's over the last 26 weeks. It also estimates what you would keep each month if you halved the extra returns, and says that this is an estimate. Impact shows a dash, because the figure is an estimate, not a measured amount.
The product sales check compares what each product sold in the last 4 complete weeks with the same weeks last year. A fixed statistical test makes sure the change is not chance. When a product's sales fell, the title says its sales fell sharply against last year. When they rose, it says its sales rose sharply against last year. The text gives the product's sales in the last 4 weeks and in the same weeks last year, and the change in percent. For a fall, Impact shows the sales the product lost each month. For a rise, Impact shows a dash. A rise usually ranks below a fall, because it asks for an optional push rather than a fix. The check compares each product with its own sales last year, not with the rest of your store. A fall across the whole store can therefore show on several products at once.
The cost check looks at how much of your sales in the last 13 weeks came from products with a cost per item in Shopify. When less than half did, you see Most of your sales come from products with no cost price. The text gives that share in percent and says that Commander estimates margins for the rest, so profit figures are less exact. When many of your costs share one cost-to-price ratio, the text also says that this looks like a default margin, not real costs. Impact shows a dash, because a data fault has no direct revenue figure. To fix it, add the cost per item in the Pricing section of each product or variant in Shopify admin, starting with your best sellers.
The new customers check counts your first-time customers in the last 90 days and in the same 90 days a year earlier. A customer counts once, on the day of their first order. When the count fell by a clear margin, and by more than normal ups and downs would explain, you see Fewer new customers than a year ago. The text gives both counts and the fall in percent. Impact shows a dash, because a change in customer count has no direct revenue figure. To act on it, check which channels brought fewer first orders, starting with paid ads and organic search.
The win-back check uses the customer groups of your latest Growth Audit. It looks for customers who used to buy often and have not ordered recently. When enough of them stopped, you see Customers who used to buy are drifting away. The text gives how many customers slipped and what winning back 1 in 10 of them is worth each month, and says that this is an estimate. Impact shows a dash, because the figure is an estimate, not a measured amount. To act on it, send these customers a win-back offer by email or SMS, starting with those who bought most often.
The ad spend check looks at your Meta ads and your Google ads that open a product page. It checks each platform on its own. For each one, it adds up what those ads cost over the last 28 days and the purchases the platform reports from them. When the ads for a product on one platform cost more than that platform says they brought in, and the spend is large enough to matter, you see a finding. The spend counts as large enough only when, at what your other ads on that platform cost per purchase, it should have brought several purchases, so a few quiet days do not raise a finding. Ads that only just started are left out until the platform has had a few days to credit their orders. The title names the product and says its ads cost more than they bring in. A product gets one finding, also when both platforms flag it. The text gives the spend and the purchase value the platform reports, and it names the platform: "Meta reports", "Google Ads reports" or "Meta and Google Ads report". A platform where the product's ads pay for themselves is not counted in these figures. Impact shows a dash. The difference is the platform's own figure, not an amount Commander measured in your orders. Before you cut the ads, check the product page, the price and the stock, and compare the platform's purchases with your own orders for that product. If a platform reports no purchase value at all for your ads in those 28 days, its purchase tracking is likely not set up, so the check skips that platform.
The bundle check uses the orders of your latest Growth Audit. It looks for two products that are bought together far more often than chance would give. The title names both products and says they sell well together. The text gives how many orders in the last 180 days had both, how many times more often that is than chance, and what a bundle could add each month. It says that this is an estimate. Impact shows a dash, because the figure is an estimate. To act on it, offer the two products as a bundle, or show the second product on the first one's page.
A finding of this kind closes by itself when it no longer holds on newer checks. The GA4 finding also closes by itself on a later check, once GA4 has recorded your orders again for a while. The cost finding closes on the next check once at least half of your sales come from products with a cost per item. The new customers finding stays open while the fall is still large, and closes once newer checks show only a small fall or none. When another finding already says that GA4 tracking is broken, Commander closes the store-check finding, so one fault shows one card. The GA4 purchase-tracking finding follows the same rule: it does not show while another open finding covers the same fault, and it closes when one opens. The GA4 purchase-tracking finding also closes by itself once GA4 has recorded purchases again on two checks in a row. A GA4 finding about a change in visits, engagement, conversion or revenue closes by itself when the change no longer shows in the data for two checks in a row. That close does not count as fixed: the number may still be where it moved to. A stock finding also closes when the product is archived, set to draft, or set in Shopify to keep selling when out of stock. When the data a check needs is missing or late, the check makes no card and closes none. No card from a check is therefore not an all-clear. Commander keeps only a few stock, returns, product sales, ad spend and bundle findings open at once, the strongest first. When an open ad finding already covers the same ad, ad set or campaign, the ad spend check makes no card of its own. The win-back and bundle findings follow your latest Growth Audit. When that audit is no longer recent, these checks make no card and close none, and once it is much older their open findings close.
