Your advertising account reports more conversions. Your attribution tool credits the campaign with sales. Your payment records show fewer paid orders than either. The next media invoice is due, and the team wants to know which campaigns to cut.
Cutting is a decision about the campaigns. Before making it, check the measurement chain the decision rests on. Advertising platforms, attribution tools and payment systems often count different things under the same name, and a campaign can look wasteful or profitable because of how a report counts, not because of what buyers did.
This guide is for a growth lead running Google or Meta campaigns for an online product or education business. It asks one practical question: what does each report actually count, and is it the sale you are paying for? It uses HYROS documentation as the worked example, and what IZZY learned connecting HYROS for a client.
Answer in 60 seconds
- Define the sale you are paying for: a qualified lead, a first paid order or retained revenue. Keep them separate.
- Trace one real purchase through the advertising account, the attribution tool, the CRM and the payment records.
- Read your attribution tool’s definitions. In HYROS, “Sales” includes refunds, “Profit” is revenue minus ad cost, and “new customer” means new since HYROS was installed.
- Check where the sale signal comes from. A thank-you page visit is not a payment you can reconcile.
- Confirm events in the destination. A tracking call that never arrives fails silently.
- Check which conversion actually drives bidding, and whether the same sale enters twice.
- Only then decide: repair a defect, wait for outcomes to mature or test a bounded change in spend.
In this article
- Define the sale you are paying for
- Trace one purchase through every report
- Know what your attribution tool counts
- What we learned connecting HYROS for a client
- Give bidding the right signal
- Then decide: repair, wait or change spend
1. Define the sale you are paying for
A cheaper form submission is useful only if the people submitting it are suitable and progress towards the outcome the business needs.
Choose one primary commercial outcome for the review. For a product sold directly online, that might be a first paid order from a customer who has not bought before. For an education product sold after a consultation, keep qualified enquiries, booked calls and payments visible as separate stages.
For a defined campaign and period, a useful descriptive measure is:
Observed media cost per attributed new paying customer = eligible media spend ÷ verified new paying customers credited to that campaign under the agreed attribution rule.
This is a media-cost measure, not a complete acquisition cost or evidence of causality. Add creative, software and management costs separately when evaluating the total investment. If no verified customers exist for the period, report the spend and that absence rather than an invented acquisition cost.
The definition has to be explicit. In an acquisition journey IZZY built, activation meant completing the journey, answering the required questions and buying the product, and a refund required a person to correct the state. A report that counts anything less is counting something else. The same discipline applies to product activation: define it as real value before you measure it.
If the offer, economics or customer definition is still unsettled, start with the paid-advertising readiness guide. If the campaign’s own choices were never really made, resolve that campaign decision debt first.
2. Trace one purchase through every report
Before comparing dashboard totals, choose one genuine paid order and reconstruct its path. This is a diagnostic exercise you can run in your own records.
| Record to inspect | Question to answer | Decision it informs |
|---|---|---|
| Customer history | Did this person buy before the tracking system was installed? | Whether the order represents new-customer acquisition |
| Payment or order | Was payment completed, and which identifier links the records? | Whether the counted outcome was a paid sale |
| Refund or cancellation | How much value remained at the review date? | Whether the revenue supports the acquisition economics |
| Attribution report | Which model, window and filters credited the sale? | Whether disagreement comes from reporting definitions |
| Platform conversion | Which action, value and identifier were received? | Whether bidding receives the intended outcome |
| Sales handoff | What happened between enquiry and payment? | Whether the next repair belongs in sales operations |
Use the same currency, time zone and reporting boundaries. Check whether a report groups outcomes by advertising interaction date or transaction date. Match the unit too: customers, orders and payment events are different counts.
Record the discrepancy precisely. “The customer bought before installation but appears as new” calls for a different action from “the payment never reached the conversion feed”.
One trace can reveal a failure. It cannot establish completeness. Extend the check across the relevant variations, including returning customers and refunded orders, before treating the integration as reliable. For an online shop, the same trace runs from product data through checkout to fulfilment, as in our ecommerce conversion audit.
3. Know what your attribution tool counts
Write down how each report treats new customers, repeat purchases, recurring payments, refunds and costs. A familiar metric name does not mean two systems measure the same outcome.
