Algeria AgencyDigital marketing
Analytics: five numbers, defined, every month
A useful report fits on one page and carries the same five numbers every month. The rest describes the agency’s activity, not yours.
This page closes the digital marketing pillar and it handles what arrives at the end of all the others: the report. It is the most produced object in this trade and the least read, and the two facts are connected.
The standard report runs to twelve pages, contains curves, heat maps, breakdowns by device and by city, and enables no decision at all. It describes the activity of whoever wrote it rather than the result of whoever receives it.
The useful report fits on one page, carries five numbers, the same ones every month, with their definitions written beside them. It is dull to produce and it is the only kind a director actually uses.
This article says which five numbers, what the tools count wrongly and in which direction, why attribution does not resolve in this market, and what the law asks before anything is measured at all.
What a report is for
A report does not exist to inform: it exists to enable a decision. That is the test to apply to every line, and it eliminates most of them.
Take any chart from your last report and ask: if this figure were twice as high, what would we do differently tomorrow? If the answer is "nothing", the line is decorative, however accurate it is.
The decisions actually available in a ten- to fifty-person business are few: spend more or less on a channel, write a page, change a price, answer faster, stop something. Five or six levers, and a report has to feed those.
That explains why long reports get read once: they contain information with no lever attached. The reader can do nothing with it, learns that this document asks for no action, and stops opening it.
The format constraint follows from the rest: one page, five numbers, one sentence of comment per number, and a recommendation at the end. What does not fit that shape is a one-off analysis, not a monthly report.
Five numbers, and their definitions
The five numbers vary by activity, but their nature does not: they follow the journey described in the conversion-rate article, from first contact to money received.
The first is the volume of enquiries: messages, calls, forms, added together. The second is the first-response delay. The third is the number of quotations or orders. The fourth is the number of orders accepted and paid for. The fifth is the advertising spend for the period.
What matters more than the list is the definition written beside each number. What exactly does "enquiry" mean: a message even without a question? A missed call? Two messages from the same person the same day?
That definition has to be decided once and then stay put. A report whose definition changes mid-year produces false comparisons nobody detects, and it is the commonest way a table lies without any figure being inaccurate.
Add a sixth line that is not a number: what changed this month. A campaign launched, a page modified, an employee away, a stock-out. Without that line the other four cannot be interpreted — it is the log described in the testing article, applied to the report.
Why this page carries no chart
Nearly every article in this series carries a dated, placed figure. This one carries none, for a reason belonging to its subject rather than to a general rule.
An article about measurement that displayed figures from elsewhere would do exactly what it holds against reports: substitute an impressive statistic for the one that decides. The numbers that count here are yours, and they are in your inbox.
There is in any case no published series to lean on for this particular subject. Rate comparisons, average bounce figures, session durations by sector are produced by tool vendors, on their own client base, in other markets.
The useful consequence is an exercise rather than a reading: open the last report you received and count how many of its figures come from your business rather than from a platform. The proportion is often one in ten.
What we produce is the reverse: four numbers out of five come from your own records, and the fifth — the spend — comes from the invoice. None of them depends on a tool that could stop working.
What tools count wrongly, and in which direction
Measurement tools are not wrong at random: they err in known directions, and knowing the direction is enough to avoid a bad decision.
They undercount visitors. Blockers, refused consent, strict browsers and messaging apps’ embedded browsers remove part of the traffic from the count. The size varies and it is never nil.
They overcount people. The same person who looks at you on their phone and then on the office computer is counted twice, and a conversation resumed a week later creates a new session with no new visitor.
They overcount conversions when those are clicks. A click on a call button is not a call, and the conversion-rate article refuses that equivalence for precisely this reason.
And they lose the end of the journey. The conversation, the quotation, the delivery and the payment happen off the site, therefore outside the tool. That is the part that decides, and no setting brings it back into the dashboard.
The conclusion to draw is not to give up measuring, but to know which way each error pushes before comparing two periods. A figure wrong in a known direction is still usable for watching a trend; one wrong in an unknown direction is usable for nothing at all.