Findings from your audits
When Commander publishes a Creative Audit or a Media Audit report for your store, it can add a few findings from that report to this feed. Fixed rules choose them, with no AI. AI writes none of the numbers. Each card repeats the wording and the numbers of the report, so the card and the report always agree.
A Creative Audit finding is about your own ads. It can say that one opening line carries a large share of your ads, that rivals target an audience you reach less, or that rivals lean on a kind of proof that your ads lack. Impact shows a dash, because a finding about your ad mix has no money figure. The (i) tooltip says so.
A Media Audit finding is a change to make this week on one ad platform, for example to scale a winner, to cut spend that loses money, to move spend out of a weak placement or audience, to fix a landing page, or to add new creatives to a proven line. The text gives the money at stake each week, as the report shows it, and the first thing to do. Impact shows a dash, because that money is spend to move or fix, or what your ads earn now, not a revenue gain. The (i) tooltip names the amount and says which it is. The card still ranks by that money. After a weekly update, a card can also say how the amount moved since the last report.
Select a card to open it. The link opens the report at the section that holds the finding. Use the report to see the full picture behind the card.
Commander adds only a few cards per report, the strongest first. It leaves out findings that rest on checks Commander has not yet calibrated, findings that are still in a learning phase on the platform, and findings that another finding in this feed already covers, for example a budget finding on the same ad set. It adds no card from a report that is incomplete.
A newer report replaces the cards of the older one. A card closes by itself when the newer report no longer holds its finding. A finding that the newer report still holds keeps its card and your status on it. A card that you moved to another status stays open.
Ideas from your Growth Audit
The ideas of your latest Growth Audit lead this feed, in the audit's own order. They are the same ideas as What to do next on Home, so Home and this feed are one list. Fixed rules turn each idea into a card, with no AI. The top-ranked ideas are High priority, the next ones Medium, and the rest Low. Every idea sorts above the other findings.
An idea card differs from other cards in a few places:
- In place of the priority pill it says Growth audit · idea N, with the idea's rank in the audit.
- In place of Impact it shows Worth: with what the idea adds a month by month 3 and at full effect, in your store's currency. This is the audit's own money, the same as on Home.
- Once you start the idea on Home, the card shows Started, then Live when Commander sees the change in your data. The queue buttons give way to Open, which goes back to where the idea started: the customers, Create, Ad performance or the idea in the Growth Audit.
You start, finish and discard an idea on Home. See Home: your store's numbers and what to do next. Each weekly update re-ranks the ideas. An idea that leaves the list closes by itself, unless you started it. An idea you started, finished or discarded keeps your status.
Findings about your NPS
Commander also checks your NPS survey answers once a day. Fixed rules and statistics do this, with no AI. These findings show under the CX chip, and Impact shows a dash.
- Your post-delivery NPS is below its usual level (or pre-delivery): your NPS over the last 30 days is lower than your store usually gets, by at least twice its usual swing. Commander compares it with earlier 30-day periods of your own answers and checks each moment on its own. The text gives your score now, your usual score, how many periods that usual score comes from, and the detractor shares. When your store does not yet have enough earlier periods with enough answers, Commander compares the last 30 days with the 30 days before instead, and the finding says Your post-delivery NPS fell. To act on it, open Insights > CX > NPS, read the newest low-score comments, and check what changed in that time: a carrier, a product batch, a price or offer, or the survey itself.
- Customers who bought a product rate you lower after delivery than your other customers: over the last 12 months, the buyers of that product gave you a clearly lower NPS after delivery than everyone else, and they still do in recent months. Commander counts only post-delivery answers here, because customers rate a product once they have it. When your store has enough other products with enough answers, the product must also sit far below your typical product. When it does not, Commander uses a fixed gap instead. The text gives both scores and both detractor shares, and how far the product sits below your typical product. To act on it, read the low-score comments for that product on the NPS tab (Product breakdown, which shows post-delivery answers), then check its product page, size or description, packaging and delivery.
These findings close by themselves when newer answers no longer show the gap. A product finding also closes when the product is archived or deleted. When Commander cannot read your NPS answers, it makes no finding and closes none.
While the feed is empty and a source is still preparing, a notice says what is pending.
Findings about customers moving between segments
Commander groups your customers into segments, from your best customers down. Each day it counts how many customers moved into and out of each segment. Fixed rules and statistics do this, with no AI. AI writes none of the title, the text or the numbers.
You see a finding in two cases, each judged against the segment's own usual week:
- An at-risk or lost segment grew more than usual, for example At-Risk or Churned. The title says by how much in the last 7 days, such as "At-Risk grew 14% in 7 days, more than usual". A larger rise shows with a higher priority, up to Medium. A growing at-risk segment never shows as High.
- Your VIP segment shrank more than usual. The title says by how much, such as "VIPs fell 6% in 7 days, more than usual". This finding always shows with a high priority.
