Get your The Reconciliation Method tracking to 95%+ accuracy guaranteed in writing.
Every tracking project we run follows the same ten steps in the same order, starting from one question: what does your business count as a conversion, and which system holds the real record of it? This page documents the method end to end, including the step almost everyone skips — telling attribution differences apart from actual data loss. It's written so you can audit our work, or run it yourself.
Free30 minutesNo obligation
The Reconciliation Method · what breaks
What goes wrong without a method
- 01
Everyone starts by fixing tags, not defining the conversion
"Purchase" means one thing in the store backend, another in Stripe, and a third in the CRM after cancellations. Tags get rebuilt against a definition nobody wrote down, so nothing ever reconciles.
- 02
Attribution differences get misdiagnosed as broken tracking
Meta counting a 7-day-click conversion that GA4 gives to organic is not data loss — both systems are behaving as designed. Rebuilding a pixel to chase that gap spends the budget and fixes nothing.
- 03
The rebuild goes wider than the actual problem
With no audit scoping the failure, teams replace everything at once — new container, new dataLayer, new consent tool — and afterward nobody can say which change moved which number.
- 04
Nobody checks the finished work against the backend
The setup gets called done when events appear in a debug view. Appearing in a debug view and matching your order table are two different claims, and only one is worth paying for.
The Reconciliation Method · the fix
What the method gives you that a tag rebuild doesn't
One number everyone in the room argues from
Finance, the media buyer and the agency all reconcile to the same system of record. Disagreements turn into questions about attribution windows rather than about whether tracking works at all.
You know which gap is fixable and which isn't
The audit separates events that never left the browser from conversions two platforms both legitimately claim. Only the first category is an engineering problem, and we say so plainly.
Platforms learn from outcomes, not from form fills
Qualified stages returned from the source of truth — paid order, kept appointment, closed deal — become the conversion the bidding algorithm optimizes toward.
The work stays auditable after we leave
The conversion definition, the source of truth, the dedup keys and the validation results are all written down. Your next vendor inherits a document instead of a mystery.
The Reconciliation Method · architecture
The ten steps, in the order they have to run — The Reconciliation Method
Each step exists because the next one depends on it, which is why the order isn't negotiable. Reordering it is how projects end up rebuilding tags against a conversion definition nobody ever agreed on.
- 1
Steps 1–2: Define it, then name the record
We write down what counts as a conversion and which single system — store backend, payment processor, booking tool or CRM — holds the real record of it.
- 2
Steps 3–4: Map the lifecycle, audit the signals
The conversion is mapped from first touch through cancellation, refund and rebill, then browser, server, consent and CRM signals are inventoried across the same set of real orders.
- 3
Steps 5–6: Split the gap, then rebuild narrowly
Conversions the platforms merely attribute differently are separated from conversions no platform ever received, and only that second column justifies touching the setup.
- 4
Steps 7–8: Dedupe on event_id, check vs. orders
Browser and server pairs are collapsed by a shared event_id, then what the platforms report is joined to real backend orders on transaction_id.
- 5
Steps 9–10: Return outcomes, then monitor
Qualified backend stages — paid, kept, closed-won — go back to the platforms, and alerts then watch event volume, dedup rate and match quality daily.
The Reconciliation Method · diagnose
Run the first five steps on your own setup this week — The Reconciliation Method
None of this needs our access — a spreadsheet, one month of orders and one platform export will get you through the step everybody skips.
- 01
Write the conversion definition in one sentence
State what counts, at which lifecycle stage, and what doesn't — refunds, cancellations, staff orders, test checkouts. If two people on your team write different sentences, you've already found something.
- 02
Name one system as the record, and one field
Pick Shopify orders, Stripe charges or CRM closed-won — one, not a blend of all three. Write down the exact field you'd sort by and the timestamp you'd date a conversion to.
- 03
Pull last month's orders and one platform export
Export the source-of-truth records for a settled month, then the same month's conversions from one ad platform with the transaction identifier included. Identical date boundaries on both sides.
- 04
Join on transaction ID, not on totals
A lookup formula is enough. Backend rows with no platform event are candidates for real data loss; platform rows with no backend order are deduplication or phantom-conversion problems.
- 05
Sort the misses into attribution vs. data loss
A conversion another channel got credit for still appears somewhere in your reporting — that's attribution, and it's normal. A conversion no platform ever received, in any report, is technical loss. Only the second list deserves engineering budget.
- 06
Check whether the losses cluster
Group the unmatched orders by browser, device, payment method and consent state. Losses concentrated on iOS Safari, PayPal or denied consent point at one mechanism rather than a general collapse, which changes what you'd fix.
The Reconciliation Method · prove
How we prove the rebuild actually worked — The Reconciliation Method
The method closes with the same join it opened with, re-run against the rebuilt setup. Events appearing in a debug view is not the finish line; a settled month of orders reconciling is.
Source of truth
Store backend orders · payment processor charges · CRM closed-won
- Re-run the step-four signal audit on a fresh set of real orders, confirming each signal now reaches the platform it should.
