Algeria AgencyArtificial intelligence
Artificial intelligence
Assistants, document extraction and automation built on your own data — with a written boundary and a hand-off to a person.
Artificial intelligence is sold as a capability and bought as a decision: is this task, in your business, worth handing to a machine that is right most of the time? For some tasks the answer is obviously yes and for others obviously no, and working out which is which is what we do before writing anything.
The test is the cost of a mistake. Sorting incoming messages, summarising a long document, drafting a reply: an error is visible and corrected in seconds. Quoting a price, approving a file, sending a commitment on the company’s behalf: an error surfaces later and is expensive. So we automate the first group freely and refuse to leave the second without a person in the loop.
Technically we build much the same thing every time: a system that answers from your real material — pages, catalogue, procedures, message history — and says it does not know when it does not. An assistant that invents a plausible answer does more damage than a contact form, because the customer believes it.
Finally, we will often tell you the problem is not an AI problem. A business answering the same question twenty times has a missing page before it has anything else, and fixing that costs a day rather than a project. We say so first, and it is not the most profitable answer for us.
What we usually find
- You answer the same questions about prices, hours or lead times every day.
- Your team retypes data that already exists in a document you received.
- You were shown an impressive demonstration, and nobody knows what it would cost per month.
- You tried a consumer tool and it confidently said something false about your business.
What changes
A written boundary
What the system handles alone, what it passes to a person, and what it may never do. Written before go-live rather than discovered after.
Answers that cite their source
Every answer points at the material it came from, so a mistake is fixed by fixing the source instead of by guesswork.
A predictable monthly cost
Cost per request and volume measured on your real traffic, with a ceiling. A foreign-currency bill tripling without warning is the genuine risk in these projects.
Explore
AI chatbots & virtual assistants
An assistant on WhatsApp, Messenger or your site that answers from your real content, and hands over the moment it leaves its subject.AI voice & call agents
An agent that answers, books an appointment or takes a simple order, and hands over to a person the moment the call leaves the script.AI-powered automation
Enquiry qualification, record enrichment and handoffs between your tools — automated where a mistake can be undone, and only there.Custom AI integrations
Connecting a model to your site, CRM or point of sale — without replacing what already works, and without locking you to one provider.AI analytics & dashboards
Dashboards that reconcile your sources, explain where they disagree, and refuse to conclude when the data will not carry it.AI search & recommendation
Site search that forgives typos and handles your customers’ three languages, plus recommendations drawn from your real sales.AEO (AI engines)
Being cited by ChatGPT, Gemini, Perplexity and Google AI Overviews when they answer instead of a page of results.
What you get
Scoping by cost of error
Your tasks ranked by what a wrong answer costs, which is what decides where automation stops and a person starts.
An assistant grounded in your material
It answers from your pages, procedures and catalogue, in French, Arabic and English, and hands over when it cannot find something.
Document reading
Extraction of the useful fields from incoming orders, invoices and forms, with a confirmation screen before anything is saved.
Triage and drafting
Incoming requests classified by subject and urgency, and draft replies your team edits instead of writing.
Integration with your tools
Wired into what you already use — website, mail, CRM, spreadsheet — rather than delivered as one more interface to open.
Logging and measurement
What was asked, what was handed over, what was got wrong. Without that log nobody can say whether the system works.
How we work
Look at the real traffic
A hundred messages you genuinely received, with the answers that were given. Better material than any specification.
Settle the scope
What gets automated, what stays human, and the hand-off rule. This stage ends in a document, not in a prototype.
Build one slice
A single use case, put in your team’s hands before any customer sees it, with a measurement of how much it handles alone.
Open up and watch
A gradual go-live, the log read weekly at first, and corrections made to your material rather than to the model.
Is this the right fit for you?
This is for you if
- A real repetitive volume: dozens of near-identical messages or documents a week.
- Your information exists in writing somewhere, however badly organised.
- You accept that a useful system is sometimes wrong, and you want to decide where.
This is not for you if
- You want to replace a person in month one. That is not what happens, and promising it would be a lie.
- Your answers exist nowhere in writing. They have to be written first — and that is content work, not AI work.
