Why Ad Measurement Is Incomplete: Consent and Modelling

The conversion count in the ad dashboard is the single number most businesses rest their budget decisions on. That number is assumed to reflect real demand one for one. The assumption is wrong, and nobody hides it — the dashboard itself says so, on separate lines, in footnotes and with the word "estimated". The problem is that those footnotes never make it into the decision meeting.

In Turkish-language sources, measurement is handled almost entirely from the setup side: how to place the tag, how to build the consent banner, which box to tick. That content is saturated and mostly current. What is missing is reading the result the advertiser sees once setup is finished: why some conversions disappear, what "modelled" means, which compensating mechanism is worth it for whom, and what decisions taken on broken measurement cost.

The text below is not a setup guide. It describes where the measurement chain breaks, how the break looks in the dashboard, and with what confidence interval you should be reading the number in front of you.

Why the number you see in the dashboard is not the same as reality

For a conversion to land in the dashboard, several interdependent things have to work at the same time: the user has to carry an identifier when arriving from the ad, that identifier has to be storable on the site, it has to be readable again at the moment of conversion, and the record has to be transmitted to the ad account. When any link in the chain breaks, the conversion is not counted even though it happened.

That fragility is not new, but in recent years it has been under pressure from three directions at once: browsers restricting third-party identifiers, operating systems putting in-app tracking behind permission, and consent regulation pushing measurement tags to fire only after consent. All three work the same way — the measurable share is narrowing.

The conclusion is not pessimistic, only honest: the number in your dashboard is not a count but an estimate. How good an estimate depends on your setup. What you need when deciding is less the number itself than the direction it is skewed in.

What breaks in the chain when cookie consent is refused

A visitor who passes the consent banner with "reject" does not leave your site; measurement tags simply do not run at full permission in their browser. On Google's side, the arrangement carrying that preference through to the measurement layer goes by the name Consent Mode v2; whatever the name, the logic is the same — consent state passes to the tags as a signal and changes how they behave. In practice three things happen at once.

First, no persistent identifier can be written that would recognise the visitor in a later session. The same person comes back the following day and looks like two different users. Second, the tie between the ad click and the conversion weakens; even if the conversion happens, which campaign it came from cannot be recorded and does not land in the report. Third, remarketing lists are not fed; list size stays below expectations and campaigns targeting that audience cannot go live.

In the dashboard this usually looks like this: clicks normal, conversions low, remarketing lists smaller than they should be. The team reads that picture as "the campaign is performing badly" and starts changing things on the bidding side. Because the change will be read with the same broken measurement, the discussion goes round in circles.

The critical distinction: the conversion of a visitor who refused consent is not failing to happen, it is failing to be seen. The two are identical in the dashboard and completely different in reality. How to verify by hand whether a conversion is genuinely being recorded, and the diagnostic order in full, we set out step by step in our article on ad spend going out with no conversions.

What a click identifier is, and where it goes missing

When an ad is clicked, an identifier for that click is usually appended to the destination address; on the Google Ads side this is the gclid parameter you see in the address bar. Its function is simple: to let that click be matched to a conversion that happens later.

The typical places the identifier goes missing are these:

• Redirect chains. If the address passes through a shortener, a campaign tool or an old 301 rule, the parameter can be dropped along the way.
• A change of domain. If the form or the payment step moves to a different domain, in most setups neither the session nor the identifier travels with it.
• Scripts blocked before consent. If the script that is supposed to store the identifier does not run before consent, the parameter appears on the page but is never written anywhere.
• Server-side caching. If the query parameter is ignored while the page is served from cache, the measurement layer never sees the parameter at all.
• A long delay. If the time between the click and the conversion exceeds the period the identifier is stored for, no match can be made. In services with long decision cycles this happens more often than people assume.

The quickest way to understand the loss is a manual test: click your own ad, note the parameter in the address bar, browse the site through to the conversion step and check whether the parameter is still being held. A parameter dropped during a redirect shows up in the first ten seconds of that test.

What a "modelled conversion" means, and how to read one

To close the unmeasurable share, ad platforms use modelling: they start from the pattern in observable conversions and produce a statistical estimate for the part that cannot be observed. In the dashboard this usually arrives inside the total figure, flagged separately with the word "estimated".