The text gives the segment's size at the start and the end of the 7 days, a usual week for it, and where most of the customers came from or went to. Impact shows a dash, because a move between segments has no direct revenue figure. To act on it, open Insights > Customers > Transitions to see who moved, then build an audience in Performance > Audiences: a win-back message for a growing at-risk segment, or a reason to come back soon for your remaining VIPs. See Transitions.
Commander needs several weeks of daily counts first, so a newly connected store gets no such finding for a while. The 7 days end the day before the latest daily count, so a move shows one day later. A day when most of your customers changed segment at once, for example after a change to how segments are set up, does not count. Commander keeps only a few of these findings open at once, VIP findings first. A finding closes by itself once newer checks show the segment is back within its usual range.
Findings about ad tracking
Commander checks whether the clicks on your Meta and Google ads carry the ad id. Without it, Commander cannot credit an order to the ad that brought it. Fixed rules do this, with no AI.
When a large share of an ad platform's spend in the last 28 days runs on ads without the ad id, you see one finding for that platform. The title starts with the platform name, for example Meta:, and says what is wrong:
- a share of your ad spend runs on ads Commander cannot credit orders to;
- some ads send shoppers to a page your store does not serve;
- some Meta ads lack the standard UTM parameters. Orders do not need these for credit, because Commander credits an order by the ad id, but the parameters keep your reports in other tools consistent;
- Google: new ads will not carry an ad id, when the Google account has no tracking suffix.
The share leaves out Google Shopping and App ads. Their clicks cannot carry an ad id, so you cannot fix them with link tags. Here "link tags" means the ad id. The text says so when such ads ran.
The text lists every problem it found. Impact shows a dash, because a tracking gap hides which ads sell but does not cost sales by itself. The Proposed action is Add the ad id to the link tags, with the steps for that platform.
Commander checks again after each ad sync. The finding closes by itself when the share falls well below the level that opened it. When Commander cannot read a platform's ads, its data is out of date, or none of its ads ran in the last 28 days, the finding stays as it is. The same holds when only Google Shopping or App ads ran. If that lasts for some days, the finding closes and does not count as fixed. For Shopping and App ads alone, the note says that no ad the share applies to is running.
Findings about competitors
When a competitor you track starts to run several new ads on a messaging angle you own, you see (competitor) is moving into your "(angle)" angle. You own an angle when it makes up a large share of your own recent ads, and Commander has analysed enough of them. The text says how many recent ads the competitor runs on the angle, as a minimum: Commander counts only the competitor ads it has analysed. It also gives your own share of the angle. Fixed rules make this finding, with no AI. Impact shows a dash, because the finding has no money estimate. Commander keeps only a few of these findings open at once, the strongest first. A competitor whose website is your own store never gets this finding.
The finding closes by itself when the competitor stops launching new ads on the angle. When Commander has not refreshed that competitor's ads lately, the finding stays open as it is. To act on it, compare the competitor's ads in Competitor Synthesis, then decide whether to refresh your own ads on that angle or to move away from it.
When a competitor you track starts to launch new ads much faster than it did before, you see (competitor) is launching new ads faster than before. Commander compares the competitor's typical week of the last 4 weeks with its typical week of the 4 weeks before, by the date each ad first ran. Ads with the same headline and text count once, so a competitor that posts many copies of one ad does not look like a surge. The text gives both numbers. When Commander can read the angle of the competitor's new ads, the text also says which angle most of them use and the share, or says that the new ads split across several angles. A typical week moves only when most weeks move, so this finding appears about two weeks after a real surge starts. Commander judges a competitor only when it has followed it for long enough, counted from its first ad, and refreshed its ads every week, so a gap in the refreshes never looks like a surge. Fixed rules make this finding, with no AI, and Impact shows a dash. Commander keeps only a few of these findings open at once, the largest rise first. A competitor whose website is your own store never gets this finding. The finding closes by itself when the competitor slows down again. When Commander has not refreshed that competitor's ads lately, the finding stays open as it is. To act on it, open the competitor in the Competitor Library and look at its newest ads.
When most of the competitors you track run an ad angle, some of their ads on it have run for a long time, and your own ads use little or none of it, you see Competitors run “(angle)” ads and you run very few. The text says how many of your competitors run the angle, how many of their ads have run for 30 days or more, and your own share of it. Fixed rules make this finding, with no AI. It closes by itself when Commander no longer sees the gap. If you dismiss it, it does not come back for a while. The finding also feeds your Creative ideas: each insight-based idea there names the finding it comes from.