HYROS documentation is a good example of why this matters. In its metrics guide:
- Total Revenue includes recurring sales.
- Revenue is first-time sales revenue, excluding recurring revenue.
- Sales is the total number of sales, including refunds.
- Profit is total revenue minus cost, where Cost is the cost of your ads over the selected period.
So the Profit column is not operating profit after product costs, fulfilment, payment fees and software. The Sales count does not tell you how many purchases kept their value. Reconcile both with the business records the decision depends on.
Customer history matters too. HYROS’s reporting guide defines its new-customer filter as customers who made their first purchase since HYROS was implemented, unless existing customers were imported with at least one sale event. Without that import, “new to the reporting tool” and “new to the business” are different populations.
The same guide explains that an attribution window of None lets old clicks still count, while a set window excludes clicks outside it. Compare these settings before reading a higher attributed-sales total as stronger campaign performance.
These are documented definitions. They do not tell you how a particular account, integration or refund process is configured. That is what the trace in section 2 is for.
4. What we learned connecting HYROS for a client
IZZY connected HYROS to the acquisition funnel of an online education business, from registration through booked sales calls to online payment. Four lessons apply to any attribution setup.
The tool could not see the journey the way the site was built
The funnel was built as a single-page app, as many modern funnels are. In this integration we found no supported way to record its screen changes as page views in HYROS: the tracking script guards against being loaded twice, and we found no page-view endpoint. The journey was recorded instead as custom events tied to the visitor’s email.
Two consequences follow. Pages a visitor sees before giving an email cannot be linked to that visitor at all. And every event is only as reliable as the email match. Decide which steps actually inform a decision before instrumenting them.
Matching on email also means sending personal data to the attribution vendor and, through conversion feedback, to the ad platforms. Treat that as a data-sharing decision, with a clear purpose, vendor terms and suppression rules, as you would for automated CRM outreach.
Fire-and-forget tracking fails silently
Tracking calls are usually fire-and-forget: the page sends the event and moves on, so the visitor is never blocked. That also means nothing on the page shows when an event goes nowhere.
It happened twice in this setup. One tracking route was never executed by the hosting platform; requests to it returned an ordinary web page instead of reaching HYROS. Later, the site called an automation webhook before the workflow behind it had been created. The first was fixed by sending events through an n8n webhook that holds the API key server-side; the second by creating the missing workflow.
The review lesson: confirm events in the destination system, not in the code or the browser. A response is not a delivery. The same rule runs through our agent harness framework: an attempted action is not a confirmed one.
Where the sale signal comes from matters
The first version reported each sale to HYROS when the buyer reached the thank-you page. That counts a visit to a confirmation page, not a payment the business can reconcile: a failed charge or a later refund never travels through that route.
In the version that followed, sales reached HYROS from the payment platform through n8n, server-side, and the thank-you page no longer took part in attribution. If your attribution tool learns about sales from a page in the browser, find out what it misses. Each fact should come from the system that is the authority for it, a principle we set out for company systems of record.
Record the sales step, not only the ad and the purchase
For a product sold after a call, the click and the payment are not the whole path. Here, a booked call became its own event that could be crossed with purchases, and each sale could be tied back to the person who handled the call.
Without that middle stage, a campaign that fills the calendar with unsuitable calls looks the same as one that fills it with buyers until the payments arrive.
These are integration lessons, not a performance result. We report no change in ad spend, sales or attribution accuracy.
5. Give bidding the right signal
Improving a report and changing the signal used for bidding are separate interventions. Establish which one you need.
Both platforms offer native routes for downstream outcomes. Google Ads provides enhanced conversions for leads, connecting later customer outcomes with website lead information. Meta’s Conversions API for CRM sends lead stages from your CRM back to Meta for its Conversion Leads goal; Meta currently documents that goal for Lead Ads instant forms only. Check these routes before assuming a separate attribution tool is required to return outcomes to the platforms.
Then inspect the campaign goal itself. Google uses primary actions for bidding when the standard goal is selected, but actions inside a custom goal are used for bidding even when marked secondary. A dashboard that shows the right event does not prove the campaign optimises towards it. Google conversion-action settings
Choose a signal that is useful and timely. Google’s value-based bidding guidance asks you to weigh accuracy against conversion delay, and suggests an earlier stage, such as a qualified lead, when the final sale arrives too late. Keep the paid outcome available for evaluation whichever stage guides bidding.
