At the end of August the owner of an air conditioner repair shop in Almaty is looking at the table by campaign: a completed order from Google Search cost 3,333 ₸ in ads, one from Instagram 7,500 ₸. The decision looks obvious: "In September, all 100,000 ₸ goes to Google." But 3,333 ₸ is the average cost of an order at 40,000 ₸ of spend. It says nothing about what orders will cost if you put two and a half times as much into search. The numbers in this lesson are made up for the example.
The average cost of an order and the cost of the next one
The average cost of an order is total spend divided by total orders. A decision about the next money needs a different number: added spend divided by added orders. Economists call it marginal cost; put simply, it is the cost of an added order. Within one campaign you find it through budget steps, and when there are several channels, months in which the budget was split between them differently come to the rescue.
The shop has two such months. In July the media buyer gave 80,000 ₸ to search and 20,000 ₸ to Instagram; in August he moved money into Instagram, where leads were cheaper. Both months ended with 20 completed orders, and each order leaves the shop 8,000 ₸ before ad costs.
| Google Search | Total | ||
|---|---|---|---|
| July: spend and orders | 80,000 ₸, 16 | 20,000 ₸, 4 | 100,000 ₸, 20 |
| August: spend and orders | 40,000 ₸, 12 | 60,000 ₸, 8 | 100,000 ₸, 20 |
| Average cost per order in July | 5,000 ₸ | 5,000 ₸ | 5,000 ₸ |
| Average cost per order in August | 3,333 ₸ | 7,500 ₸ | 5,000 ₸ |
| Cost of an added order | 40,000 / 4 = 10,000 ₸ | 40,000 / 4 = 10,000 ₸ | - |
In both channels, the 40,000 ₸ above the lower level brought four more orders, at 10,000 ₸ each. An order leaves 8,000 ₸, so the top 40,000 ₸ in either channel returned 32,000 ₸ of contribution and lost 8,000 ₸. The averages hide this: in July both channels delivered orders at 5,000 ₸, well below the contribution.
In Instagram, part of that 10,000 ₸ is not the ceiling but the ad. In August only 40 of the 160 leads from there were a fit: the rest wanted a new air conditioner or lived outside the city, beyond the technician's area, so the plan rewrites the ad.
A four-order difference between months happens without any budget change: one or two cancelled jobs and a couple of hot days add up to the same. So 10,000 ₸ is a suspicion, not a price. The totals, though, are clear: moving 40,000 ₸ from search to Instagram did not change the total number of orders, 20 and 20. The top money seems to have worked at around zero, and a growth step is needed to check it. One more caveat: an order is credited to a channel by the last click. Someone who saw the ad on Instagram could later find the shop on Google, and their order went to search.
Now the owner's plan. In July, at 80,000 ₸, the search campaign lost only 3% of impressions to budget, which means at its current bids it was already showing for almost every search available to it. Another 20,000 ₸ would go into pricier clicks or weaker queries, so expect 16-18 orders from search. Instagram drops out along with its four July orders, and perhaps with some of the orders credited to search. Even at August demand the plan gives no more than 16-18 orders and 28,000-44,000 ₸ after ads instead of 60,000 ₸.
The picture asks the same question with another business's round numbers: an extra 100,000 ₸ is judged by the customers it will add, and that number is uncertain.
Where a channel tops out
The closer the budget gets to a channel's limit, the more the next order costs. The shop has three such limits.
The first is the number of people searching for air conditioner repair. Ads can capture a large share of them, but search demand has a ceiling. How close a campaign is to it shows in the impression share columns in Google Ads. If Search lost IS (budget) is close to zero, the budget is no longer holding impressions back, and new money goes into higher bids, where every click costs more, or into new keywords, further and further from "air conditioner repair". For example, into "buy an air conditioner", and the shop doesn't sell new units.
The second limit is in Instagram. Even if you show ads only in the districts you serve, there is a finite number of people there, and Meta's system takes the cheapest delivery opportunities first and then moves on to more expensive ones. For the Highest volume strategy Meta says it plainly: costs can rise as you increase budget.
The third ceiling is the business itself: leads beyond what the technicians can handle in a day are paid for, but there is nobody to do the work. That is why you find out the business's capacity before setting the budget.
Base, growth step and test
It helps to split a monthly budget into three parts with different rules.
The base is the channels that already pay off, at the level where they paid off with a margin to spare. For the shop that is 40,000 ₸ in search (12 orders in August) and 20,000 ₸ in Instagram (4 orders in July). If the months are similar, together that is about 16 orders, 128,000 ₸ of contribution and 68,000 ₸ after ads. More than the shop made in July or August with 100,000 ₸.