Findings about search demand
Commander follows search terms for your best-selling products, in your store's home country. There are two findings, and a product has one of them at a time. When searches for a term rose toward the same coming months in each of the last 2 years, you see (product): a (month) peak is ahead for searches for "(term)" in (country). The text gives the searches a month, as Google Ads counts them, and the rise in each of the two years. When interest in a term rises or drops sharply on Google Trends, you see (product): searches for "(term)" are up (change)% in (country), or are down. That text compares the last 7 days with the 28 days before. Fixed rules make both findings. An AI model only names the search terms, once for each product. Both texts say what Commander knows about the product's ads. Impact shows a dash, because the finding has no money figure. A product that already has an open finding that it sells more than its ads gets no search demand finding. See Search demand.
Ad patterns that rarely win
When one kind of hook, offer, angle or format rarely appears in your winning ads on one platform, but does appear in your weak ones, you see (Field) “(value)” rarely wins in your (platform) ads. Commander counts each creative once, even when it runs in several ad sets. It looks only at recent ads with enough impressions. On a store whose ads report sales value, a winning creative has a platform-reported return on ad spend of 3 or more, and a weak creative has at least 50 clicks and either a return on ad spend under 1 or under 0.3% clicks per impression. On a store whose ads report no sales value, a winning creative gets 2% or more clicks per impression, and a weak creative has at least 50 clicks and under 0.3% clicks per impression. Commander tests every pattern at once and shows a finding only when the gap is unlikely to be chance alone. It also checks that the pattern shows up in more than one of your campaigns. In more than one of those campaigns, creatives with the pattern must run next to creatives without it, and winning creatives must run next to ones that did not win. The gap must also hold across those campaigns taken together when each creative is compared only with creatives in the same campaign. One campaign can carry most of that gap. On a store whose ads report sales value, a pattern whose creatives report no sales at all gets no finding. The text gives the counts and says how many campaigns the pattern runs in. Fixed rules make this finding, with no AI. Impact shows a dash. Commander keeps only a few of these findings open at once, the strongest first. If a finding already says a hook works well in your ads, you see no "rarely wins" finding about hooks. The same goes for offers, angles and formats. The finding closes by itself when the gap closes, when your recent ads stop using the pattern, or when the pattern no longer passes the campaign check.
Repeat orders dropped
Commander also checks your repeat orders each week: orders from customers who bought from you before. Fixed rules do this, with no AI. It compares the last 4 weeks with your usual level over the 26 weeks before. When repeat orders fell well below that level, by a clear margin and by more than normal ups and downs would explain, you see Repeat orders dropped below their usual level. The text gives both weekly numbers and the fall in percent. It also says whether returning customers' share of your orders fell. When your order history reaches back far enough, Commander first checks the same weeks a year earlier. A dip that also happened then makes no finding, because it is likely seasonal. When there is not enough data for that check, the text says so. A finding also appears when repeat orders fell by 30% or more, even if your weekly numbers swing a lot. A finding that is already open stays open through a seasonal dip, and the text then says the same weeks last year also dipped. While a finding is open, Commander keeps comparing against your usual level from when it first appeared, so a lasting drop does not quietly become the new normal. Impact shows a dash, because a change in order count has no direct revenue figure. To act on it, check stock, price and shipping changes on the products customers reorder most, and send a reorder reminder to customers who bought before. The finding stays open while repeat orders stay low, and closes by itself once they are back near the level they had when it appeared.
Products with many low-star reviews
Commander also checks the star ratings of each product's reviews. Fixed rules do this, with no AI. When a product with enough reviews gets a large share of 1 and 2 star reviews, and it still got a review recently, you see (product): (share)% of its (count) reviews are 1 or 2 stars. This finding shows under the Products chip. It counts stars, not the tone of the review text, so its share can differ from the sentiment figures on the Sentiment tab.
The text gives how many reviews of the product have 1 or 2 stars, against the share across all your reviews, the product's average rating, its most common complaints and the date of its last review. When the product got enough reviews lately, the text also gives the share of 1 and 2 stars among those recent reviews. When the product's most common complaint is also the most common complaint across all your reviews, the text says so. Impact shows a dash, because the finding has no money estimate, and the finding always shows with a low priority.
To act on it, read the product's 1 and 2 star reviews and fix the cause: the product page, sizing, packaging or the product itself. Then watch the share in new reviews. The share in the title covers every review to date, so it falls slowly after a fix.
Commander keeps only a few of these findings open at once, the products with the most low-star reviews first. A product that a finding about its low rating and its ad spend already covers gets no second finding. Neither does a product that holds nearly all of your reviews, while a finding about your store's reviews as a whole is open. The finding closes by itself when the share no longer holds, when recent reviews are clearly better, or when the product is archived or deleted. When Commander cannot read your reviews, it makes no finding and closes none.
Order the feed
The feed shows the highest priority first. The Sort menu changes the order. Choose Priority, Impact, Confidence or Recency. Impact sorts by revenue impact. Recency puts the newest first. Findings without a value for the sort you chose come last.
Open a card
Each card shows its Proposed action without opening. When a finding has more than one recommended action, select See more context to read every step. Select See less context to close the list.
