Algeria AgencyDigital marketing
Trialling Snapchat: the stopping rule is written before the first dinar
On a platform whose figures will settle nothing, the only honest method is to decide in advance what would make you stop.
The article that accompanies this one exists to decide whether to go at all, and concludes that for most Algerian businesses the answer is no. This one addresses the minority who answered yes.
It does not resemble the other steering articles in this series, and that is deliberate. On Facebook, on Google or on Instagram the numbers eventually arbitrate: you read a cost per result, you compare, you decide. Here they will not.
The platform reports views and interactions rather than customers; most of what happens on it happens privately, out of reach of any dashboard; and nothing you produce there stays online. A trial launched without a stopping rule therefore never stops — it is forgotten.
The method fits in one sentence: write down before you start what would make you say no, then hold to it. Everything else in this article is only the business of making that sentence applicable.
The platform will not decide for you
On the four other platforms in this pillar the discussion always ends in the same object: a table. You read a spend, a number of results, a cost per result, and the decision very nearly makes itself.
That table exists here too, and it is misleading. It counts views, swipes and opens — gestures rather than intentions. A swipe up is cheap and says almost nothing about the person who made it.
To which is added the particularity the companion article develops: the platform is private. What becomes of your advertising after it is seen — a screenshot sent to three friends, a conversation, a visit three days later — reports back nowhere.
The practical result is that at the end of a quarter you will have abundant figures and no answer. That is precisely the situation in which a budget extends indefinitely: not out of conviction, but because nothing ever said to stop.
Hence the inversion proposed here. Since the instrument will not conclude, the conclusion is written in advance. This is not distrust of the platform; it is the direct consequence of what it measures.
One question, written before you begin
A trial that sets out to learn whether "it works" cannot end, because "it works" has no definition. A trial that sets out to learn one precise thing ends on its own.
The question has to be phrased so that the answer is yes or no. "Do customers under twenty-five arrive mentioning the agreed word?" can be answered. "Is this platform of interest to us?" cannot.
One question, not three. A trial with three questions produces three half-answers that contradict each other, and the final decision is then taken on whichever one suits.
Write it on the quotation, in the minutes, or simply on the first page of the notebook. What matters is that it is legible in three months to somebody other than you — including you three months later, who is not quite the same person.
One usable as it stands: "Over ninety days, at least twenty people arrive mentioning the agreed word." It contains a threshold, a duration and an instrument. All three are necessary, and one is almost always missing.
The budget is a ceiling you accept losing
The difference between a budget and a ceiling is not accounting, it is psychological, and it is what decides the end of the trial.
A budget is a sum whose return is expected. Exhausted without one, the natural reaction is to extend it "to see", because stopping would mean recording a loss.
A ceiling is a sum you decided in advance could be lost entirely without consequence for the business. Exhausted, it is not extended: it was the price of the answer, and you have the answer.
The right amount is therefore the one you can lose without "should we continue?" becoming a financial question. At most of the shops that consult us that is a modest sum, and the modesty is a property of the method rather than excessive caution.
One practical note: include production in the ceiling. Distribution often costs less than the time spent making what is distributed, and a trial that counts only the distribution understates its real cost by a margin that is anything but negligible.
Ninety days, and why neither thirty nor a year
Thirty days is not enough here, for a reason that is not statistical. On a platform where private word of mouth does the work, the effect arrives delayed: the person sees, keeps, talks about it, comes back later.
A year is too long, for the opposite reason. Beyond a quarter nobody remembers what we wanted to know, the person who was handling it has changed role, and the trial has become an ordinary line of expenditure renewed without being read.
Ninety days also covers a complete commercial cycle for most of the activities concerned, which avoids concluding on an exceptional period. A trial falling entirely within Ramadan or entirely within the return to school measures the season as much as the platform.
A useful clarification, because another article in this series proposes six weeks: that is a different object. Six weeks is enough to eliminate one cause on a page you modify yourself, where the effect is immediate and the channel already known. Here a whole channel is being tried, its effect delayed and its instrument absent.
Fix both dates in a shared calendar on the same day you decide to run the trial: the start date and the decision date. The second is the more important of the two, and it is the one nobody ever writes down.
What has to exist before the first dinar
Four things, and none of them optional: an advertising account that belongs to you, a named person, a single destination, and a way of recognising somebody who came from there.
