Why automated strategies burn the budget, and how to stop it
Contents
Share
An automated strategy “burns” the budget more often because of Direct’s calendar than because of a single button. The weekly cap runs Monday through Sunday, the average price is held for that same calendar week, and changing the cap mid-week opens a new ceiling on top of money already spent. Click fraud enters this mechanic in one narrow way: conversions Direct has marked as fraudulent do not train the strategy. The rest of the week’s spend is explained by the budget, the average price and a restart.
The sequence below separates ordinary strategy behaviour from spend you should stop by hand. Figures from the help pages stay as published. The worked numbers at the end are an example, not a norm for your campaign.
What the strategy treats as learning
Click fraud and sklikivanie matter here as a boundary. Conversions recognised as fraudulent do not teach the strategy. Invalid clicks whose conversion already has that status do not feed bid selection for that conversion. Who clicks ads, and why does not add a new motive to learning: the antifraud note simply leaves recognised fraudulent conversions out of training.
“Maximum conversions” strategies with an average target, including average conversion price or a share of ad spend, set bids from a conversion forecast. The average is held for a calendar week. On single days it can sit above the target or below it. The help text names 10 conversions a week as the learning minimum, 70 for app promotion, and asks you to judge the result after 7–14 days.
Ten, seventy and a one-to-two-week window are learning conditions, not a line after which clicks become fraud. A campaign with six conversions this week is not “broken by bots”. It simply lacks the signal the strategy uses to hold an average.
A bid adjustment changes the target, not the report. At 200 ₽ per conversion and a +50% adjustment on a group, that group’s average conversion price is 300 ₽. If you need to stay on a set price, the help text suggests paying for conversions rather than for clicks.
Whether Yandex refunds click fraud is a different article. Stopping a fast week, below, is about the weekly cap, the restart and the average price.
The strategy price and the reported CPA are different numbers
You read the current conversion price in Report Wizard, using CR, CPA and “Conversions”. The price set on “Maximum conversions” limits future conversions. Reported CPA can land above that limit or below it.
A week in which reported CPA is higher than the strategy price does not prove the strategy “ran away because of bots”. The help text allows the gap in both directions. It asks you to judge the result after 7–14 days, and the average over the calendar week, not over Monday evening.
Write two numbers side by side: the limit in the strategy settings and the CPA in Report Wizard for the same calendar week. If they diverge on day three, that is not yet a reason to change the budget. A mid-week budget change starts a separate mechanic.
How much the strategy may spend in a day
For these strategies the budget help recommends a weekly budget of at least 10 × weekly margin per conversion for “Maximum profit”, and at least 10 × conversion price for “Maximum conversions”. The minimum weekly budget is 300 ₽.
The multiplier of ten is a recommendation so the strategy has something to work with. It is not a profitability line and not a sign of a burn. A budget of ten conversion prices can be profitable or loss-making; the budget help does not calculate that.
The maximum spend in a day is counted in Moscow time and depends on the schedule:
a one-day schedule — up to 100% of the weekly budget and any carry-over;
two days — up to 50%;
three or more days, or no schedule limit — up to 35%.
A campaign that runs all week should not take more than 35% of the weekly cap on an ordinary day. About a third on Wednesday is the help-text ceiling, not an accident. If more left the account and you changed the weekly budget or the payment model the day before, look at the restart, not at “bots in the bids”.
When a budget change opens a second cap
Direct spreads the weekly budget from Monday through Sunday and does not exceed it for the week, except on a restart. A restart happens in three cases: you change the weekly budget, you change the payment model, or you move the start date into the future.
Stopping and starting again, changing the strategy, changing the limit type, changing the click price or changing the conversion price do not restart spend. Switching the campaign off in the evening and on in the morning is not a new budget. Changing the conversion price is not a restart either. Changing the weekly amount is.
If you change the budget mid-week, money already spent is not subtracted from the new limit. On a click-payment restart day, up to 35% of the old budget plus 35% of the new one can leave. On conversion payment, up to 100% of the old budget plus 100% of the new one.
A help-text example, not a spending norm: the budget is 7,000 ₽, 4,000 ₽ is spent by Thursday, and the budget is changed to 5,000 ₽. From Thursday through Sunday you can spend up to another 5,000 ₽. For the calendar week the account can show more than the original 7,000 ₽, and that is the restart behaviour the help text describes.
Change the budget at the start of the calendar week. Then the old limit does not stack with the new one for the remaining days.
Underspend carries at most 30%, and it does not pile up
On click-payment strategies, underspend carries to the next week, but by no more than 30% of the weekly budget you set. The first week has no carry. Balances do not accumulate: only last week’s underspend is used. Carry is spent first, then the new week’s budget.
The week’s ceiling is up to 130% of the budget you set, and only because of last week’s underspend. Over several weeks the help text promises cumulative spend no higher than the sum of those weeks’ planned budgets. Carry does not retrain the strategy and does not restart spend.
Carry does not apply to the account’s shared balance. When total spend hits the account limit, impressions stop, even if the strategy still has a weekly remainder and a carry.