Go to the ad a finding is about
A finding about one of your ads names that ad under its title. To see its numbers:
- Select Open in Ad performance under the ad's name. The page opens in a new tab.
- Read the ad's row. The page shows the By ad view, with the ad's name in the search field and its row highlighted.
- Select the X in the field (Clear search) to see every ad again.
See Ad performance.
Act on a product finding
A finding about a product that sells more than its ads shows links under its description. Create an ad for this product and Create a dedicated ad open the Creative Wizard with the product selected. See this product's ads opens Ad performance in a new tab, By product. See Underpromoted products.
Rescan Data
Rescan Data reloads the feed from Commander. It also refreshes the ad name and the Open in Ad performance link on findings about your ads, so a finding points at the ad as it is named today. A short message gives the number of ad findings it updated, and nothing shows when none needed updating. If Commander cannot refresh them, it shows "Couldn't refresh ad findings" and reloads the feed anyway. It does not start a new analysis. New findings appear on their own as your data updates.
When findings update
The feed shows no update time. Commander updates findings on its own. If GA4 is connected, Settings > Store Status lists a Daily insights analysis row with its last run. Findings about customers, products and orders come from a separate job. The Store findings row in Store Status shows its last run. Some findings come from other jobs on other schedules, so the feed has no single refresh time. See Store Status.
Exact output
The finding card
Each card shows these parts.
| Part | What it tells you |
|---|---|
| Priority pill | The tier, with its rank and its word: P0 · Critical, P1 · High, P2 · Medium or P3 · Low. The priority score follows it, as a whole number. |
| Category tag | The kind of finding, in short form. |
| Recurring note | Plain text, N× recurring, when Commander has raised the finding more than once. |
| Channel chips | Ads, Email, Website or Other: where you act on the finding. |
| Created | How long ago Commander raised the finding, such as Created 3d ago. |
| Source icons | The data sources behind the finding. Some findings show none. Hover or focus the icons to see every source and what it provides. On a star-rating finding the tooltip also says how many reviews the finding is based on, in each half of the period and across how many products. |
| Early signal badge | Shows on a GA4 finding about a change that is not yet clear enough to act on. It sits to the left of the status badge. It goes when the change firms up. |
| Good news badge | Shows on a finding that reports something that improved, such as a fall in your cost per new customer. It sits to the left of the status badge. |
| Status badge | Where the finding is in your workflow. It sits last in the top right corner. See Work through findings. |
| Title and description | What Commander detected. |
| Ad line | Shows only on a finding about one of your ads. It names the ad as your ads account names it, then the platform and the ad ID. The campaign and the ad set follow when Commander knows them. A headline that differs from the name shows underneath. |
| Open in Ad performance | Shows with the ad line. Opens the ad on the Ad performance page in a new tab. |
| Impact: | An estimated revenue range, or one value when Commander has no range. A dash means Commander has no revenue estimate. |
| Confidence: | How sure Commander is, as a percentage. A dash means Commander has no value. |
| Proposed action box | The first recommended action. A line says how many more steps wait under See more context. It shows only when the finding has recommended actions. |
| Effort: | Low, Medium or High: how much work the first action takes. It shows only when Commander has a value. A finding with the Early signal badge has no Effort: line, because it asks for no action yet. |
| Queue button | The next step for the finding: Add to queue, then Move to To do, then Start, then Mark resolved. Each press moves the finding one step along. On an On hold finding the button reads Resume. |
| Put on hold, Back to To do, Reopen | Smaller buttons beside the queue button. Put on hold parks a To do or In progress finding. Back to To do moves an In progress or On hold finding back to To do. Reopen shows on a Resolved finding and moves it to In progress. See Work through findings. |
| Dismiss | Opens a list of reasons: Expected, Not actionable, Not relevant, Duplicate and Other. Choose one, add a note if you want, and confirm. |
| Restore | Shows on a dismissed finding. Moves it back to New. It clears the reason and the note. |
| Merged findings | A card that holds merged findings lists them under Encompasses N related findings. Select Split beside one to take it out as its own card. |
The Confidence percentage is coloured by level. Read the number, not the colour.
Impact and its tooltip
Impact is a range, not a promise. The figure can be followed by / mo estimated impact for a monthly estimate, lifetime estimated impact for an estimate over the customers' lifetime, or one-off estimated impact for a single promotion's total. A small flask icon sits beside the label. It has no click action. Hover over it to read that impact estimation is still being calibrated. Use the range to rank your work.
Hover over the (i) icon next to the range to see how Commander calculated it. With a revenue range, the tooltip shows the formula, its inputs and the result. Without one, it explains why. For an ad finding, it lists the three inputs named in Prerequisites. When the finding has its own low and high estimate, the tooltip adds a Range width note.
The expanded card
The expanded card shows All recommended actions as a numbered list. The top row holds the category, channels, sources and status.
