Why “how much does a month cost” is the wrong question
Advertising is sold not by the month but by the click at auction. The same budget buys different things in different niches: where a click costs 30 rubles, thirty thousand gets you a thousand visits; where it costs 600 rubles, fifty. In the first case you will have an answer by the end of the month; in the second, a collection of random events.
So you count from the other end: how many clicks you need for a conclusion to be a conclusion, multiplied by the cost per click. The resulting amount is the price of answering the question “does Direct work for me”.
How many clicks you need for a conclusion
Suppose your landing page converts at three percent: three out of a hundred people leave an inquiry. A typical level for services.
- 20 clicks. 0.6 leads expected. The probability of getting none at all is about 54%. So more often than half the time, perfectly fine ads will show zero. No conclusion can be drawn at this volume.
- 100 clicks. Three leads expected. A zero here already means something: at three percent it happens in about five cases out of a hundred. This is the first point where a negative result becomes information.
- 300 clicks. Nine leads expected. Now you can not only say “works or not” but also estimate the cost per lead.
Hence the rule: a hundred clicks per landing page is the minimum for a “yes or no” answer, three hundred for a “how much” answer.
A twenty-click test is no cheaper than a hundred-click test. It is simply worthless and tells you nothing — the money is spent and there is no answer.
Where to get the cost per click without spending a ruble
Direct has a “Budget forecast” tool — it is available before you launch campaigns and without topping up your account. Enter key phrases, choose a region, and it will show an estimate of the cost per click and traffic volume. Two caveats: it is a forecast, not a fact (a new campaign usually pays more), and it calculates from bids, not from your niche as a whole — good enough for an order of magnitude, not for planning down to the ruble.
That is enough to check the order of magnitude. If the forecast says 40 rubles per click and you need 300 clicks, the test costs about 12 thousand. If it says 500 rubles, the test costs 150 thousand, and that is a different conversation, better had before launch than in the middle of the month.
The full calculation, in numbers
Take a service with an average order value of 40,000 RUB and a forecast cost per click of 60 RUB. The cost rows are the number of clicks from the row above multiplied by 60 RUB; cost per lead is spend divided by the number of leads.
| Step | How many |
|---|---|
| Clicks for “yes/no” | 100 |
| That costs | 6 000 ₽ |
| Clicks for cost per lead | 300 |
| That costs | 18 000 ₽ |
| Leads at 3% | 9 |
| Cost per lead | 2 000 ₽ |
The last row is what the whole calculation was for. Two thousand per lead with a forty-thousand order value is a conversation about what share of leads turn into deals. One in five — a customer costs ten thousand and the ads pay off. One in twenty — they don’t, and the work needed is not on the ads but on sales.
All the numbers in the table except the cost per click are arithmetic, not our data. Plug in your own and get your own answer.
Three ways to waste a test budget
Splitting it across five campaigns. Twenty clicks each means five tests, none of which will show anything. It is better to test one direction all the way than five halfway.
Stopping after three days. Demand in most businesses follows a weekly cycle: weekdays and weekends behave differently. Less than two full weeks measures part of a week, not demand.
Changing settings every day. Every change resets the accumulated statistics and restarts the learning of automated bidding strategies. A test is a period when you don’t touch the campaign, you watch it.
What a test will never show
Seasonality. One month doesn’t distinguish your niche from your month. Testing in July a business that peaks in September, a negative result may be a property of July.
Customer quality. A lead and a deal are different events, with weeks between them. A test gives you the cost per lead, not the cost per customer.
The ceiling. Cost per lead over the first hundred clicks is almost always worse than three months later: the campaign hasn’t learned, the ads haven’t been weeded out, the negative keywords haven’t been collected. A poor test result is the lower bound, not a verdict.
If after the calculation the test costs more than you are prepared to spend, that is a result too: it means Direct is not the first channel for your niche. Our page on contextual ads and paid social describes how we calculate this before launch, not after.