Algeria AgencyDigital marketing
Running a Facebook campaign: the structure, the wait, one change
Campaigns break mostly from being managed too much. How to structure one, how long to leave it alone, and what to change when you change something.
Most Facebook campaigns that fail here fail neither for lack of budget nor for bad targeting. They fail because they are attended to too much: cut into too many pieces, looked at the next morning, and modified before the system has had enough to learn from.
The instinct is understandable. A campaign spends money continuously and displays figures in real time, which supplies both the worry and the means to act. The result is a campaign perpetually reset, which has never got past its first phase.
This article is about running one: the starting structure, the budget expressed in the right unit, the learning phase and what restarts it, the first seventy-two hours, reading numbers that lag, the single-change rule, creative fatigue, raising the budget, and the local calendar.
The article that accompanies this one is about what a campaign actually produces here — messages — the hour that follows, the comments under the ad, and cash on delivery. None of that is repeated here, and it is worth reading first: it describes what you are buying, this one describes how you drive it.
The account you inherit
A campaign almost always starts in an account that has been used before, and the first hour is not spent creating but taking inventory. What is still running, what has been paused for months, what is spending with nobody watching: all three exist in most accounts.
Start with the spend. Open the last thirty days at the level of the whole account rather than the campaign, and check that the total matches what you believed you were spending. The gap, when there is one, almost always comes from an old boosted post that was never stopped.
Then look at what is being measured. An inherited account often contains several conversion events defined at different times, some of which no longer fire, and a campaign optimised on a dead event spends without direction. An event that has recorded nothing for a month is dead until proven otherwise.
Finally check the saved audiences and their dates. An audience of people who wrote to you two years ago is not an audience, it is an old list. Interaction-based audiences have a window, and too long a window dilutes exactly what it was meant to concentrate.
Write the result of that hour down somewhere. That one page — what is running, what is measured, what is paused and why — is what will let you know in three months what you changed, and it is the thing nobody does because it produces nothing visible on the day.
Fewer campaigns than you think
The reflex is to split: a campaign per product, an ad set per city, another per age band, another per interest. It looks rigorous and it is the main cause of campaigns that never learn, because the budget divides into portions none of which reaches the threshold that lets the system work.
The useful rule is the opposite and fits in a sentence: as little splitting as possible, and each split must correspond to a decision you will actually take. If you are not going to cut a city’s budget because it costs more, that city does not need its own ad set.
For an advertiser starting out the sensible structure is almost always the same: one campaign, one broad ad set, several ads inside it. The ads are the only place where variety costs little, because they share the same learning instead of fragmenting it.
Splitting is justified later, and for two reasons only: when two audiences have very different values and you need to be able to spend more on one, or when an external constraint imposes it — stock available in a single wilaya, an offer that ends on a given date.
What never justifies a split is the need to know. You can read the breakdown by age, sex and region in the reports without having separated anything; separating in order to observe means paying for an extra learning phase to obtain a figure you were already given.
The budget: the number exists, but not in dinars
The question asked most often is "what budget does it take to make this work", and it has a precise answer that is not an amount. The threshold the system imposes is expressed in conversion events per week per ad set — a number the platform publishes in its own documentation and has already moved.
That is why we print no amount here, and it is not situational caution. Translating that threshold into dinars requires your cost per result, which depends on your product, your margin, your offer and what your competitors are paying at the same moment. A published amount is that conversion done at somebody else’s exchange rate: exact for them, false for you.
The multiplication is within your reach and takes a week. Launch with a modest budget, leave it alone, record your real cost per result, and multiply it by the number of weekly events the platform asks for. The product is your threshold, in dinars, for your business.
That calculation often produces an uncomfortable figure, and this is useful information rather than an obstacle. It means your budget will not run two ad sets, and therefore that the structure in the previous section was not an aesthetic preference: one properly fed ad set beats three that learn nothing.
If even one ad set cannot be fed, the honest conclusion is to optimise on a more frequent and shallower event — a message rather than a purchase — while watching the quality behind it. A system receiving enough signal on an approximate event works better than one receiving almost none on the right one.
The learning phase, and what restarts it
A campaign that starts goes through a period where its results mean nothing. The system tries combinations, spends unevenly, and displays costs that rise and fall with no relation to the quality of what you made. Judging during that period is judging an average calculated from three values.
The important point is elsewhere and is poorly known: certain modifications reset that period. Changing the targeting, the optimised event, the main creative, or the budget by a large amount restarts the learning, and the campaign returns to its statistical starting point.
That is the mechanism behind accounts where nothing ever works. Somebody modifies every two days, each modification restarts the period, and the campaign never leaves it. From outside it has been running for six months; in reality it has started again eighty times.
The practical consequence is a discipline rather than a technique: decide in advance what you allow yourself to change and how often, and write it down. A written rule holds against Tuesday morning’s anxiety; an unwritten good intention does not.
Note too that not all modifications are equal. Adding an ad to an existing set generally has no effect on the learning, while replacing the one receiving most of the delivery does. When you do not know, assume the modification resets: the error in that direction costs waiting, the error in the other costs the campaign.