Search and filter controls
| Control | What it does |
|---|---|
| Search agents, insights... | Matches the title, description, category and channel. |
| Status pills | Show one status at a time: All active, New, Considering, To do, In progress, On hold, Resolved or Dismissed. |
| Category | Filters by the kind of finding. The list offers the categories in your feed. |
| Source | Filters by data source, such as Shopify, Meta or GA4. The list offers the sources in your feed. |
| Confidence | Filters by how sure Commander is: 90+, 70-89, 50-69 or Below 50. |
| Channel | Filters by Ads, Email, Website or Other. |
| Sort | Orders the feed by Priority, Impact, Confidence or Recency. |
| Area: All, CX, Customers, Products, Ads | Filters by area. All shows every area and is on when you pick none. Select one or more of CX, Customers, Products and Ads to show only those areas. Select All to clear your choice. CX gathers the review and NPS findings, except the low-star review finding about one product, which shows under Products. Customers holds the other customer findings, such as retention, acquisition and segments. |
| Show more | Sits below the last card when more findings match than the page shows. Shows the next set of findings and moves focus to the first new card. |
| Select & Merge | Sits at the right end of the Area row. Starts merge mode, where you tick related findings and join them. See Work through findings. |
Within one filter, you see findings that match any choice. Across filters, a finding must match all of them. The page address keeps your search, status, filters, Area choice and sort, so a reload or Back keeps the view.
What to do with it
Top findings tells you where to look. It does not run the fix. See Work through findings to track your decision.
| What you see | Decision or action | Owner |
|---|---|---|
| A P1 · High or P0 · Critical finding with a revenue Impact range | Read the Proposed action, open See more context for the other steps when the card has them, and decide what to do first. | ecommerce manager |
| An ad finding with a Meta or Google logo | Decide whether to change the ad or the budget. You make the change in Meta Ads Manager or Google Ads, or in Create or Performance. | media buyer |
| A finding that names one of your ads | Select Open in Ad performance to check the ad's numbers. Use the ad ID to find the same ad in Meta Ads Manager or Google Ads. | media buyer |
| A recommended action that calls for a new creative | Plan the new creative in Create. | creative strategist (the person who plans and makes ad creatives) |
| A Customers finding | Decide whether a follow-up campaign is needed. You run it in your email tool. | retention marketer (the person who runs email and repeat-purchase campaigns) |
| Repeat orders dropped below their usual level | Find out why customers stopped coming back: check stock, price and shipping on the products they reorder most. Send a reorder reminder from your email tool, outside Commander. The Repurchase Signals tab lists customers who are due, and Export to CSV gives you the file. | retention marketer |
| A product that may run out of stock soon | Reorder it in your supply process, or add a pre-order or back-in-stock notice in Shopify. | ecommerce manager |
| A product that is returned far more often than your other products | Read its return reasons, then check its size guide, photos and description in Shopify. | ecommerce manager |
| A product whose sales fell sharply against last year | Check its stock, price, product page and ads for a change that explains the fall. | ecommerce manager |
| A product whose sales rose sharply against last year | Check that its stock covers the demand. Consider more ad budget for it while demand is high. | media buyer |
| Your NPS is below its usual level, or fell in the last 30 days | Read the newest low-score comments on the NPS tab and find what changed in that time. | ecommerce manager |
| Customers who bought a product rate you lower after delivery than your other customers | Read that product's low-score comments, then check its product page, packaging and delivery. | ecommerce manager |
| searches for a term are up on a product | Check that the product's ads and stock keep up. When it has no live ad, test one that uses the search term. | media buyer |
| searches for a term are down on a product | Check the cost per order of the product's ads and hold budget increases. Do not start a new ad on this signal. | media buyer |
| A competitor is moving into your angle | Compare its ads in Competitor Synthesis, then refresh your own ads on that angle or move away from it. You make the ads in Create. | creative strategist |
| A competitor is launching new ads faster than before | Read its newest ads in the Competitor Library. Note the angles and hooks it repeats, then plan one new creative in Create on an angle you do not run yet. | creative strategist |
| Competitors run an angle and you run very few | Make a creative on that angle in Create, and test it next to your best ad. | creative strategist |
| rarely wins in your ads | Make your next test creatives with a different hook, offer, angle or format, in Create. | creative strategist |
| An Impact dash | Rank the finding by its priority pill and Confidence instead, or choose Confidence in the Sort menu. | ecommerce manager |
| A notice about data being prepared | Connect the missing data source. | ecommerce manager |
Expected value and measurement
Top findings can help you decide what to work on first and see signals from several sources in one ranked list. To check whether it helps:
- Open Insights > Top findings and select the Resolved status pill to see the findings you closed.
- Open the page that holds the metric the finding is about:
- Ad finding: Performance > Ad performance. See Ad performance. Compare Gross profit after ad spend or Matched revenue.
- Product finding: Insights > Products > Product List. See Product List. Compare Return %, Revenue or Margin.