- Join rebuilt platform events to backend records on transaction ID over a settled window; every unmatched row gets a reason code.
- Confirm the ratio clears 95% against the named source of truth — and that it isn't above 100%, which is an equal-severity defect.
- Verify browser and server pairs deduplicated inside each platform's own interface for named orders, not inferred from the payload.
- Re-test granted, denied and analytics-only consent to confirm the denial path suppresses exactly what it is supposed to suppress.
- Leave alerts on event volume, dedup rate and match quality, so drift after handoff surfaces as an alert, not a surprise.
The Reconciliation Method · free audit
What we check in the free audit
- Write the conversion definition in one sentence
- Name one system as the record, and one field
- Pull last month's orders and one platform export
- Join on transaction ID, not on totals
- Sort the misses into attribution vs. data loss
- Check whether the losses cluster
Checked against
Store backend orders · payment processor charges · CRM closed-won
Free30 minutesNo obligation
The Reconciliation Method · when not to buy
When the full method is overkill
Ten steps is a lot of process, and there are setups where it buys you almost nothing.
- One checkout path, one ad platform, and the platform count already lands within a few points of your order table monthly. Do steps one and two, stop.
- You aren't running paid media yet. Get spend on one platform first — a reconciliation with nothing to feed is documentation, not a return.
- You're mid-replatform. Wait until the new checkout and order pipeline are live, or everything reconciled against the old one has to be redone.
- Your real conversion happens offline and nothing records it yet. Build that record first; the method has nothing to reconcile against until you do.
The Reconciliation Method · what we deliver
What the method produces
A written conversion definition
One paragraph per money event stating what counts, what doesn't, and at which lifecycle stage it is counted. You sign off on it before anything gets built.
A named source of truth per conversion
Store backend, payment processor, CRM, booking system or call platform — one system per conversion, with the exact field and timestamp we will reconcile against.
A conversion lifecycle map
First touch through order, cancellation, refund and rebill, so it's clear which stage each platform should be told about and which stages it should never see.
A four-way signal audit
Browser, server, consent and CRM signals inventoried side by side across the same set of real conversions, showing precisely where each signal stops.
The attribution-versus-loss split
A table separating conversions the platforms merely disagree about from conversions no platform ever received. We only engineer against the second column.
A reconciliation ledger and live monitor
The validation run that closes the project, plus alerts watching event volume, dedup rate and match quality against backend counts from then on.
Written guarantee
95%+ tracking accuracy
Measured against Store backend orders · payment processor charges · CRM closed-won. Miss it and we keep working at no cost.
FAQ
The Reconciliation Method questions, answered
Because a name fixes the order of operations, and the order is the whole point. Naming it also gives you something to hold us to — if we start rebuilding tags before we've named your source of truth, we've skipped our own method and you can say so.
Step five: separating attribution differences from technical data loss. Almost every "our tracking is broken" conversation starts with two numbers that were never supposed to match — a platform's click-attributed count and a backend order count. Until you split those apart, you don't know whether you have an engineering problem or a reporting misunderstanding.
No, and I'd rather be clear about it than sell you something. Those orders still happened and the money still arrived. What you lose is the platform's ability to learn from them, which shows up later as worse targeting and misread campaign performance. That's a real cost, but it isn't the same as lost sales, and any vendor conflating the two is doing it on purpose.
That happens often — phone sales logged in a spreadsheet, deals closed in email, a CRM where stages get set inconsistently. When there's no reliable record, step two becomes building one before anything else proceeds. Sometimes that means a webhook into the CRM, sometimes it means agreeing that the payment processor is the record and the CRM is decoration.
Yes, and the steps are here for that reason. The hard parts aren't secret: getting the business to agree on one conversion definition, and having the discipline to audit before rebuilding. What usually stops teams is time, not capability — the signal audit alone means pulling matched records from four systems for the same set of orders.
If you have one checkout path, one ad platform, modest spend, and your platform's conversion count already lands close to your order table each month, the method will mostly confirm you're fine. Run steps one and two yourself, compare a month of orders, and if the numbers agree, keep your money. We'd rather tell you that than sell a rebuild nothing required.
The build is usually 2–4 days once the audit is done. The audit itself depends on how many systems hold pieces of the truth — a single store backend is quick, a CRM with custom stages plus a call platform takes longer. Monitoring runs from the day we hand off, not from the day something breaks.
More on The Reconciliation Method
- GuideHow Accurate Is Your Conversion Tracking? Here's How to Actually Measure It
- GuideHow We Fix Any Conversion Tracking Problem With First-Party Server-Side Tracking (And Guarantee 95%+ Accuracy)
- GuideServer-Side Tracking Explained: What It Is, How It Works, and Who Actually Needs It (2026)
- GA4 reconciliationGA4 source-of-truth reconciliation
- Tracking qualified leads not form fillsHow do I track qualified leads instead of form fills?
- GA4 source/medium incorrectWhy is my GA4 source and medium wrong?
Free · 30 min
Want the method applied to your setup?
WhatsAppFree30 minutesNo obligation
