The Ghost in the Machine: Why Your Offline Conversion Uploads Are Failing and What to Do About It

90 days. That is the entire window you get to upload a Google Ads offline conversion before the GCLID ages out and your closed deal becomes invisible. Miss it, and Google Ads will tell you the conversion never happened, even though your CRM says the contract is signed and the money cleared.

I have spent the last few years cleaning up conversion pipelines for B2B teams, and the single most expensive bug I find is not a broken pixel. It is a silent one. An offline conversion upload that returns "success" in the API, shows zero errors, and still moves no data into the campaign that earned the lead.

This is not a "fix your error codes" post. Google's docs already list the error codes, dry as sand. This is a post about the ghost in the machine: the gap between "my CRM closed the deal" and "Google Ads thinks that click went nowhere." That gap has a precise location in your pipeline. You can find it.

The honest read is that offline conversion failure is not a reporting problem. It is an algorithm-training problem. When your highest-value conversions never reach Google, Smart Bidding optimizes toward whatever cheap signal it can still see. The fix is architectural, and DataCops exists because the upstream pipeline that feeds these uploads is almost always the thing that is actually broken.

Quick stuff people keep asking

Why are my Google Ads offline conversions not uploading? Usually one of four things: the GCLID expired past the 90-day window, the conversion action type does not match the upload type, your timestamps are in the wrong time zone, or the GCLID never got captured at the lead-form stage in the first place. The upload "succeeds" on three of those four and still imports nothing.

How long do you have to upload an offline conversion? 90 days from the click for GCLID-based uploads. For enhanced conversions for leads, you are matching on hashed email or phone, so the click-ID clock matters less, but the conversion still needs to land inside the action's lookback window.

What does "GCLID expired" mean? The Google Click Identifier attached to that lead has aged out. Google will not associate a conversion with a click older than 90 days. B2B sales cycles routinely run 90 to 180 days. So your best, most considered deals are structurally the ones most likely to fail. That is the cruel part.

Why do my uploads succeed but show no data? Because "success" in the offline conversion import API means "your file was syntactically valid and accepted," not "this conversion was attributed." A row with an expired GCLID, a wrong action name, or a future-dated timestamp passes ingestion and then quietly gets discarded. No error. No data.

What is the UPLOAD_CLICKS type error? Your conversion action in Google Ads is configured for one import method and your upload uses another. Upload a GCLID-based row against an action set up for enhanced-conversion data uploads and the type mismatch kills the row. The action has to be created as the right type before the first upload, not patched after.

How do I debug offline conversions in Google Ads? Work the pipeline backwards in stages, not the error log. Confirm the GCLID was captured at form submit. Confirm it survived the trip into your CRM field. Confirm the upload job ran. Confirm the conversion action type matches. Confirm the timestamp and time zone. The failure is almost always at one specific stage, and naming the stage is the whole job.

What is the difference between online and offline conversions? Online conversions fire from a browser event the moment they happen. Offline conversions are events that happen away from the site, a sales call, a signed contract, and get matched back to the original ad click later, by GCLID or hashed identifier. Online is real-time and lossy at the edges. Offline is delayed and lossy in the pipeline.

The ghost-in-the-machine pipeline: where the failure actually lives

Here is the thing nobody tells you. Offline conversion tracking is not one system. It is a chain of five handoffs, and a break at any link looks identical from the Google Ads UI: no data. The skill is locating the break.

Link one: capture. The GCLID has to be read from the landing-page URL and written into a hidden field on your lead form. If your form is on a subdomain, or a third-party form embed, or an SPA that re-renders the URL before the script runs, the GCLID silently never gets captured. Every downstream step then works perfectly on a value that does not exist. This is the most common failure and the hardest to see, because the upload file looks fine. It just has blank GCLIDs.

Link two: storage. The GCLID lives in a CRM field, Salesforce, HubSpot, a custom field. It has to survive lead merges, deduplication, and sales reps editing records. CRM admins routinely map the field wrong, or a dedupe rule overwrites the GCLID-bearing record with a cleaner-looking duplicate that has no GCLID. The deal closes on the record without the click ID.

Link three: the clock. Your sales cycle is 110 days. The GCLID window is 90. The deal closes, the upload runs, and the GCLID is 20 days expired. The row is accepted and discarded. This is not a bug you can fix with better code. It is a structural mismatch between how long humans take to buy and how long Google will remember a click. Enhanced conversions for leads is the real workaround here, because it matches on hashed email instead of a decaying click ID.

Link four: the type and the name. The conversion action in Google Ads must exist, must be the correct upload type, and the action name in your upload file must match it character for character. A trailing space, a renamed action, a sandbox-versus-production mismatch, all kill the row.