When you add or change a conversion-feedback integration, validate the existing feed and goals first. HYROS’s own setup guide warns that tracking errors push inaccurate data back to the platforms, that account-wide Google conversion goals used alongside its API will double the conversions entering the account, and that the change can reset learning, so it should be tested on a few campaigns before going account-wide.
6. Then decide: repair, wait or change spend
First check whether the outcomes have had time to arrive. Google explains that recent performance can look weaker because spend is reported immediately while some conversions are still pending. Google conversion-delay guidance
Then connect what the trace found to an action:
| What the trace found | Next useful action |
|---|---|
| An event that never arrives, or a sale signal that is not a payment | Repair the feed; pause the affected route if continued spend would repeat the failure |
| Customer history or report settings explain the discrepancy | Correct the interpretation before changing allocation |
| Suitable leads exist, but payment outcomes are incomplete | Review their progression and wait for the agreed evaluation point |
| Unsuitable enquiries or stalled calls | Repair the promise, qualification or sales follow-up before buying more of them |
| Mature, verified outcomes miss the agreed economic boundary | Test a reduction or stop the route within a written loss limit |
If the trace points to unsuitable enquiries or a broken contact path rather than measurement, work through the five checks before spending more on ads.
For any spending change, write down the hypothesis, maximum acceptable loss, evaluation point and reversal condition. For a bidding test, keep the offer, creative and landing page stable; Google’s experiment guidance cautions against changing them at the same time. Recording each change and judging it at the level the evidence supports is covered in IZZY’s post-launch campaign guide.
Attribution is not the same as additional sales
Attribution allocates credit for an observed outcome. Incrementality asks how many outcomes happened because of the advertising.
A higher attributed total cannot answer the second question on its own. Research comparing observational methods with randomised Facebook advertising experiments found that the methods often did not reproduce the experimental effects. The study is several years old; use it for the methodological distinction, not as a current benchmark. Gordon and colleagues, advertising-measurement study
If the spending decision depends on causal lift, assess whether a properly designed experiment is feasible, with a defined control, outcome and allocation. A bidding comparison is not automatically a no-ad holdout, and sparse evidence may leave the question open.
Start with one unresolved sale
Choose a paid order the reports disagree about. Trace its history, payment, retained value, attribution settings and platform feedback. Classify the discrepancy, then write the smallest repair or test it justifies.
At the next review, check whether that repair or test changed verified paid customers, media cost per paid customer and retained value. Continue, adjust or stop according to those outcomes and the boundary agreed before spending.
Trace one sale with IZZY
Bring the order your reports disagree about, the advertising accounts, the attribution tool and the payment and CRM systems involved. We will trace it with you, classify the gap and show the smallest repair: a reporting setting, a feed that needs to move server-side, or a change to what drives bidding.
Connecting advertising, CRM and payment systems with n8n is the scope of our n8n AI Automation service. Choosing between native platform routes and an attribution tool is part of our Advisory work.
Frequently asked questions
No. Analytics can find a discrepancy or guide a test. The financial effect depends on the change you make. Check paid outcomes and an appropriate comparison before claiming savings or a sales increase.
Use a meaningful stage with reliable, timely data. Google’s guidance weighs outcome accuracy against delay. If qualified leads guide bidding, keep checking how they become paid customers.
From the system that records the payment, server-side, with an identifier that matches the lead. A browser event on a thank-you page can miss failed charges and refunds, and can fail silently.
Not necessarily. Google and Meta both offer native routes, with their own limits. Whether they meet your reporting and operating needs depends on the gap your trace found.
Sources and evidence limits
Sources and linked IZZY pages were checked before publication. HYROS documentation supports the stated definitions and setup cautions; Google and Meta documentation support the platform capabilities described. None of these sources establishes a financial result for your account.
Section 4 describes an integration IZZY built for a client, who is not named. It reports what was built and changed, not a performance result: no savings, sales increase or attribution-accuracy figure is claimed. Observations about HYROS reflect that integration and may change as the product evolves.
HYROS is used as an example. IZZY receives no compensation for its inclusion, and this article is not an independent accuracy benchmark of the tool.