Why does tracking work in preview but fail in production?
Preview mode is not the same environment as a real customer session, and four specific differences account for almost every case: preview usually runs without a consent banner, without an ad blocker, on a desktop browser rather than the in-app or mobile browser your customers use, and often against a staging URL whose structure differs from production. A real test purchase on a real device, with the banner present and nothing disabled, is the only test that means anything — and it is the step most often skipped.
Free30 minutesNo obligation
Preview works, production fails · what breaks
What this looks like when it's happening to you — Preview works, production fails
- 01
Tag Assistant shows everything firing
And the platform receives nothing.
- 02
Works for you, not for customers
Because you are not testing what they experience.
- 03
Works on desktop, fails on mobile
A narrower version of the same problem.
Preview works, production fails · the fix
What changes once it's fixed — Preview works, production fails
A test that actually predicts production
Real device, real order, banner present.
Environment differences identified
So preview stops giving false confidence.
Verification that holds
Confirmed against your own order records rather than a debugger.
Preview works, production fails · architecture
Why your conversion data was wrong in the first place — Preview works, production fails
Fixing this specific fault is worth doing. But it is worth understanding why it happened, because the same underlying weakness produces a different symptom every few months — and the reason your conversion data is wrong is almost never the tag itself.
- 1
Your conversion depends on a browser that is being restricted on purpose
A browser-side tag has to survive the page loading, the visitor staying, the script executing and the request leaving. Every one of those is now contested. The order in your database is certain; the browser event reporting it is not, and that gap is where wrong numbers come from.
- 2
Safari caps the cookie your attribution depends on
Intelligent Tracking Prevention limits cookies written by JavaScript to seven days, whatever expiry you set. Any customer who takes longer than a week to decide comes back as a stranger, so the sale is recorded and credited to nobody. A cookie your own server writes in an HTTP response is not subject to that cap — which is the single largest practical difference between browser-side and first-party server-side tracking.
- 3
Ad blockers remove the tag before it can run
A blocked pixel does not error, retry or warn anyone. The visitor buys, the order lands in your store, and no platform ever hears about it. First-party events sent from your own subdomain aren't filtered the same way, because they are not a third-party request.
- 4
Consent banners stop tags, sometimes when they didn't need to
Where consent is genuinely required, losing the event is correct. What is not correct — and what we find repeatedly on US-majority accounts — is an EU-style prior-consent banner applied to every visitor, discarding conversions no law asked you to discard. A server container enforces the real decision at the container rather than hoping each vendor script honours a flag.
- 5
The checkout can end somewhere your tag isn't
Wallets and hosted payment pages can be abandoned after the money moves. The customer paid; the confirmation page never rendered; the browser event never fired. Sending the purchase from the order record instead means the conversion arrives whether or not anyone came back to your site.
- 6
What that adds up to
Each of these is a separate route to preview works, production fails, and they compound. First-party server-side tracking doesn't patch them one at a time — it moves the money event off the browser entirely, so the order in your backend is what gets reported. That is why we reconcile against your own order records and put 95%+ accuracy in writing: with the event sourced from the order, the number is checkable rather than hopeful.
Preview works, production fails · normal or broken
Some of this is supposed to happen — Preview works, production fails
Half the time nothing is broken and the number is being read wrong. Knowing which is which stops you paying to fix something that was never a fault.
Normal, not a fault
Preview showing more events than production
Expected, because preview has no consent banner and no blocker.
Some production traffic never producing events
Blocked or denied users. Correct, not a fault.
Actually broken
No consent banner in preview
Tags that would be blocked for a real visitor fire freely for you.
No ad blocker in preview
A blocked request in production simply doesn't exist.
Desktop preview versus in-app mobile reality
Especially for social traffic, where most sessions are in-app.
Staging URL structure differing from production
URL-based triggers match in one and not the other.
Preview works, production fails · diagnose
How we find out why does tracking work in preview but fail in production
Always in this order. Each step eliminates a class of cause, so by the last one there is usually a single explanation standing.
- 01
Complete a real order on a real device
Consent banner present, nothing disabled. The single most useful test available.
- 02
Repeat with an ad blocker enabled
To see what that segment of your traffic actually produces.