A modelled conversion is not a fake number; but it is not an observed number either. Reading it correctly means holding on to the following distinctions:

• Usable for trend. The direction between months — rising or falling — can be read from modelled data too, because the model works the same way each time.
• Weak for a single decision. The answer to "should we shut this campaign down" should not rest predominantly on a handful of modelled conversions.
• Handle with care in cross-channel comparison. If two channels do not use the same modelling approach, the two numbers are not on the same scale.
• Model quality depends on the data. If the observed side is weak, the model's foundation is weak too; modelling is not a solution that replaces measurement but a correction that sits on top of it.

The practical rule: ask to see the observed and the modelled portions on separate lines in the report. If they cannot be separated, lower by one notch the precision of any decision you take on that number.

How much do enhanced conversions make up for, and who is it not worth it for

Enhanced conversions aim to raise the match rate by sending information you already collect from the user (an email address, for instance) to the platform in hashed form. When the cookie-based chain breaks, part of the conversion can be recovered through that second route.

The cases where it is worth it are fairly clear: if you collect a genuine piece of identifying information at the moment of conversion (a form, a sign-up, an order), if your conversion volume is regular, and if you have a technical resource to set it up correctly and maintain it, the gain is measurable.

The cases where it is not worth it are just as clear and usually go unwritten. If your conversion is only a phone call or a tap on directions, there is no identifying information to send. If your monthly conversion count is in single figures, an increase in match rate will not register meaningfully in the number. If there is no technical resource to maintain the setup, a half-finished implementation does not improve measurement, it complicates it. And most importantly: every mechanism that sends user data has to be consistent with your privacy notice and your consent arrangement. A compensating mechanism built without securing that consistency creates a bigger problem than the one it solves.

The phone call: the most frequently lost conversion channel

In businesses such as clinics, estate agencies, law firms, plumbing services and hotels, most demand arrives by telephone. On the measurement side this channel is noticeably more fragile than a form conversion — because the call itself happens outside the browser.

The routes used in practice are these:

• Counting a tap on the number as a conversion. The easiest method. Its limit is that a tap does not prove a conversation took place. A user who touched the wrong number or changed their mind is counted too.
• Platform reporting of calls made through the ad. Platforms have their own reporting mechanisms for calls arriving via ad assets; because the scope and level of detail can vary by country and by account type, what is actually reported has to be verified from your own dashboard's documentation.
• Dynamic number insertion. Call tracking software that shows a different number on the site depending on the visitor's source ties the call back to that source. It gives the most detailed data; in return it brings an additional software cost and responsibility for processing personal data.
• Manual logging. The person answering the phone asks "how did you find us" and writes the answer down. It looks primitive, but kept up consistently it is often the most reliable method at small volumes.

Which route to choose depends on volume and on the length of the decision cycle. In a business receiving a few calls a month, manual logging is enough; where hundreds arrive, it becomes inconsistent. For businesses where the telephone is the main channel, we set out how this arrangement joins up with the page side on our solution page for dental and aesthetic clinics.

Offline conversions: writing closed sales back into the account

A customer who fills in an enquiry form and signs a contract three weeks later appears in the ad account only as a "form". Yet the business outcome is not in the form, it is in the contract. The way to close that gap is to write the closed record on the sales side back into the ad account.

The precondition is not technical but operational: the form record and the sales record have to be matchable through the same identifier. If your sales team copies the enquiry into its own spreadsheet by hand and the source is not recorded, there is no link to write back. That is why offline conversion projects usually begin as advertising projects and end as record-keeping discipline projects.

When the write-back works, the gain is this: the goal the bidding algorithm chases stops being "a form" and becomes "a closed sale". That raises the quality of automation directly. Why the decision about how much to hand over to automation depends on a clean signal we dealt with separately in our article on PMax and AI Max.

The relationship between your consent text and your measurement setup

Every compensating mechanism on the measurement side processes personal data. The technical setup and the legal texts therefore have to agree; when they do not, the resulting risk is greater than the measurement loss.

