Here’s a mistake we see constantly. A business owner opens three tabs before a Monday meeting: GA4 shows one conversion count for last month, the Meta dashboard claims credit for nearly all of those sales by itself, and the bank deposit implies fewer orders than either tab. Someone asks which one is right. Nobody knows. So the team quietly stops using any of them, and marketing decisions go back to gut feel.
The mistake is treating the disagreement as a bug to fix. Those three numbers will never match, because they are not measuring the same thing. The fix is deciding, in writing, which number answers which question, and then building a report your team actually reads. That is a process problem more than a tooling problem, and you can solve most of it yourself in an afternoon.
Why don’t GA4, ad platforms, and my bank ever agree?
They disagree because each system counts different events, over different time windows, with different rules for who gets credit. GA4 models sessions and applies its own attribution. Ad platforms count anyone who saw or clicked an ad and later bought, on their own lookback window. Your bank counts settled cash. Three definitions, three numbers, every single month.
Break it down and the gaps stop being mysterious:
- Attribution windows. Meta might claim a purchase seven days after a click and one day after a view. GA4 might credit the same purchase to organic search because that was the last click. Both platforms report the sale. Your bank saw it once.
- Ad blockers and consent. A meaningful slice of visitors block analytics scripts or decline cookies. GA4 never sees them. The order system does. GA4 will almost always undercount real sales, and that is normal.
- Timing. GA4 records when the purchase event fired. Your processor records when the money settled, which can be days later, minus refunds and chargebacks that analytics never subtracts.
- Double counting by design. Every ad platform grades its own homework. Add up the conversions Meta, Google Ads, and TikTok each claim and you will routinely get more sales than you actually had. Each one is answering “did people we touched buy?” and a single buyer can be touched by all three.
Once you accept that each system answers a different question, you can stop chasing a match that will never come and start assigning jobs.
What should be your source of truth?
For revenue and order counts, the source of truth is your order system: Shopify, your POS, your CRM, or your invoicing tool. Money is the metric nobody argues with. GA4 and the ad platforms then become directional tools for understanding where that revenue came from, while the order system keeps score.
In practice, that means a simple hierarchy:
- Revenue, orders, refunds: the order system, reconciled against the bank monthly.
- Where demand comes from: GA4 (or a privacy-friendly alternative), read as trends and proportions rather than absolute truth. If organic traffic is up by a third and paid is flat, that story holds even if the raw counts are undercounted.
- Whether a specific channel is working: the platform’s own dashboard, compared against itself over time. Meta’s numbers this month versus Meta’s numbers last month is a fair comparison. Meta versus GA4 is not.
Write this hierarchy down. One page. When someone asks “which number is right?” the answer becomes “right for what?” followed by a pointer to the page. We build this same hierarchy for clients as part of our analytics and reporting service, and the written version does more good than any dashboard we ship, because it ends the weekly argument.
How do you define metrics so people stop arguing about them?
Pick a handful of metrics, write a one-line definition for each that names the system it comes from, and never let two metrics share a name. Most reporting fights are two people using the word “conversions” to mean different things, and a definitions page ends the fight.
A definition worth writing looks like this: “Revenue = net sales from Shopify, after refunds, calendar month.” Not “revenue = what we made.” The definition names the source, the treatment of refunds, and the time window. Do the same for “lead” (a form fill? a qualified phone call? a booked appointment?), for “conversion rate” (of sessions? of users? on which pages?), and for “return on ad spend” (whose attribution?).
Two rules keep the list honest:
- If a metric cannot be pulled the same way twice by two different people, it is not defined yet.
- If nobody would change a decision based on a metric, drop it from the report. Track it privately if you like, but the report is for decisions.
This is also where you fix GA4 itself. Most GA4 setups we audit mark far too many events as key events, so “conversions” includes newsletter signups, outbound clicks, and scroll milestones alongside actual purchases. Trim key events to the actions that map to money, and the GA4 number starts meaning something again. If your order data lives in one tool and your marketing data in another and nothing connects them, that is a plumbing problem worth solving too. We wrote about it separately in why your tools don’t talk to each other.
What does a monthly report people actually read look like?
One page, five to eight numbers, each compared to the prior period and to the same month last year, with two or three sentences of plain-language commentary about what changed and what you will do about it. If it takes longer than five minutes to read, it will stop being read by March.
The structure we use:
- The money row first. Revenue or leads from the source of truth, versus last month and last year. Seasonality is real for almost every business, and the year-over-year comparison is the one that catches it. A Tampa restaurant comparing March to February is mostly measuring Gasparilla hangover and spring tourists, not marketing.
- Two or three demand signals. Organic traffic trend, paid spend and its platform-reported return, email revenue. Directional, labeled with their source.
- Commentary that admits uncertainty. “Paid looks stronger, but a third of it is view-through credit, so we are watching one more month before scaling” builds more trust than a wall of green arrows. The fastest way to make a team distrust a report is to make every month look like a win.
- One decision. Each report should end with something you will start, stop, or keep doing. A report that never changes a decision is a ritual, and rituals get skipped.
Whoever assembles it should pull from the definitions page, not from memory. The first month is tedious. By the third month it takes an hour, and you will have something most businesses never get: a set of numbers the whole team argues from instead of about.
Common questions
Should I just ignore GA4 since it undercounts?
No. Undercounted data that is consistently undercounted is still useful for trends, proportions, and comparisons over time. Ignore GA4 for “how much did we sell” and use it for “where is attention coming from and what do people do on the site.”
Which ad platform’s conversion numbers should I believe?
Believe each platform about its own trend, not its absolute total. If platform-claimed conversions rise while revenue in your order system stays flat, the platform is taking credit for sales you would have had anyway. That comparison, platform claims versus real revenue, is the honest test.
Do I need a data warehouse or a BI tool for this?
Almost certainly not at small-business scale. A spreadsheet fed from your order system, a definitions page, and a disciplined monthly habit beat an expensive dashboard nobody opens. Add tooling when the manual pull genuinely takes too long, and start with connecting the systems you already have.
How long until the numbers feel trustworthy?
About three monthly cycles. The first report surfaces the disagreements, the second confirms your definitions hold, and the third gives you your first clean month-over-month comparison. Trust comes from consistency, not from any one tool.