A growth step is an addition to a proven channel to find out where an order becomes more expensive than its contribution. 10,000 ₸ is the average over the whole 40,000 ₸ between the August and July search budgets, and the first 20,000 ₸ of it may well pay off. So the shop raises search from 40,000 to 60,000 ₸ and writes the rule in advance: the addition must bring at least three orders on top of the twelve (20,000 / 8,000 = 2.5), otherwise the search budget goes back to 40,000 ₸. Even over a month, three orders sit at the edge of normal variation. The rule protects against a loss rather than measuring the price: if the step brought one or two orders above twelve, it is held for another month. A step never runs shorter than a month: in two weeks search brings the shop six or seven orders.
Finally, a test is money for a question you don't have an answer to yet: a new channel, a new offer, a new audience. The shop tries 2GIS: in the channel selection table it is the second channel for summer repairs. You size a test from the answer you need. With 8,000 ₸ of contribution, a 20,000 ₸ test pays for itself from three orders. A test like that separates "doesn't work at all" from "worth another month", but two or three orders won't tell you the real cost of an order in 2GIS. The rule is written before launch: three orders or more means another month at the same amount, zero or one means stop, two means you decide by how many of the 2GIS leads were a fit.
2GIS catches the same people as search, so the step rule looks at search and 2GIS orders together. Otherwise 2GIS may simply take some orders away from search, and the step will look worse than it is. Together they must bring six orders more than the search base: three for the step and three for the test, that is 18 at August demand. So the extra 40,000 ₸, step and test together, is judged as one block, and this rule replaces the separate step rule above.
There is no ready-made split like 70/20/10 here, and splitting evenly doesn't work either: the base is limited by the cost of the next order, the test by the answer you need.
How to split a monthly budget
The shop's budget for September. The heat is easing off, and search will bring less than in August. Compare against impressions: if impressions for the same queries are down by a third, expect about eight orders from the search base, not twelve. The thresholds in the table are set at August demand and move with the base: with eight base orders, search and 2GIS together need 14, not 18.
| Part | Channel | Amount | Based on | What we decide at month end |
|---|---|---|---|---|
| Base and growth step | Google Search | 60,000 ₸ | 12 orders on 40,000 ₸ in August | Search and 2GIS together: 18 orders or more means keep 60,000 ₸, 16-17 means another month, fewer means back to 40,000 ₸ |
| Base | Instagram, ad rewritten | 20,000 ₸ | 4 orders at 5,000 ₸ in July | How many orders and how many leads were a fit |
| Test | 2GIS | 20,000 ₸ | Second channel for summer repairs | 3 orders or more: another month; 0-1: stop; 2: decide by how many leads were a fit |
| Total | 100,000 ₸ |
If the step and the test bring no orders at all, at August demand that comes to 16 orders and 28,000 ₸ after ads, the same as the worst case of the owner's plan, while the base alone at 60,000 ₸ would have left about 68,000 ₸. The step and the test are the price of knowing, a month from now, whether search has room to grow and whether 2GIS works. Whether to pay it is the owner's call; your job is to show him both options in numbers.
For any business the order is the same; the only thing that changes is what you already know about the channels:
| What you know about the channels | Base | Growth step | Test |
|---|---|---|---|
| No channel tested yet | None | None | One channel, as in the Meta lead test |
| One channel pays off | That channel at the level where it paid off with a margin | A step in the same channel | A second channel |
| Two or more channels pay off | Each channel at the level where it paid off with a margin | The channel with the cheapest added order | A new channel or a new offer |
Channels are compared by contribution, not by revenue and not by leads: in the online school's report, Meta and Google each brought 150,000 ₸ of revenue, but after ads Meta was 30,000 ₸ in the red while Google stayed in the black. So before you move money, compare three things for each channel: how much it leaves after ads, what the next order will cost, and where the channel tops out. Last month's average result does not promise the same from the next slice of budget.
Large advertisers get the "budget vs orders" curve from a marketing mix model: it shows where each channel hits saturation, and Google's Meridian has a scenario that allocates budget to a target marginal ROI. A shop with twenty orders a month has nothing to build one on; two months and one step are enough.
Exercise: the cost of the next order in your project
Take a client project, your own test or a friend's business that has at least one ad channel.
The exercise is done when every budget line has an amount and a rule for the end of the month. If you divided total spend by total orders, you got the average cost, not the cost of an added order. Say spend went up by 30,000 ₸ and you got two more orders. That is 15,000 ₸ per added order, and with 8,000 ₸ of contribution that part of the budget is losing money, however good the average looks. If something else changed between the periods (the creative, the price, the season), write it down next to the numbers: the difference in orders may not come from the budget. A test cheaper than two orders at break-even cost won't answer anything (for the shop that is 16,000 ₸), so it is better to postpone it until you have the money. And a growth step and a test in the same channel, or in two channels that catch the same demand like search and 2GIS, can't be told apart in one month: keep just one, or count their orders together, as the shop does.