- 03
Repeat through each payment method
Redirect behaviour differs.
- 04
Compare staging and production URL structure
Against every URL-based trigger.
- 05
Reconcile against backend orders
The only check that isn't affected by the environment you tested in.
Preview works, production fails · prove
How we prove it's actually fixed — Preview works, production fails
The tag firing again proves the tag fires. These prove the numbers are right.
Source of truth
Your backend order records
- A live order completed on a real device with the consent banner present, verified end to end.
- The same test repeated with an ad blocker enabled, so the blocked-traffic case is understood.
- Every payment path tested individually.
- Reconciliation against backend orders over a closed window.
- Reconciled against your backend order records for a fully closed window, matched on ID rather than on totals.
- No record counted twice, checked across the whole window rather than spot-checked.
- The same check repeated 30 days later, because a fix that holds once and drifts is the harder failure to catch.
Preview works, production fails · free audit
What we check in the free audit
- Complete a real order on a real device
- Repeat with an ad blocker enabled
- Repeat through each payment method
- Compare staging and production URL structure
- Reconcile against backend orders
Checked against
Your backend order records
Free30 minutesNo obligation
Preview works, production fails · when not to buy
When you don't need us for this — Preview works, production fails
Most of these you can do yourself, and we would rather you did.
- You can place one real order on your phone with the banner up — that's the whole test, and it costs you the price of one order.
- You already reconcile against your backend, which is immune to this class of confusion.
- You have no consent banner and no meaningful blocker share, in which case preview is closer to reality than usual.
Preview works, production fails · what we deliver
What we do about it — Preview works, production fails
Test as a customer
Real device, real order, consent banner in play, nothing disabled.
Test each payment path separately
Wallets and hosted pages behave differently from card fields.
Verify against the backend
Reconciliation, not a debugger, is the proof.
Pricing
Pick the package that fits your ad spend
Browser-side starter
$400One-time fee · lifetime ownership
We fix the fault in your existing browser-side setup and verify it on a live transaction. Honest disclosure: this repairs what you have, it does not make it durable. If you spend $1K+/month on paid media, the fault will recur the next time Safari, a consent banner or a checkout change moves under you — the server-side tier is what stops that.
- Root cause found and fixed, not worked around
- Verified on a live transaction, not in preview
- Reconciled against your own order records
- Written handover documenting what changed and why
1st-party server-side
Recommended$800One-time fee · most popular
The tier that removes the whole class of failure rather than this instance of it. Your money events stop depending on the browser: they are sent from your order records through a first-party server container on your own subdomain, so iOS, ad blockers, consent banners and interrupted payment redirects can no longer take them away. Built in 2–4 days. You own the stack outright — no monthly SaaS.
- First-party server container on your own subdomain
- Purchases sent from your order records, not the browser
- Meta, Google, TikTok, Microsoft and LinkedIn all fed server-side
- Deduplicated on a shared event ID, verified at every destination
- First-party cookies beyond Safari's 7-day cap on JavaScript-set cookies
- 95%+ conversion accuracy and 8+ Meta EMQ, guaranteed in writing
Full 1st-party stack
$1,800One-time fee · lifetime ownership
Every paid-media channel server-side, end to end, plus the offline loop. For advertisers whose spend makes an attribution leak more expensive than the build. Pay once, own everything.
- Everything in 1st-party server-side
- Pinterest, Snapchat, Reddit and LinkedIn CAPI
- CRM and offline conversion loop, so closed revenue reaches bidding
- Refunds and cancellations reversed at every destination
- Klaviyo identity stitching, BigQuery and Looker Studio
- Monitoring, so the next break surfaces in days rather than a quarter
FAQ
Preview works, production fails questions, answered
It doesn't — it accurately reports what happened in that session. The session just isn't representative of a real customer's.
Yes, at least once. It tells you what that share of your traffic actually produces, which is otherwise invisible.
For catching outright breakage, yes. For confirming production behaviour, no — the URL structure and consent behaviour usually differ.
One real order, on a phone, with the consent banner present, on production, verified against the order record afterwards.
Free · 30 min
Get preview works, production fails diagnosed properly
Free30 minutesNo obligation