- Customer finding: Insights > Customers, tab All Customers. See All Customers. Compare Churn Risk or Days Since Order.
- Compare the metric before and after you acted, over periods of the same length.
The Impact range is an estimate. You cannot isolate one finding's effect when other changes ran at the same time.
Limitations and exceptions
- Impact and Confidence are estimates. Commander is still calibrating impact estimation.
- A recommended action can be wrong. Check it against what you know about your store. Text that an AI model writes can be wrong too.
- Ad findings without all three inputs show no revenue range.
- Some findings have no money figure by design. A budget decision can show a dash for Impact.
- Findings about customers, products and orders state an association. They show that two things go together, not that one causes the other.
- Scheduled work can pause. Commander refreshes findings on its own, and that work uses credits up to your scheduled work budget on the Billing page. When the budget is used up, the findings stay as they were. A line under the page title reads Updated on (date); paused at your limit. See Scheduled work uses credits.
- The two product-mix findings count orders from customers with an account, over the last 12 complete months. A store with less history than that is counted over all the history it has. Each order counts once, with every line it holds, and orders from staff accounts are left out. For the finding about products with no brand or category, products no longer sold do not count. The finding about one product in a large share of your orders counts every product that sold, including products you no longer sell. A store with few orders, or data that Commander cannot yet rely on, gets no finding from them. That is not an all-clear.
- Those findings can be late. They update less often than the others. A change in your store can take a while to show, and to clear.
- The stock check reads stock as Shopify last reported it. It does not check products whose stock Shopify does not track, or that Shopify keeps selling when out of stock. A store with a data feed and no stock levels gets no stock finding, because Commander has no stock figure to read. A product that shows 0 in stock without a sign that Shopify tracks it is treated the same way: Commander does not call it out of stock.
- The returns check needs returns in your order data. When Commander sees no returns for your store, the check makes no finding. That is not an all-clear.
- The product sales check compares with last year, not with the rest of your store. A fall across the whole store can show on several products, with one cause. A store with less than a year of order history, a gap in its weekly sales, or late order data gets no finding from it. That is not an all-clear.
- The cost check leaves out some products. It skips products stored without variants, and products Commander synced without their cost. When those products make up too much of your sales, the check makes no finding. That is not an all-clear. It also makes none while your Shopify catalog is still syncing.
- The new customers check counts customers its own way. Cancelled orders, orders with no value and orders from internal accounts do not count, and nor do customers who asked Shopify to erase their data. The count can therefore differ from other customer counts in Commander and in Shopify. A store with few new customers a year earlier, order history that does not cover that period, or late order data gets no finding from it. That is not an all-clear.
- The win-back and bundle checks run only on a recent Growth Audit. A store with no Growth Audit, or whose latest audit is not recent, gets no finding from them. That is not an all-clear. The money on both cards is an estimate: it assumes a share of customers come back, or a share of buyers take the bundle.
- Findings from your audits start only after a report publishes. A store without a Creative Audit or Media Audit report gets none. A report that is incomplete, or that is not yet reliable for your store, gives no card. That is not an all-clear. Read the report itself for the full list.
- A card from an audit shows no money in Impact. The report states the money in its own wording. Open the card to read it there.
- The win-back check does not use your margins. Its estimate is in sales, not profit.
- The repeat orders check counts orders its own way. It uses the same cleaning as the new customers check, and it counts orders from customers who bought before, not customers. Orders from guests, with no customer account, do not count as repeat orders. A subscription order from a returning customer counts as a repeat order. A store with too little order history, too few repeat orders, mostly guest orders, a week with no orders, or late order data gets no finding from it. That is not an all-clear. The finding names no product.
- The ad spend check uses the ad platforms' own purchase figures. Meta and Google Ads can count a purchase your orders do not show, or miss one they do. Google Ads counts every conversion you set up to count as a conversion, not only purchases. The check reads only ads that open a product page, and it does not leave out free gift products. It does not read Performance Max campaigns. A store with neither Meta Ads nor Google Ads connected, with late ad data from a connected platform, or whose currency Commander has no rule for, gets no finding from it. So does a store whose ad spend is not yet converted to the store currency, or whose ads open no product page that Commander knows. That is not an all-clear.
- Findings about customers moving between segments start only after several weeks of daily counts. A store without that history gets no finding of this kind. That is not an all-clear. These findings cover only at-risk and lost segments that grow and the VIP segment that shrinks. A day with no daily counts for your store is left out, and so is the day after it, because that day holds the moves of two days. A segment with a name of your own, which Commander cannot match to its standard segments, gets no finding. A small segment, or a small change in a large one, gets none either.
- The low-star review finding needs reviews that Commander matched to a product. A review your review app sends without a product Commander can match does not count for any product, so a store with few matched reviews gets few such findings. That is not an all-clear. The finding names only products that are still active in your catalog.