Attribution does not resolve here
The question "where do our customers come from" looks technical and is not. In this market it has no reliable answer inside a tool, and saying so beats selling a model.
The typical journey runs like this: the person sees an ad, does not click, searches your name two days later, opens a conversation, asks three questions, disappears, and comes back a week afterwards through a link sent by a friend. No tool reconstructs that.
Last click — the default model — attributes everything to the final step, which is the search on your name. So it systematically credits what cost nothing and discredits what made the name known.
The honest substitute is the one the public relations article recommends: the question asked of every new enquiry, written down by hand. "How did you hear about us?" costs five seconds, and over three months it produces a picture no model manufactures.
That method has its faults: people remember badly, they name the last thing they saw, some do not answer. It is imperfect and it is the least wrong available, which is the criterion that counts.
Consent, and what the law asks
Before measuring, you have to have the right to measure. Processing personal data is regulated in Algeria, and a browser identifier attached to a behaviour is personal data.
In practice that means three things. The person has to be told what is collected and why, in readable language. They have to be able to refuse without the site ceasing to work. And refusing has to be as easy as accepting.
A banner offering only an "accept" button is not compliant, and it is very widespread. A bar covering the screen on a phone until it is accepted is the same problem plus a conversion problem.
The consequence for the figures is direct and has to be accepted: a compliant site measures less than a non-compliant one. That is not a loss of measurement quality, it is the real measurement of what was permitted.
We do not give legal advice and this article is not any. For large-scale processing or anything touching sensitive data, the question belongs with the Autorité nationale de protection des données à caractère personnel rather than with an agency.
The properties belong to you
The rule is the same as for advertising accounts, the business listing and hosting, and it is repeated here because measurement is where it is most often forgotten.
The measurement property has to be created with an account belonging to your company, and the agency invited into it. The reverse — the agency creates, you are invited — means that on the day of separation you lose the entire history.
The loss cannot be recovered. Measurement data does not transfer from one property to another: if you start again from zero, you start with no point of comparison, and the first two years of your site disappear.
Check as well who can delete the property, and export once a year. An annual file of your five numbers, kept at your premises, is worth more than a database you do not control.
Finally, note the configured retention period. Many accounts are set to a short retention by default, which silently erases the older history — and nobody notices before they need it.
The figures that serve nothing
Some measures appear in every report and carry no decision. Naming them once saves receiving them for two years.
Impressions and reach: they count opportunities to be seen, not interested people, and they simply go up when you pay more. The article on advertising budgets explains why they reassure and do not inform.
Bounce rate and session duration: they depend so heavily on how the tool is configured and on the kind of page that they compare neither over time nor between sites. A page that answers the question in ten seconds produces a bad figure on both.
Follower count, mention count, number of ranked keywords: three counters that rise with activity and have no demonstrated relationship with sales. The article on online reputation says the same of the average rating.
And comparisons with "the sector average", which come from another market when they come from anywhere. A business compares itself to itself, at constant definitions, and to nothing else.
When the table contradicts the workshop
It happens regularly that the report announces a good month and the shop saw very few people, or the reverse. The tie-break rule is simple and unpopular: believe the workshop.
The reason is that the tool’s numbers describe technical events while the till describes money. When the two diverge, it is nearly always because the first is measuring something other than what its label announces.
The way forward is to trace the definition rather than argue about the gap. What exactly does that line count, since what date, with which filter? In most of the cases we have taken over, the divergence came from a forgotten change of definition.
The second reflex is to check what changed: a duplicated tag counts everything twice, a tag deleted during a rebuild makes a whole channel disappear, a forgotten internal-traffic filter inflates the figures with your own team.
And if everything is correct, the divergence is real information: you have more enquiries and fewer sales, which moves the subject from measurement to response delay or delivery, both handled by this group’s hub.
The monthly conversation, rather than the dashboard
The permanent dashboard, available online at any hour, is a much-requested and rarely-consulted deliverable. We have installed several and we know the internal statistic: two visits in the first month, none afterwards.
What works instead is a short conversation, on a fixed date, with the page in front of everybody. Thirty minutes, once a month, with the person who decides the spending — not with an intermediary.