The account has to be created in the name of the business, with a business email address. That is what the article on the keys to your page develops in detail, and it is settled at the moment of creation: afterwards it repairs badly, and sometimes not at all.
The named person is not an organisational detail. The companion article says it plainly: the fifth platform is not paid for in budget but in attention, and attention is already the scarce resource. If nobody is named, the trial will be run by whoever has time, which means intermittently.
The single destination avoids the commonest error, which is to send people to the homepage. A trial needs one door so that what passes through it is attributable. A page, a number, a physical address — but one.
The means of recognition is the subject of the next section, and it is the one most trials omit. Without it the other three preparations are worthless: you will know that people came, and never know from where.
The agreed word is your only instrument
The principle is old and has no better substitute here: the advertising contains a word, a phrase or an offer found nowhere else, and you count the people who say it.
What it measures is narrow and honest. It does not count everybody who came because of the platform — many will forget to say so — but it never counts anybody else. It is a floor rather than a total, and a floor is exactly what a stopping rule needs.
It has to be easy to say and impossible to confuse. An alphanumeric code reads badly, is remembered worse, and turns into three variants at the till; a short, slightly incongruous phrase remembers itself.
The instruction to the team fits on one line and has to be given before the start, not after three weeks: ask every new customer how they heard of the shop, and note the word when it comes. One column in a notebook is enough.
The word must be published there and nowhere else. Used at the same time in the window or on another platform, it stops measuring anything at all, and the trial finds itself with no instrument whatever without anybody noticing.
The geofilter is trialled in one evening
The format particular to this platform — the one where your customers distribute your advertising — is not trialled the same way, because it is not a campaign but an event.
It attaches to a place and a duration: a perimeter drawn on a map, a band of hours, and a design that has to stay legible sitting on somebody else’s photograph.
The corresponding trial is therefore a single unit: one evening, one wedding, one opening, one match. One occasion, a cost known in advance, and a question that answers itself the following morning rather than after a quarter.
What is measured is not a click. It is the number of people who used it, which the platform reports, and above all what you observe yourself: did anybody mention it to you, did you see those faces again.
Two drawing notes that save the trial. The filter has to work small and at an angle, because it will sit on an image whose framing and light you do not choose. And it should name the place rather than praise the business: people gladly circulate where they are, never your advertising.
Three readings, taken by hand
Three, not thirty. One at the start, one in the middle, one at the end, each on the same sheet and with the same columns.
The opening reading is taken before the first distribution, and it is the one systematically forgotten. Without it everything that follows is a figure with no reference: twenty visits a week means nothing if nobody knows how many there were the week before.
The middle reading, at forty-five days, has one function and it is internal policing: to check that the instrument works. If nobody has noted a single agreed word, the problem is not the platform but an instruction that was not followed — and there are forty-five days left to put it back.
The closing reading is taken on the date written in the calendar, not in the week somebody thinks of it. A measurement taken at a convenient moment is a chosen measurement, and it is worth nothing.
Three columns are enough: the date, the number of people who mentioned the word, and the cumulative spend. Everything else is available elsewhere and will not enter the decision.
What does not count as a result
This section exists because the end of a trial is the moment one becomes lenient, and because the platform supplies abundant material for leniency.
Views do not count. They measure paid distribution, which is what you bought rather than what you obtained. A trial that concludes on a number of views concludes only that it spent its money properly.
Screenshots and shares do not count either, flattering though they are: they signal that something was liked, and "being liked" was not the question written at the start.
The opinions of those around you do not count. The nephew who thinks it is very good, the supplier who says he saw it go past, the employee who feels things have changed — three sincere judgements, none of them related to a customer.
Only what the written question asked counts, read the way decided at the start. That discipline looks excessive until the day one sees it is the only thing separating a trial from a spend of mood.
The decision day is in the calendar
The decision is taken on a fixed date, in one sitting, with the sheet of three readings and the written question set beside it. It is not taken as one goes along.
Three outcomes, and only three. The threshold is met: you continue, and the trial becomes an ordinary budget with its own follow-up. The threshold is not met: you stop. The threshold is half met: you run the trial once more, with one thing changed and the same duration.
The third outcome degrades fastest, so it is limited to one repeat. A trial extended twice is no longer a trial; it is a subscription nobody has taken responsibility for.
The person who decides has to be the person who pays, and that is not obvious. In many shops the person who ran the trial has come to care about it, and asking them to conclude in favour of stopping is asking them to conclude that they wasted three months.