The first seventy-two hours: touch nothing
The rule is blunt and that is what makes it usable: after launch, nothing is modified for three full days. Not the budget, not the targeting, not the ads. You may look — it costs nothing and it is reassuring — but you do not touch.
The reason is twofold. The first is the learning described above. The second is that day-one figures are systematically bad and systematically misleading: delivery begins with the people cheapest to reach, who are not the most interested, and the cost per result on day one rarely resembles that of week two.
There are two exceptions and they are technical, not commercial. You stop immediately a campaign whose ad contains a factual error — a wrong price, a wrong number, a date that has passed — and you stop immediately a campaign pointing at a page that does not work. In both cases you correct and relaunch; that is not management, it is repair.
What needs preparing during those three days is not in the ads manager but beside it: the person answering the messages, what they answer, and how fast. The article accompanying this one explains why that is the real bottleneck; the seventy-two hours are the right moment to deal with it, since you have nothing else to do.
Reading numbers that are behind
The figures displayed are not final, and the gap is wider than people think. A conversion is attributed to the ad that produced it, but the person may have seen the ad on Monday and written on Thursday — the result then appears in Monday’s column, several days after you read that column.
The consequence is that a figure recorded today for yesterday is incomplete, and will improve on its own. Plenty of advertisers have cut a campaign on a cost per result that, three days later, had become acceptable in a history they were no longer consulting.
The reading rule is therefore simple: never judge a period that has not been closed for at least the length of your attribution window. And always look at a rolling period rather than a day — seven rolling days show a trend, one day shows noise.
On top of that there is a local distortion here that the manager cannot correct: the conversation continues off the platform, the order is confirmed by telephone, and the refusal at the door is never reported back. The manager’s figure is therefore structurally optimistic, by a proportion only you can measure.
Measure it once and keep the ratio. If seventy-five of every hundred declared orders complete, your real cost per result is the manager’s divided by that rate. The ratio moves slowly, so recalculating it quarterly is enough — but without it every judgement you make rests on a number that is not yours.
One change at a time
When the observation period has closed and a decision is called for, the discipline is one rule: a single modification between two readings. Changing the visual and the targeting on the same day produces a result nobody will be able to explain, and therefore nobody will be able to reproduce.
It is counter-intuitive because modifications appear free: you can change everything in three minutes. But what costs is not the action, it is the observation period you then have to wait out. Two simultaneous changes cost the same time as one and return half the information.
The order matters as well. Start with what is most likely responsible and cheapest to undo — almost always the creative, then the offer itself, and only after that the targeting. Targeting is what advertisers modify first and it is rarely the problem, especially with a broad audience.
Keep a record, however minimal: one line per change, with the date, what was changed and why. Three months later that list is the only thing separating a managed account from a disturbed one, and it is also what lets you go back without guessing.
Creative fatigue
An ad that works stops working, and that is a due date rather than a failure. The same people see it several times, recognise it, stop pausing on it, and the cost per result rises slowly while nothing else has changed.
The signal reads before the cost rises, in the average frequency — the number of times one person has seen the ad. When it climbs steadily while the reach stalls, you are paying again for the same people, and fatigue is near even if the result still holds.
The answer is not to remake the whole campaign but to hold a reserve. Three or four creatives ready, produced at the same time as the first, allow a replacement without crossing a production period during which the campaign degrades. It is the only part of this work that gains from being done in advance and in a batch.
A variation is not a new creative. Changing the colour of a band or recropping the same photograph does not reset recognition; it takes another idea, another angle, another foreground. The question to ask is: will somebody who saw the previous one recognise this as the same ad?
Finally, a tired creative is not always deleted. An ad paused for several weeks often becomes usable again, because the audience has renewed and memory is short. Keep them and rotate rather than starting from nothing each time.
Raising the budget without breaking what works
The most dangerous moment in a campaign is when it works. The natural reaction is to triple the budget, and that is very often what breaks it: a large increase restarts the learning, and the campaign that was producing good results becomes a campaign that is searching again.
The method that holds is dull: increase in moderate steps, letting the campaign settle each time before the next step. You climb more slowly than you would like, but you climb, whereas an abrupt increase often takes you back down.
You also have to accept that a cost per result rises with the budget, and that is not a degradation of the work. A larger budget has to reach more people, therefore people less immediately interested. The question is never "has the cost risen" but "is the cost still below what the sale returns".
That is why a campaign’s ceiling is not an advertising figure but a management one: it is the margin on what you sell, less what handling the order costs. As long as the real cost per accepted order stays under that ceiling, climbing is rational; above it, climbing is an error paid for in volume.
The Algerian calendar
The cost of advertising is not stable across the year, and it varies here for local reasons foreign calendars do not describe. Ramadan moves browsing hours towards the evening and the night, changes what people buy, and raises advertising competition in the categories concerned.
The start of the school year, the two Eids and family celebration seasons produce the same effect in other sectors. In both cases the practical consequence is identical: the cost per result rises during the period for everybody, and a campaign judged on those weeks looks bad when it is simply expensive.