Link five: the timestamp. Conversion time has to be in a format Google accepts and in the right time zone. A conversion dated in the future, even by an hour because of a UTC-versus-local-time slip, gets rejected. A conversion dated before the click is rejected. Time-zone mismatch between your CRM export and your Google Ads account is a classic silent killer.

Run a real audit and you find the breakage clusters at link one and link three. Capture and the clock. Not the upload code everyone obsesses over.

Now the part that matters more than the reporting. Layer 5 of the data problem. When your closed-won deals never make it to Google, Smart Bidding does not stop optimizing. It optimizes on what is left, the cheap form-fills, the low-intent newsletter signups, the lead-magnet downloads. It learns that those are your conversions, because as far as it can see, they are the only ones you have. It pours budget toward the audiences that produce more of them. Your real buyers, the 110-day enterprise deals, get less budget, because the algorithm was never told they exist.

That is the ghost. Your CRM is full of revenue. Your ad account is training itself on the cheap stuff. The two never met.

And the deeper reason this keeps happening: the data is flowing through a pile of disconnected, third-party scripts and CRM integrations with no isolation and no validation before it leaves your infrastructure. The GCLID gets handed from a form embed to a CRM connector to an upload script, and nobody owns the chain end to end. DataCops fixes the upstream side of this: a first-party data pipeline running on your own subdomain, capturing the click identifier and session truth at the source, before any third-party handoff can drop it. When the capture layer is yours and is first-party, the GCLID is not at the mercy of an embed that loaded too slow or a CMP race condition. CAPI delivery to Google and Meta then ships from clean, validated data instead of from whatever survived the relay race.

A diagnostic framework: the four-question audit

When a B2B team tells me "our offline conversions are not working," I do not open the error log. I ask four questions in order. The first one that gets a "no" or an "I don't know" is your failure point.

Question one. Pull ten recently closed-won deals from your CRM. Do all ten have a non-empty GCLID field? If some are blank, your failure is at capture, link one. Fix the form. Nothing downstream matters until this is yes.

Question two. For the deals that have a GCLID, how old is the GCLID at the moment the deal closed? If your median is past 75 days, you are losing deals to the 90-day clock and you should move to enhanced conversions for leads, which matches on hashed email and is not chained to the click-ID expiry.

Question three. Does the conversion action in Google Ads exist, is it the correct upload type, and does its name match your file exactly? If you cannot answer all three with a confident yes, that is your failure.

Question four. Export one conversion row and check the timestamp. Is it in the past, and in your Google Ads account time zone? A future date or a time-zone slip rejects the row silently.

Four questions. The break is almost never where the error log points, because the worst failures produce no error at all.

The Meta CAPI parallel

Same disease, different host. On Meta, offline events fail to match for the mirror-image reasons: weak or missing match keys, no hashed email or phone or external ID on the event, and timestamps outside the attribution window. Meta will happily accept an offline event with thin matching parameters and then quietly fail to attribute it. Your Event Match Quality score drops, attribution thins out, and Meta's algorithm, just like Google's, starts optimizing on the cheap signals it can still see.

The root cause is identical. Events handed between systems with no validation and no isolation, so the match keys degrade in transit and you find out months later when ROAS has already slid.

Decision guide

B2B sales cycle longer than 90 days? Stop relying on GCLID uploads. Move to enhanced conversions for leads, which matches on hashed email and survives long cycles.

Uploads "succeed" but show zero conversions? Run the four-question audit. Start at capture. Do not touch the upload script until questions one and two are clean.

Lead forms on subdomains, embeds, or an SPA? Treat GCLID capture as broken until proven otherwise. This is where the ghost lives. Move capture to a first-party layer.

CRM is Salesforce or HubSpot with active dedupe rules? Audit whether your dedupe logic preserves the GCLID-bearing record. It usually does not.

Running paid on both Google and Meta? Fix the upstream capture once, feed both. A clean first-party pipeline solves the Meta match-quality problem and the Google upload problem in the same move.

Already done all of the above and ROAS is still soft? Your uploads may be landing but carrying bot-contaminated and low-intent noise. The next audit is data quality, not pipeline plumbing.

You are debugging the wrong layer

The mistake I see on every one of these calls is the same. The team treats offline conversion failure as a Google Ads problem and spends a week in the error log. The error log is the last place the failure shows up and the least useful place to look. The failure happened three systems upstream, at a form embed or a CRM field, and it left no error because the upload was syntactically perfect. It just carried nothing, or carried something expired.

Reframe it. This is not a reporting bug. It is the algorithm being trained on your worst leads because it was never shown your best ones. Every week that runs, Smart Bidding gets more confident about the wrong audience.

So go pull ten closed-won deals from your CRM right now. Check the GCLID field. If even three of them are blank, you have just found the reason your Google Ads bidding has been quietly optimizing against you, and you found it in two minutes, in the one place you were not looking.