How much you lose to click fraud: a calculation for your own campaign
Contents
Share
Your campaign’s loss is not someone else’s percentage times your spend. The fourth Bot Baseline report from the ANA association and White Ops forecasts 5.8 billion dollars globally for 2019, for video, display and other CPM formats. The TAG report for the United States in the first half of 2023 gives an IVT share of 0.82% in channels where the whole TAG chain is certified, and 1.19% in uncertified ones. Neither is Yandex Direct spend, and neither is search.
How to calculate budget loss from click fraud is where you calculate from your own clicks. Here is the order that keeps 5.8 billion, 20–35% or 0.82% out of that sum. Separately: Direct returns spend on invalid traffic it has already recognised to the shared account balance. That is not an estimate of what remains in spend.
What the ANA and White Ops report measured
The measurement ran from 1 August to 30 September 2018: 50 ANA member companies, 2,400 campaigns, 27 billion impressions. Search and paid social were not in the study. Losses from fraud that got through and was paid, the authors forecast at 5.8 billion dollars globally for 2019, for video, display and other CPM formats on desktop and mobile. They estimate fraud attempts at 20–35% of impressions over the year, and the paid remainder as less. The data is from the vendor White Ops. The ANA press release is dated 1 May 2019.
The 2019 forecast is built from the August–September 2018 measurement. It is not Yandex Direct statistics and not a share of Russian search. Multiplying a week of Direct search spend by 20–35% takes an attempt rate on CPM impressions that did not include search, and calls it your money. The authors already separate the paid remainder from attempts: it is smaller.
What the TAG benchmark measured
The TAG US Fraud Benchmark for 1 January to 30 June 2023 was collected by The 614 Group from agency holding companies and their MRC-accredited measurers. There was no sample: every impression passed in was reviewed, 871.9 billion, for display and video on desktop, in the mobile web, in apps and in CTV, in the United States. The IVT share in channels where the whole TAG chain is certified is 0.82%. In uncertified channels it is 1.19%. IVT in the report is GIVT plus SIVT, by the classification of the measurers DoubleVerify, IAS and Moat.
0.82% belongs to the certified channels of that sample, not to “all internet advertising”. It is not Yandex Direct spend and not search. You cannot label your CTR or your CPA with these shares: CTR and CPA are calculated from your clicks, impressions and goals. Metrica robotness is not this benchmark’s IVT either: robotness is a different mark, on different data.
What is already not a loss
Direct returns spend on invalid traffic to the shared account balance. If the report has already recognised the traffic, the matching spend is a refund, not a hole to multiply by an industry percentage. The refund does not estimate the remainder. The remainder, if you calculate it, is calculated separately and labelled as your calculation, not as a figure from ANA or TAG.
Order for your own campaign
Write down the period and the system: Direct, Google Ads or an impression network. Put a foreign report only next to its limits.
If the system is Direct search, do not take 5.8 billion, 20–35% or 0.82%. Search was not in the ANA report. 0.82% is certified display and video channels in the United States for the first half of 2023.
Separate invalid traffic already recognised. Its spend was or is returned to the shared account. Do not put that sum into “losses still sitting in the budget”.
Do not use the refund as an estimate of the remainder. The remainder needs your own denominator: your clicks and your spend for the period. The method is in the budget-loss article.
If you still want to show a benchmark to a manager, label the sample in full: dates, what was included, what was not, whose vendor or measurer it is.
Calculate CTR, CPA and robotness from your own reports. Do not replace them with IVT.
Close with one line: “the refund of what was recognised is on the shared account; the industry percentage was not applied to our spend; we calculate the remainder separately”.
An example you cannot file as a loss
A teaching example of how a figure comes out wrong.
A week of Direct search spend is 100,000 ₽. Someone multiplies by 0.82% and gets 820 ₽. That is not the campaign’s loss. 0.82% is the IVT share in TAG-certified channels of the display and video sample in the United States, not of this search.
The same spend times 20% is 20,000 ₽. The 20–35% in the ANA report is fraud attempts on impressions over the year, not the paid remainder, and not search. The authors call the paid remainder smaller. 20,000 ₽ does not go into a Direct loss report.
Of the same 100,000 ₽, suppose 4,000 ₽ was already recognised as invalid and returned to the shared account. Those 4,000 ₽ are not added to “we will still lose”. The refund has already taken them out of spend. How much remains beyond them, this example does not calculate: the refund does not measure the remainder, and foreign percentages are not applied to it.
