Paying an AI supplier from Algeria

A developer in Oran finishes a prototype on Thursday evening. They open the supplier’s subscription page, take out the company card, and the payment is refused. They try another card, then a director’s, then a cousin’s in France. By Monday the project is three weeks behind and nobody has written another line of code.

This is the first real obstacle to an artificial intelligence project in Algeria, and it is not technical. Paying for a service billed abroad is a service import, regulated, with an order of steps that cannot be improvised and a tax that appears in no quote.

This article describes the regular route, what it asks, what it costs in delay, and what the arrangements in circulation are really worth. It does not cover data law — what you are allowed to send to a foreign model is a separate decision, taken before this one.

This is the characterisation that governs everything else, and it surprises because nothing is delivered. There is no goods, no container, no customs: there is an invoice issued by a foreign company for a service consumed in Algeria, and that is exactly what the service-import regime covers.

The fact that the service is available immediately, by simply creating an account, changes nothing about its characterisation. Ease of access is a property of the product; the applicable regime is a property of the financial operation that pays for it.

This holds for a whole AI project’s chain, not just the model: hosting abroad, the monitoring tool, the vector database, the transcription platform. A business that prepares its file for one supplier and discovers four more foreign-currency invoices three months later starts over four times.

The first job is therefore an inventory, and it fits on a page: every foreign service the project consumes, what it bills, how often, and in which currency. The same inventory that precedes any integration, with one column more.

The bank examines a file, not an intention. It expects an invoice or a contract describing the service, the supplier’s full identity, the amount, the currency, the frequency, and the authorisation obtained beforehand. A subscription form displayed on screen is not an invoice.

What it refuses is predictable and nearly always the same thing: a file where the payee is not the party on the contract, a variable amount with no ceiling, or a service described in three words. None of those refusals is arbitrary; each corresponds to a missing document.

The most underestimated point is variability. A usage-based subscription — you pay what you consume — is hard to domicile cleanly because the amount is not known in advance. A fixed-amount commitment, even slightly more expensive in total, is handled far more easily, and that is a trade to settle with the supplier before signing.

Talking to your bank early beats preparing a file in a vacuum. Practices vary between institutions on what is asked and in what form, and a twenty-minute conversation at a branch often replaces two rounds of documents. And on the day the first invoice arrives, that tax is one of the six lines the reconciliation brings out.

Since the administrative cost attaches to the operation rather than to the amount, the variable that matters is not the monthly price but the number of currency outflows per year. Twelve payments of a hundred units cost twelve procedures; one payment of twelve hundred costs one.

This inverts the usual instinct. A prepaid annual commitment is generally more expensive on cash flow and cheaper in work, and in this market the second half weighs more than usual because it mobilises people rather than money.

The same reasoning recommends consolidating suppliers. Four services of a hundred units at four different companies make four files; the same total at one supplier who resells them makes one. A regular intermediary, invoicing from abroad under its own name, is not an arrangement — it is one more supplier, with a contract.

What must not be done is splitting to stay under a threshold. Cutting an annual expense into twelve to avoid one procedure produces twelve procedures, and it is the one case where the instinct to economise costs precisely what it believes it is saving.

We build the inventory above with you and write the technical description of the services bought — what each does, what it is used for in your business, at what volume — in a form a bank can examine. It is the document technical teams know how to produce and that files rarely contain.

We also design the project around this constraint, which changes real decisions: preferring a fixed-amount commitment, grouping several needs at one supplier, or running locally the part that does not need to leave. Those trades are cheaper taken at the start than corrected afterwards.

We put no irregular arrangement in place, and we do not pay a subscription on your behalf with our own means. The second offer is frequently requested and looks obliging; it creates a dependency you only leave by redoing the project, and we prefer to say so at the moment it is asked for.

We give no tax opinion. The characterisation of a foreign-currency subscription, the fate of the tax and the accounting treatment of the expense are your accountant’s work, and we stop at describing what is bought. We will tell you when the question has changed trade.