- The "rarely wins" finding needs enough winning and weak creatives on one platform. New stores and stores without measured ads get none. When you advertise on two platforms, a pattern only one platform can show (such as carousels) gets no finding. A pattern you keep in campaigns of its own, such as a campaign with only videos, gets no finding: Commander cannot tell the pattern apart from the campaign. That is not an all-clear. A pattern that few of your creatives use gets no finding.
- The bundle check names only products that are still active in your catalog. A pair with an archived or deleted product makes no card.
- The NPS findings need enough survey answers. A store with few answers, or answers on only a few days, gets no finding. That is not an all-clear. When a store does not yet have enough earlier 30-day periods with enough answers, the NPS drop finding looks for a rise in detractors against the 30 days before: then, when promoters only turn into passives, the score can fall with no finding. The product finding counts only post-delivery answers, so a store that surveys customers only before delivery gets no product finding. That is not an all-clear either. It names only products that are still active in your catalog, and only a few of them at once, the strongest first.
- A store with little traffic or few orders gets fewer GA4 findings. Commander makes no GA4 finding about a change it cannot tell from normal ups and downs. A finding about conversion or revenue needs enough orders. That is not an all-clear.
- GA4 findings about the conversion rate, visits and the number of orders need 6 full weeks of GA4 history. Commander judges the change against how much your store's own weeks go up and down, so a store with fewer than 6 usable weeks before the weeks it compares gets none of these three. A week with a missing GA4 day, or one that holds Black Friday to Cyber Monday, does not count toward the 6. While such a gap lasts, a finding that is already open stays open instead of closing by itself, and no new one of these three shows. That is not an all-clear.
- GA4 conversion and revenue findings wait until Commander can check that GA4 records your orders. A store that connected GA4 recently gets none in its first weeks, and neither does a run where that check cannot run. Findings about engagement, such as the bounce rate, and about active users still show, but visits and orders findings need the 6 weeks above. That is not an all-clear.
- Dismissing a GA4 finding hides the same change for a while. Commander holds back a later GA4 finding about the same number moving the same way, such as sessions falling again. A finding about another number still shows, and so does a full finding about the same change after you dismissed its Early signal. The dismissed change stays held back even when a finding about another number of the same kind shows in the meantime. Marking a GA4 finding resolved works the same way.
- The GA4 purchase-tracking finding is not instant. It first shows on the morning after the third day with almost no purchases. GA4 can also process a day late, and a day that is only partly processed holds the finding back instead of raising it. A store with few orders gets no finding of this kind. That is not an all-clear.
- The GA4 check needs GA4 history and fresh GA4 data. A store that connected GA4 recently, or whose GA4 data is late, gets no finding from it. A GA4 tag that records nothing at all can look like late data, so it can also show no finding.
- The search demand findings read search data for one search term and one country. The Google Trends index is relative, not a number of searches. The monthly counts are rounded, and the newest month is one to two months old. The term is generic, and the country is your store's home country. A peak in past years is no promise of one this year. A store whose home country Commander cannot tell gets no finding. That is not an all-clear.
- A dismissed or resolved finding about ad budget does not come straight back. Commander holds it back for a while. A finding with a money estimate returns only when the estimate has grown well beyond the dismissed figure. One without a money estimate returns if the condition still holds.
- A dismissed or resolved finding can come back as a new card if the problem continues.
- A finding you have not worked on can close by itself after a while, sooner for a lower priority. If the problem is still there, Commander can raise it again as a new finding. A finding you moved to another status stays open, On hold included.
- An older finding can show a revenue range with no breakdown. Its (i) tooltip then says so.
- The feed shows a limited number of findings at once. Select Show more below the last card to show the next set. A line above it gives the count showing. A new search, filter or sort starts the list again at the first set. Show more does not fetch findings beyond the ones Commander already loaded for the view.
- Category, Source, Confidence and Channel filter the findings already loaded on the page. A matching finding beyond the loaded set does not show, so the list can look empty or short.
- Only some findings name an ad. A finding about several ads, a product or a customer group has no ad line and no Open in Ad performance link.
- Ad performance shows the ad only for the selected dates. If the ad had no spend in those dates, the page says it has no row. Widen the dates in the date picker. Select Dismiss to close the notice.
When the feed is empty
| What you see | What it means |
|---|---|
| No insights match your filters | Select Clear filters. |
| No active insights right now | Nothing needs attention. |
| "We're still finishing ..." | The sentence names what is pending, for example your Google Analytics data. Wait. |
| "Connect your Google Analytics data to fill in this section." | Connect it under Settings > Integrations. |
| Failed to load insights. | Select Retry. |
No active insights right now shows only when nothing is pending. A new store can see few findings.
Demo Mode
In Demo Mode the feed shows a sample set of findings. The sample set includes one ad pattern that rarely wins. Rescan Data only shows "Not available in demo mode", and the feed does not change.