30% and 130% are the carry rules for click payment. For conversion payment the help text describes a restart instead, not this carry. Do not apply “up to 130%” to a campaign that pays for conversions.
One calendar week, in order
Walk the steps on Monday, and again on Thursday if spend suddenly speeds up. Each step answers one question.
Note the payment model: clicks or conversions. It decides both the daily ceiling and whether a 30% carry exists.
Write down the weekly budget and compare it with the recommendation: for “Maximum conversions”, at least 10 conversion prices and at least 300 ₽. A smaller budget means the strategy is outside the help-text recommendation. That is not fraud by itself.
Check the schedule. One show day — a daily ceiling up to 100% of the week and carry; two days — up to 50%; three or more, or no limit — up to 35%. Count the day in Moscow time.
Open Report Wizard and copy CR, CPA and “Conversions” for the current calendar week. Put the strategy price next to them. A gap in either direction is allowed.
Count conversions since Monday. For an average target the help text names 10 a week, 70 for apps. Fewer than that is a reason not to judge the strategy before 7–14 days, not a reason to declare a burn.
Check whether this week you changed the weekly amount, the payment model or the start date. If you did, money already spent is not inside the new limit. On the restart day, click payment can take two slices of 35%; conversion payment can take up to 100% of the old budget and 100% of the new one.
If you pay for clicks and last week was not the first, check underspend. Carry is at most 30% of the budget you set, older weeks do not stack, and carry is spent first. This week’s ceiling is up to 130%.
Separately, see whether the shared account balance is exhausted. A weekly carry does not raise it.
Fraudulent conversions are outside training. Heavy spend and too few conversions for learning are two plots: the antifraud filter and the weekly cap. Do not merge them into one cause until steps 1–8 are done.
Change the budget on Monday. The same edit on Thursday opens a second limit for the rest of the week.
Example: Thursday and a new limit
This is a teaching example using the help-text numbers, not a measurement of your campaign.
The campaign pays for clicks. The weekly budget is 7,000 ₽, the schedule has no limit, and an ordinary day is capped at 35%, or 2,450 ₽. By Thursday, 4,000 ₽ is spent. The budget is changed to 5,000 ₽. The 4,000 ₽ already spent is not subtracted from the new limit. From Thursday through Sunday you can spend up to another 5,000 ₽. On the restart day an extra slice of up to 35% of the old budget plus 35% of the new one is possible: 2,450 ₽ + 1,750 ₽.
If reported CPA that same Thursday is above the strategy price, that is a second, separate observation. The average has not closed the calendar week, and the help text asks for a judgement after 7–14 days.
Next week, if payment is still per click and this week was not the first, underspend carries by at most 30% of the budget you set. On a 5,000 ₽ budget the carry ceiling is 1,500 ₽, and only from last week. Carry goes first, then the new budget. Over several weeks the total should not exceed the sum of the planned budgets.
When the week still looks empty
After the steps, the fork is small.
If conversions are under 10 a week (or under 70 for app promotion), the strategy is not yet in the mode the help text calls learning. Changing the budget every day does not fill that gap: every change of the amount restarts spend.
If reported CPA stays above the limit and 7–14 days have already passed, the help text suggests holding the set price by paying for conversions rather than clicks. That is a payment-model change, and a model change itself causes a restart. Plan it on the edge of a calendar week and count, in advance, that a conversion-payment restart day can take up to 100% of the old budget plus 100% of the new one.
If the shared account is empty, impressions stop regardless of carry. Topping up the account and turning the weekly limit are different actions.
Reading one week by hand does not return money already charged and does not replace Direct’s antifraud. It only keeps you from treating an ordinary restart and an ordinary carry as “a burn that must be killed today”.
What to compare before you touch the limit
One number in the account almost always mixes two layers. The strategy layer is the limit on a future price, the weekly budget and the daily ceiling from the schedule. The report layer is CR, CPA and conversions for days that have already passed. Keep them in one table, even if the table is a notebook.
| What you read | Where it lives | How to read it |
|---|---|---|
| Price on “Maximum conversions” | Strategy settings | A limit on future conversions |
| CPA, CR, “Conversions” | Report Wizard | The current price; it may be above or below the limit |
| 10 conversions a week, 70 for apps | Condition for an average target | Not a fraud line; judgement is asked after 7–14 days |
| 35% / 50% / 100% of a day | Schedule and Moscow days | Ceiling of an ordinary day, with no restart |
| Restart | Change of budget, payment model or start date | Money already spent is not subtracted from the new limit |
| Carry up to 30% | Click payment, not the first week | Ceiling up to 130%; it does not apply to conversion payment |
| Fraudulent conversions | Direct antifraud | They do not train the strategy |
If the restart row is empty and a day still passed 35% on a wide schedule, look for a cause outside this mechanic. If the restart row is filled, recount the day as “old slice plus new slice” before you decide the campaign is being clicked fraudulently.
Bid adjustments belong in the table as one line: +50% on a group with a 200 ₽ target lifts that group’s average conversion price to 300 ₽. The account does not have to show a 300 ₽ CPA today. The average is still being collected through Sunday.