How to estimate budget lost to click fraud

The loss from click fraud is not the “invalid clicks” row. The Yandex Direct guide says you are not charged for that traffic, and the spend is returned to the balance. The loss you can actually estimate sits in a different number: spend that was charged and had no buyer behind it.

What counts as that kind of click is set in what ad click fraud is. This piece is the formula. The share in it is your estimate. The account does not print it as a ready-made percent.

The formula for what stayed in spend

Let $S$ be charged spend for the period, and $p$ the share of that spend behind which you see no buyer. The loss estimate is:

Suppose campaign spend for a month is 90,000 ₽, and you judge that 15% of that money did not go to buyers. The loss estimate is 13,500 ₽. Beside it, the “Invalid traffic” report may show that the system already returned another 20,000 ₽. Those 20,000 ₽ are not added to the 13,500 ₽. They never entered spend. A sum of 33,500 ₽ would count the same trouble twice.

Both figures, 15% and 20,000 ₽, are an example in this paragraph. Fifteen percent is not an industry norm. Twenty thousand is not a typical refund. Use your own.

If the empty clicks cost the same as the rest, the same estimate comes from the number of charged clicks $N$, the average cost per click $c$, and the share of clicks $p$:

The formulas diverge when empty clicks are more expensive or cheaper than the average. Then it is more honest to apply the share to the spend of a specific placement or campaign, not to the average CPC of the whole account. The calculator below uses the second formula: paid clicks times the cost per click times the share of empty clicks. You set all three numbers. They are not every session on the site, and they are not a share Direct printed.

On the starting numbers that is 1,000 clicks at 40 ₽ and a 20% share: a monthly loss of 8,000 ₽. The share slider is not a statistic. It is the estimate you brought with you.

Where the share comes from if the account has none

There is no ready-made percent, and it should not be borrowed from someone else’s survey. The workable move is narrower. Do not hang one share on the whole account.

Open spend by campaign and by placement. Where clicks are in spend and there is no target action, calculate $S$ for that row only. For that row, $p$ can be close to one, if the other reasons for zero leads are already ruled out: the goal records, the page opens, the offer does not explain the emptiness. How to walk those reasons before the estimate is covered in signs of click fraud. A row that lives only in Webvisor and is absent from the “Direct” report group does not enter $S$. You were not charged for it.

Cost per lead is the next check, so you do not confuse this loss with a different rise. Empty spend increases CPA, because the leads are the same and the numerator is fatter. If CPA rose because the goal undercounted leads, the loss formula will not see it: the lead count is not inside $L$. Separating those two rises is in how to calculate cost per lead.

Click-through suggests where to look, and it is not itself the share. A high CTR with empty leads is a reason to open the placement, not a multiplier for $p$. How not to treat someone else’s CTR as a norm is in what CTR is.

A visit check on the site does not print the true share. It classes a visit as suspicious or not from behavior, the browser environment, the network and the ad-click parameters, and it can show which campaign those visits bunch up in. That share still must not be added to invalid clicks already returned. Direct removed some of them earlier. You can look at the classification without closing the door on visitors: the check is counted, the person gets through. When a decision is turned on, the visit is let through, the browser is asked to solve a short task, or the visit is refused. If the check does not answer, the visit is let through, and it does not enter that hour’s loss because of an outage.