Write the decision and its date under the three readings, in two lines. It is the only document that will prevent, in eighteen months, running exactly the same trial having forgotten the first.
Stopping cleanly, and what remains
Stopping does not mean ceasing to pay and closing the tab. A badly stopped trial leaves behind an account nobody watches and a presence nobody maintains.
The advertising account is paused rather than deleted, with the payment method removed. A deleted account is hard to recreate; a paused account with a card still attached is a spend that can restart without any decision having been taken.
The public account of the business is treated like any other shopfront: either it is maintained or it is taken down. Left in place and abandoned, it tells whoever finds it that the business tried and gave up, which is exactly the reverse of the intended message.
Record the credentials in the same place as the others — that is the subject of the article on the keys to your page — including for an account no longer used. Today’s unused account is the one you will not be able to recover in two years.
What remains, finally, is the thing of most value and the one most often thrown away: the sheet of three readings and the two lines of decision. It is the only asset a failed trial produces, and it is worth its price the next time somebody proposes the same idea.
What we do, and what we refuse
What we do here is deliberately thin. We write the question and the threshold with you, we fix the two dates, we prepare the one-line instruction to the team, and we come back at forty-five days to check the instrument is turning. That is all, and it is invoiced as such.
We refuse to run a standing budget on this platform. Not on principle, but because we could not account for it: invoicing a monthly retainer on a channel we have just written is the weakest-measured in the pillar would be selling a report rather than a result.
We also refuse to state an Algerian audience figure for this platform, including when asked directly and when an estimate would do perfectly well in a meeting. No dated, public, verifiable series exists, and a figure invented here would end up quoted in a decision that is not ours.
And there is one thing you can do alone, this month, before calling us: ask your next twenty customers how they heard of you, and note the answers. If none of them names this platform and none is under twenty-five, the trial is already settled and it will have cost you a notebook.
Frequently asked questions
Ninety days — is that not too long for a trial?
It is long for a campaign and short for this channel. The effect arrives delayed, because it passes through a private conversation between the person who saw and the person who buys, and thirty days mostly measures your capacity to publish. The duration is only a cost if the ceiling is badly set: it is the amount that should be small, not the period.
How much should we set aside?
The sum you can lose entirely without the decision to continue becoming financial — production included, which is the item everybody forgets. We publish no amount, for the same reason the companion article publishes no audience: it would depend on your business and would be quoted as though it were a norm.
How do we choose the agreed word?
Short, sayable out loud by somebody walking into a shop, and slightly incongruous so that it sticks. Avoid alphanumeric codes: they turn into three different variants at the till. Above all, use it nowhere else — not in the window, not on another platform — or it stops measuring anything.
The platform gives figures. Why not use them?
Use them to check that distribution actually happened, which they do correctly. Do not use them to decide: they count gestures — views, swipes, opens — and the path to a customer runs here through private conversations no dashboard sees. They describe what you bought, not what you obtained.
What if the team forgets to note it down?
That is precisely what the forty-five-day reading exists to detect, and it is there for that and nothing else. Zero mentions at the halfway point almost always means the question is not being asked, not that the platform produces nothing. Put the instruction back and let the remaining forty-five days run; if nothing changes, the closing reading is honest.
Should we delete the account if we stop?
Pause the advertising account and remove the payment method: deleted, it recreates badly, and left active with a card attached it can restart without a decision. The public account is settled either way: maintained or taken down. An abandoned but visible account tells whoever finds it that you tried and gave up.
Where we come in
If the sentence will not come, the trial has no object, and that discovery is free. If it does come, what remains is holding the decision date when the numbers turn ambiguous.
- We put into one sentence, with you, what failure would look like, before any rate card.
- We put both dates into a calendar you share with us.
- We come back on the forty-fifth day and apply the threshold as written.
If you cannot fix the decision date in a shared calendar, do not open the trial: it will run on and stop being one.
Read next
Snapchat ads: who they serve, and who they do not
This article is a filter rather than an offer. For most Algerian businesses the honest answer here is no.LinkedIn ads: you are not buying an audience, you are buying names
Here the public that could buy from you can be counted. And what can be counted can be contacted, usually without paying.What advertising budget to set — and how to know
There is no published rate card for Algeria. Here is what the available data actually lets you conclude.
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.