What is prepared in advance is ordinary and rarely done: decide before the period whether you are taking part, prepare the corresponding creatives, and set the budget knowingly rather than discovering the rise by absorbing it. Deciding not to take part is a valid decision, and far cheaper than going in unprepared.
Note the return to normal as well, which is the most favourable moment of the year and the one nobody exploits. The week after a busy period sees competition withdraw while the attention is still there, and it is often the best return of the year for an advertiser who kept some budget back.
The report you write for yourself
Once a month, half an hour, one page. This is not administration: it is the only way to know whether the decisions taken over the weeks produced anything, because no interface tells the story of your account.
Four figures are enough, and they are yours rather than the manager’s: the month’s total spend, the number of orders actually accepted, the resulting cost per accepted order, and the share of messages answered within the hour. The last is the only indicator on that list not found in the platform, and it is often the one that explains the other three.
Beside those figures, list the month’s changes — the list kept in section 6 — and nothing else. The document’s value comes from the juxtaposition: what you changed, and what happened. Without it you remember the changes that worked and forget the others.
Close with a single line: what you will test next month. One. A report ending in six intentions is a report none of which will be carried through to a reading, and next month you will find you changed everything and know nothing.
Keep these pages. Twelve of them are worth more than any dashboard, because they contain the one piece of information the platform does not retain: why you did what you did.
What we do, and what we refuse
What we do is bounded. We build the starting structure, we produce the reserve of creatives, we keep the list of changes and the monthly report, and we record with you the accepted-order rate that corrects the manager’s figures.
We refuse to run a campaign from an advertising account that belongs to us. The account, the page and the payment method are yours and we access them as partners — it takes longer to set up, and it is what makes the end of a collaboration leave you with an asset rather than a gap, along with the history that feeds your future campaigns.
We also refuse to modify more than one variable between two readings, including when asked to move faster. Two simultaneous changes cost the same wait and return half the information, and a campaign run at that pace is one where nobody will ever know what makes it work.
And we do not promise a cost per result, for the reason set out in section 3: that figure depends on your offer and on what your competitors are paying at the same moment, and an amount announced before measuring is an invention. What we commit to is the method and the delay after which you will have your own figure. There is also work you can do without us that weighs more than the advertising itself: shortening the time it takes to answer a message.
Frequently asked questions
How much do we need to invest for a Facebook campaign to work?
The threshold is not expressed in dinars but in conversion events per week — a figure the platform publishes and has moved before. Translating it requires your cost per result, which depends on your offer and your competitors. Launch small for a week without touching anything, record that cost, multiply it by the number of events required: that is your threshold, and nobody could have given it to you in advance.
Should we split campaigns by city or by product?
Only if you will genuinely take a different decision for each. Every split divides the budget and therefore the learning, and three underfed ad sets give a worse result than one properly fed. To observe the breakdown by region or age you need to split nothing: the reports already give it to you.
The campaign started badly. Can we fix it the next day?
No, except for a factual error in the ad or a broken landing page — in those two cases you stop, correct and relaunch. Otherwise day-one figures are misleading by construction and a modification resets the learning phase. Three full days without touching it is the rule, and it is what separates accounts that progress from accounts that permanently start again.
Our cost per message has risen over the last month. What should we do?
Look at the average frequency first. If it is climbing while the reach stalls, this is creative fatigue and not targeting: the same people see the same ad and no longer pause on it. Replace it with a genuinely different creative — another angle, not a recrop — and keep the old one, which will become usable again after a few weeks paused.
Can we triple the budget of a campaign that works?
That is the commonest way to break it: a large rise restarts the learning and the campaign that was producing results becomes one that is searching. Climb in moderate steps and let each step settle. Accept too that the cost rises with the budget — the only question is whether it stays under your margin per accepted order.
Are the ads manager’s figures reliable?
They are exact about what they measure and incomplete about what matters. They report with several days of lag, so a recent day is always understated; and they ignore what happens afterwards — the order confirmed by phone, the refusal at the door. Measure your accepted-order rate once, divide by it, and recalculate it each quarter.
Where we come in
A written rule holds against Tuesday morning nerves. What it does not anticipate is the day total spend does not match what you thought you had committed.
- We reconcile account-level spend against what you believed you had committed.
- We keep a dated trace of every change, and of who made it.
- We stay reachable on Tuesday mornings, which is when rules give way.
If your campaign is under two weeks old, do not call us yet: there is nothing to read, and looking early does more harm than good.
Read next
Facebook ads: what they actually produce, and the hour that follows
Here a Facebook campaign produces messages, not orders. And the cost per message means nothing if nobody answers.Reputation on Facebook: nothing is filed, everything is in threads
There is no listing to correct here. What gets said about you lives in comment threads attached to posts that scroll away.Who really owns your page, and what happens the day you are locked out
A page is an asset almost nobody knows the holder of. The access inventory, the roles, and what is never recovered.
Let us talk about your project
A free audit, no commitment: we look at your online presence and tell you what is holding it back.