There are a few items to check in practice: that ad measurement and remarketing purposes appear explicitly in your privacy notice, that the categories named in your cookie policy are the ones actually used on the site, that the consent preference is genuinely passed through to the measurement layer, and that additional software such as call tracking appears in the texts as a party processing data. One item is often skipped: the reject option being as accessible as the accept option.

These lines are not legal advice; they are the checkpoints to look at so that your measurement setup does not contradict your legal texts. For the texts themselves you need to work with your own legal counsel.

A measurement health check: seven items to look at once a month

Seven lines looks long, but the whole set can be swept in an hour of regular checking. The value of that check is not in doing it once but in archiving it: fill in the same table every month and put them side by side, and you can tell whether a drop in a metric came from the market or from measurement. When building the same logic for organic and AI-sourced traffic, the verification steps in our article on tracking AI traffic in GA4 can be used.

What decisions taken on broken measurement cost

The cost of broken measurement is not the number of conversions lost. The real cost is the decisions made on that incomplete data, and it shows up in three forms.

The first is shutting down a channel that works. A channel that cannot be measured looks like "no return" in the dashboard and gets switched off. After it is switched off total demand falls, but the reason for the fall can no longer be measured, because the channel is closed.

The second is steering automation in the wrong direction. When the bidding algorithm is fed an incomplete signal, it optimises towards the conversion type that can be measured. In a business whose forms are measured and whose phone calls are not, the system learns to move away from the audience that converts by telephone. That does not appear in the dashboard as an error — the structure of demand simply changes, quietly.

The third is the discussion locking up. While measurement is contested, every performance meeting is spent arguing about whether the numbers are right and produces no decisions. A significant part of the loss of trust between agency and business starts here; the two sides are in fact measuring the same thing differently.

We have listed the scope of how measurement is set up and which items arrive in writing in the monthly report on the page for our ad management service. If you would like to sweep the seven items together on your own account, you can get in touch.

Frequently Asked Questions

Is the conversion of a visitor who refuses cookie consent never counted?

The visitor can convert; what goes uncounted is which ad the conversion came from. When consent is refused, no persistent identifier can be written, so the tie between the click and the conversion weakens and the record either does not land in the report at all or lands without source information. Platforms try to close part of that gap with modelling. The distinction to remember when deciding is this: a conversion not happening and a conversion not being seen look identical in the dashboard and are two completely different things in reality.

Can modelled conversions be treated like real conversions?

For reading a trend yes, for making a single decision no. A modelled conversion is a statistical estimate produced from the pattern in observable records; it is useful for reading the direction between months because the same method is applied each month. But resting a decision such as shutting down a campaign or doubling a budget predominantly on a handful of modelled conversions is risky. Ask for the observed and modelled portions to be shown on separate lines in the report; if they cannot be separated, lower the precision of any decision made on that number by one notch.

How do I detect that the click identifier is being lost?

A manual test is the quickest route. Click your own ad, note the identifier in the address bar that opens, then browse the site through to the conversion step and check that the identifier is still being held. The common points of loss are redirect chains, the form or payment step moving to another domain, scripts unable to run before consent, and cache configurations that ignore the query parameter. In services with long decision cycles, the time between click and conversion exceeding the storage period also breaks the match.

How do I measure demand that arrives by telephone?

There are four routes and the choice depends on volume. Counting a tap on the number on your site as a conversion is the easiest, but it does not prove a conversation took place. Platforms have their own reporting for calls arriving via ad assets; because the scope can vary by country and account type, verify what is actually reported from your own dashboard's documentation. The dynamic number insertion feature of call tracking software ties the call to its source, in return for cost and responsibility for personal data. At small volumes, the person answering the phone asking and recording where the enquiry came from is often the most reliable method.

Is it right to use an automated bidding strategy while measurement is incomplete?

Usually not, because automated bidding does not correct an incomplete signal, it follows it. In an account whose forms are measured and whose phone calls are not, the system learns to move away from the audience that converts by telephone, and that does not show up in the dashboard as an error; only the structure of the demand arriving changes, quietly. The defensible order is to verify measurement first, then accumulate a steady flow of conversions, and only then move bidding to automatic. Because the result of strategy changes made while measurement is contested will be read with the same broken data, they teach you nothing.