- The task produces a number: a price, a quote, a regulatory calculation. Ordinary software is more accurate and cheaper.
What we commit to
No invented answers
The system answers from your material or says it does not know. We do not take on projects that require otherwise.
You know where your data goes
Which supplier, which country, what is retained and for how long. Written before signature, including when the answer is unwelcome.
Something you can switch off
Material, logs and code belong to you, and turning the system off breaks nothing else. An assistant never becomes a dependency.
AI in an Algerian setting
Your customers write in French, Arabic and derja, often in one message and in Latin characters. Consumer tools handle that mixture badly, and it is the first thing we test against your real messages rather than against chosen examples.
These services are almost all billed by usage and in foreign currency. A project that never measured its real volume discovers its bill in month three. So we cost the price per request against your observed volumes and set a ceiling from go-live.
Finally, data leaves the country the moment a model hosted abroad is used. For some documents that is immaterial and for others it is not. We set out the local option, its cost and its limits, and the choice is yours — it is not disguised as a technical constraint.
Frequently asked questions
Does this replace staff?
Not in the first months, and rarely afterwards in the way it is advertised. What actually happens is triage: easy questions disappear and your team spends its time on the ones that deserve a person.
What happens when the system is wrong?
It has to be wrong visibly first: citing its source, so the answer can be checked. The correction then goes into your material, which repairs the answer for everybody rather than for one case.
Is our data used to train a model?
It depends on the supplier and the contract, and it is verifiable. We tell you what the one we propose says, and we favour options that rule that reuse out.
Can it be hosted in Algeria?
For some open models, yes, on a server you own. Quality is below the large remote models and the fixed cost is higher; on simple, sensitive tasks the trade-off still often lands on that side.
How long before we see a result?
A first useful slice ships in a few weeks when your material already exists. Where it does not, most of the elapsed time is writing it, and we say so up front rather than halfway through.
Do we need an AI project to be cited by AI search tools?
No, and it is a common confusion. Being cited is a matter of content and its structure — the work described on the AEO page — and has nothing to do with installing an assistant in your business.
How to start
Send us a hundred messages or documents you genuinely received last month, with the answers that were given.
We come back with the share the system would handle alone, the share it would hand over, the estimated monthly cost, and the cases where we would advise against automating at all.
What we have written on this subject
Training a model: the three cases where it is worth it
Training changes the shape of answers, almost never their content. What you must supply, and the three cases where it is the right tool.The impact assessment, before connecting an assistant to your customer data
It is done before go-live, and that alone is what distinguishes it from a report. Afterwards it describes a risk that has already either materialised or not.Artificial intelligence: what it does well, what it invents, and what you cannot hand it
Its failure mode is not breakage, it is confident error. It produces text just good enough that nobody checks it.The AI your staff already use: what leaves the business, and the one page of rules that is enough
You bought nothing and it is already in service, on personal accounts. The problem is not the tool — it is that nobody knows what goes through it.Measuring darija transcription: the protocol, and the figure we do not publish
Nobody publishes an error rate for Algerian darija, us included. Here is the protocol that would produce one, and what to know before reading it.The national AI strategy, translated for a small business
A strategy sets a direction; it imposes nothing on a business. Where it stands on 6 September 2026, what concerns you, and what already obliges you.Arabic, French, darija: which language to answer in, and to whom
The language of a message is not always the language of the expected answer. The rule is decided per channel, written on one page, and held to.Running a model on your own machine: the three numbers that decide
The number in a model’s name does not say whether it will fit. Three others do, and they are worked out before buying anything.RAG without the jargon: what an assistant on your documents still invents
Connecting a model to your documents does not make it honest. It changes what it invents, and moves the problem to what you give it to read.Agent, chatbot, automation: three things, one name
Three different products are sold under the same word. They fail differently, and the question that separates them is: who starts it, and who decides?Your first AI invoice: the lines that were not in the calculation
The estimate said one amount and the invoice says another. The gap is almost never an error: it is six lines nobody counted.The cost of a call, minute by minute
Four meters run during a voice call and none of them counts the same thing. Confusing them gives an estimate wrong by a factor, not by a margin.
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.
- No invented answers
- You know where your data goes
- Something you can switch off