The structure that holds: the five numbers and their movement, the line of what changed, one proposed decision, and one open question to the client. The open question is what surfaces the field fact the tool will never see.
The monthly rhythm is a deliberate compromise. More frequent and you interpret noise — the error described in the testing article. Less frequent and you discover a badly spending campaign too late.
One exception: the first week of a campaign gets looked at daily, not to judge the result but to check nothing is broken — an unreachable page, a wrong link, an absurd targeting. That is not analysis, it is watching.
Write the question before installing the tool
The usual order is to install measurement and then look for what it can say. It produces encyclopaedic, useless reports, because a tool answers everything and asks nothing.
The useful order starts with a written sentence: what decision do we have to take, and what figure would trigger it? "If cost per accepted order exceeds this amount two months running, we stop this channel."
That sentence determines what has to be measured and, above all, what does not. It shortens the installation, shortens the report, and makes the decision automatic on the day rather than opening a debate.
Write the threshold down before seeing the results too. A threshold set afterwards always adjusts to what one wants to conclude, and it is the decision-making version of the stopping error described in the testing article.
Three or four sentences of that kind are enough for a business. They fit on the same sheet as the definitions of the five numbers, and that sheet is this service’s real deliverable.
What we do, and what we will refuse to do
What we will refuse: putting a figure in a report whose definition we cannot write in one sentence. That removes a great many customary lines, and it is the only way to get a document anybody can argue with.
We will refuse to present impressions, reach or follower counts as a result. Those are activity measures, they rise when you spend, and putting them at the head of a report amounts to marking our own work.
We will refuse to conclude on an attribution model. In this market the journey passes through a conversation the tool cannot see, and a model claiming otherwise is a presentation rather than a measurement.
What we do: the sheet of definitions written before installation; five numbers of which four come from your own records; the line of what changed this month; the question asked of every incoming enquiry; the measurement property created in your name and exported once a year; and a thirty-minute conversation rather than a permanent dashboard.
And what you can do this week without us: take the last report you received and circle the figures on which you would have taken a decision. If there are fewer than three, you already know what to ask of the next one — and what a shorter report will look like.
Frequently asked questions
What should a monthly report contain?
One page: five numbers with their written definitions, a line saying what changed this month, a proposed decision and a question to the client. What does not fit is a one-off analysis, not a recurring report.
Are our statistics reliable?
They err in known directions: they undercount visitors because of blockers and refused consent, overcount people seen on two devices, and lose the end of the journey, which happens off the site.
How do we know where our customers come from?
Not through an attribution model: the journey passes through a conversation the tool cannot see. Ask the question of every new enquiry and write it down. It is imperfect and it is the least wrong method available.
Do we need a consent banner?
Yes, and it has to allow refusal as easily as acceptance, without the site ceasing to work. A compliant site measures less than a non-compliant one: that is not a loss, it is the measurement of what was permitted.
Who owns the measurement account?
You do, if it was created with an account belonging to your company and the agency was invited in. The reverse costs the whole history on the day of separation, and measurement data does not transfer between properties.
The table says a good month, the till says otherwise — which is right?
The till. Then trace the definition of the line concerned: in most cases the gap comes from a changed definition, a duplicated tag, or a forgotten internal-traffic filter.
Where we come in
How many figures you circled says what your report is worth. It does not say whether the ones left standing are even correct.
- We write each number’s definition in one sentence, before measuring anything.
- We check where each comes from and which cross-checks against your accounts.
- We hand back the single page, and what moved since the month before.
If you circled four figures out of five, your report is fine: keep it and buy nothing, because measurement is not your subject.
Read next
The day the measurement broke without saying so
Broken measurement does not look like an outage: it looks like a bad month. How to tell the difference, in ten minutes.A dashboard that says no
A dashboard earns its place through what it refuses to assert. Three refusals to insist on, and the hole never to fill in.A dashboard is read or it dies: the ten-minute review
Most dashboards are not bad, they are abandoned. What keeps one alive is a habit rather than a feature.
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.