How to show a benchmark without replacing your account
| Figure | Where it comes from | Where it must not go |
|---|---|---|
| 5.8 billion dollars for 2019 | ANA / White Ops forecast of paid fraud on CPM formats | The cell “losses of our Direct” |
| 20–35% of impressions | Attempts over the year in the same report; the paid remainder is smaller | As a share of your spend |
| 27 billion impressions, 2,400 campaigns, 50 companies | The sample of 1 August–30 September 2018, without search and paid social | As the reach of your week |
| 0.82% and 1.19% | IVT = GIVT + SIVT, United States, first half of 2023, display and video | As search CPA or CTR |
| 871.9 billion impressions | Every impression passed in for that sample, not a subsample | As the denominator of your account |
| Refund to the shared account | Invalid traffic Direct has already recognised | As an estimate of the remainder |
The table holds the sample limits. It is not a calculator. Your spend enters it only in the teaching example above, and only to show a multiplication that should not be filed.
Checklist before a figure enters the report
It says whether this is search or CPM, Direct or another system.
5.8 billion does not sit next to your rubles without the label “global forecast for 2019 from a 2018 measurement”.
20–35% is named as attempts on impressions, not as a paid share of your budget.
0.82% is limited to the certified TAG chain in that sample.
1.19% is not called “average fraud on the internet”.
Invalid traffic already recognised is taken out to the refund on the shared account.
That refund is not used as a measure of the remainder.
CTR, CPA and robotness come from your own reports.
The closing line does not contain the product “our spend × someone else’s percentage”.
While the product is there, the report tells someone else’s sample. Your losses start where the product is crossed out and what remains is your period, your refund and a separate calculation of the remainder.
How the figure sounds in a meeting
A manager asks for one number. You can give it only with three tails, or the number lies.
The first tail is the system. “This is a week of Direct search” is incompatible with “this is 0.82% TAG”. If both are in one sentence, cut the sentence. Search was not in the TAG benchmark. US display and video are not your search spend.
The second tail is what was done with traffic already recognised. “4,000 ₽ from the teaching example came back to the shared account” is a finished sentence about a refund. “We lost 4,000 ₽ and will lose another 4,000 ₽” is double counting. The refund does not measure the remainder, but the refund itself is not put into losses a second time.
The third tail is attempts versus paid. The 20–35% in the ANA report is attempts on impressions over the year. The authors estimate paid fraud separately and as smaller, up to the forecast of 5.8 billion dollars for 2019 on CPM formats. In a meeting you cannot say “we have 20%, like ANA” and silently turn attempts into rubles of spend. Even inside that report, attempts and the paid remainder are not the same thing. Outside it, on search that was not in the measurement, even less so.
If a tail is not spoken, the number does not go on the slide. An empty cell is more honest than 820 ₽ from a multiplication that should not be filed.
Two more recalculations that should not be filed
Both are teaching examples. Neither is for the report.
Someone divides 5.8 billion dollars by the 50 ANA companies and gets “average loss per participant”. In the report, 50 companies is the size of the August–September 2018 measurement, with 2,400 campaigns and 27 billion impressions. The 5.8 billion forecast is global for 2019, for video, display and other CPM, not a quotient of those 50. The division does not come out of the press release, and it does not carry over to your campaign.
Someone takes 1.19% as “ordinary fraud without a certificate” and multiplies Direct spend with the note “we are not in TAG”. 1.19% is the uncertified channels of that same US display and video sample for the half-year, IVT as GIVT plus SIVT at DoubleVerify, IAS and Moat. The absence of a TAG certificate on your account does not turn search into that sample.
It helps to keep the denominator of 871.9 billion impressions beside this: there was no sample, every impression passed in was reviewed. Your thousands of clicks are not in that denominator and do not “refine” a share taken from it. A refinement would be a new study, and this article does not contain one.
After the crossings-out, your own report looks modest: the period, the spend, the refund of what was recognised on the shared account, the note “the remainder is not measured by this refund”, a link to a loss calculation from your own clicks, and your own CTR, CPA and robotness with no substitution of IVT. That modesty is enough to keep 2018–2023 from being defended as if it were the current week of the account.
The dates matter on their own. The ANA measurement ends on 30 September 2018, the forecast refers to 2019, and the press release came out on 1 May 2019. The TAG benchmark covers January–June 2023. Your current week is none of those windows. Even if the formats matched, carrying a percentage across years without a new measurement would be a new claim. The formats do not match: there it is impression CPM and video, and Direct search is clicks those reports did not take into the sample. That is why the checklist forbids the product. Remove the product, and the report is about your period again. Leave it, and the report is about someone else’s year.