You receive an invoice related to an artificial intelligence tool, and half of the lines look like technical jargon. Tokens, API calls, processing volume, platform fees: each item has its logic, but nothing is obvious at first glance. Understanding AI pricing on invoices starts with knowing what each line measures and why it varies from month to month.
Structured and Unstructured Data: What Affects Processing Costs
Before reading an invoice, you need to understand what the tool processes. An invoice sent in Factur-X format (PDF with embedded XML data) requires less work from the AI than a scanned photograph taken on the fly. The format of the input document directly influences the price.
A structured file (CII, UBL, or Factur-X format) already contains identified fields: net amount, VAT, due date, IBAN. The tool only needs to verify and route. The cost per document is then low.
A standard PDF or paper scan forces the AI to engage optical character recognition (OCR), followed by a comprehension model to identify each field. An unstructured scan costs significantly more than a native XML file. If your invoice mentions an item “processing unstructured documents,” this difference is reflected there.
When trying to decipher AI pricing on invoices from Markeonbiz.fr, this distinction between formats is the first reflex to adopt to understand why two companies with the same volume do not pay the same amount.
AI Billing Lines: What Each Item Actually Covers
You may have noticed that an AI tool invoice rarely has a single price? Here are the most common items and what they mean in practice.
- Tokens consumed: the token is the basic unit. It is neither a word nor a character, but an intermediate fragment. In French, a common word can represent one to three tokens. The invoice often indicates a volume of tokens for input (what you send) and output (what the AI generates). Both are billed, sometimes at different rates.
- API calls: each request sent to the model counts as a call. Even if the response is short, the call is counted. A high volume of calls with few tokens per call can cost more than a small number of long calls.
- OCR and extraction: this item appears when the tool reads unstructured documents. It reflects the number of pages or documents analyzed through optical recognition.
- Platform or hosting fees: a fixed subscription, often monthly, that covers access to the infrastructure. This amount does not vary with your usage.
- Compliance surcharge: some platforms charge an additional fee for compliance with specific standards (legal archiving, transmission to approved platforms for electronic invoicing).
The key: distinguish variable costs (tokens, calls, OCR) from fixed costs (subscription, compliance). The former fluctuate each month, while the latter remain stable.
Electronic Invoicing in France: New Lines to Expect on Your Invoices
The French regulatory framework is evolving. Decree No. 2026-677 and the order of July 27, 2026, finalized the rules for the electronic invoicing reform. Starting September 1, 2026, a first level of structural data must be transmitted to approved platforms.
A second, more detailed level comes into effect on September 1, 2027. It will require the line-by-line transmission of descriptions, quantities, unit prices, discounts, and surcharges. For AI tools processing invoices, this means more fields to extract, validate, and transmit.
The direct consequence on your invoice: items related to data extraction and compliance will mechanically increase. Publishers who charge based on the volume of fields processed will pass on this increased granularity.
Another point deserves attention. Approved platforms will need to ensure service continuity and mobility between providers. The cost of migrating from one platform to another is now regulated, which may appear as a distinct line during a provider change.
Reading Your AI Invoice: Three Concrete Checks Before Paying
Why do some invoices seem inconsistent from month to month? Often because a parameter has changed without the user realizing it. Here are three points to check systematically.
Check the input/output tokens ratio
Output tokens often cost more than input tokens. If your invoice has jumped, first check if the volume of generated tokens has increased. A change in model (switching to a more powerful version) can also modify the unit price without changing the volume.
Compare the volume of unstructured documents
A month with more scans or non-standard PDFs increases the OCR item. If you receive supplier invoices in heterogeneous formats, asking your suppliers to send in Factur-X reduces your processing invoice.
Isolate compliance-related fees
With the rise of mandatory electronic invoicing, some publishers integrate transmission fees to approved platforms into a global item. Others detail them. Demand the breakdown to avoid paying twice for a service already covered by your subscription.
An AI tool invoice is no more mysterious than a phone bill: it mixes a fixed fee and variable consumption. The difference is that the units of measurement (tokens, calls, extracted fields) remain less familiar. By identifying the format of your input documents, monitoring the input/output ratio of tokens, and isolating compliance fees, you regain control over each line.