E-invoicing in France and AI: automate your invoices

Since 1 September 2026, every French company must be able to receive invoices electronically, and large companies and mid-sized companies (ETIs) must also issue them in that format. Many finance departments have therefore picked an approved platform and consider the matter closed. That is a misreading: the reform standardises how invoices are transmitted, not how they are processed. Matching an invoice to the purchase order and the goods receipt, coding it, handling a rejection, applying a payment: these tasks remain largely manual. The good news is that e-invoicing finally provides structured, reliable data, exactly what an AI agent needs to automate these steps. This guide explains where the reform stands, what the approved platform does not do, the six most profitable AI agent use cases across the invoice chain, and how to deploy them.
‍

E-invoicing in France: where the reform stands in October 2026

One month after launch, the reform is well established. According to the first progress report published by the French Ministry of the Economy on 12 October 2026, 75% of VAT-registered companies have designated a receiving platform, 20 million invoices have been exchanged and 70 billion euros of invoicing has gone through the new system.

The reform has two deadlines, according to economie.gouv.fr:

• 1 September 2026: every company subject to VAT must receive invoices electronically through an approved platform.
• 1 September 2026: large companies and ETIs must issue all their invoices electronically and transmit their e-reporting data.
• 1 September 2027: SMEs and micro-businesses must in turn issue their invoices electronically and transmit their e-reporting data.

Three points deserve attention from mid-sized companies. First, four new mandatory mentions have appeared on invoices since 1 September: the customer's SIREN number, the category of transaction (sale of goods, services or both), the option to pay VAT on debits where applicable, and the delivery address if it differs from the billing address. Second, the government has confirmed that no penalties will be applied before the end of 2026: the window to make your processes reliable is open, but it closes in two and a half months. Third, at the monitoring committee of 8 October, large companies reported interoperability issues between solutions and problems transmitting invoice statuses, which disrupt the automation of processing chains. In other words, the system works, but what happens downstream is not yet automated.
‍

What the approved platform does, and what it does not

An approved platform (plateforme agréée) is the mandatory intermediary that routes invoices between supplier and customer and sends the required data to the tax authorities. Its role covers transmission, format compliance and status tracking.

What the approved platform handles:

• receiving invoices in a structured format, with notification;
• checking the format and mandatory mentions;
• transmitting the lifecycle statuses of each invoice;
• sending the required data to the tax authorities.

What remains for the company to do:

• route each invoice to the right approver;
• check that the content matches what was ordered and delivered;
• match purchase order, goods receipt and invoice, then handle discrepancies;
• choose the accounts, cost centres and projects for coding;
• decide on a rejection or refusal, justify it and deal with the supplier;
• match bank statements with invoices, apply payments and chase late payers.

The second list is where time is spent and errors occur. It is also where an AI agent creates value.
‍

Why e-invoices are the best fuel for an AI agent

For years, invoice automation hit a simple obstacle: the input data was messy. Scanned PDFs, layouts specific to each supplier, approximate OCR, manual re-keying. The reform changes that. An e-invoice holds its data in a structured form, in standard formats (Factur-X, UBL, CII), where the same tag always carries the same piece of information.

For an AI agent, this changes three things. It no longer spends its effort reading the invoice, but reasoning about it: comparing, checking, deciding or alerting. All invoices arrive through a single channel, so the agent can be triggered on every receipt. And the time-stamped lifecycle statuses give it a reliable stream of events to follow each file through to payment. To understand what sets an agent apart from classic rule-based automation, see our guide to agentic automation.
‍

Six AI agent use cases across the invoice chain

1. Three-way matching: purchase order, goods receipt, invoice. This is the most time-consuming check in accounts payable. The agent retrieves the purchase order from the ERP and the goods receipt, then compares lines, quantities, unit prices and terms. When everything matches within the defined tolerance, the invoice is approved for payment. When a discrepancy appears, the agent classifies it (partial delivery, price increase, quantity error), prepares a proposal and submits it to the right person.

2. General ledger and cost accounting coding. Based on posting history, the supplier, the nature of the lines and the requester's cost centre, the agent suggests the account, the analytical axis and the project. The accountant validates in one click instead of searching. Uncertain suggestions are flagged as such, which focuses human attention where it matters.

3. Rejections, refusals and supplier disputes. Under the reform, refusing or rejecting an invoice goes through a status sent via the platform. The agent prepares the reason, drafts the exchange with the supplier, tracks the reply and follows up if needed. This is precisely where large companies currently report friction.

4. Cash application and bank reconciliation. On the receivables side, the agent matches each bank statement line with the corresponding invoice, based on amount, reference and date, then applies the payment. Only unmatched lines go back to the team. This is the use case Qolaig deployed at Pan Distribution (detailed below).

5. Payment reminders and payment terms tracking. Time-stamped statuses show where each issued invoice stands. The agent detects delays, adapts the reminder to the customer's history and the amount, and alerts the account owner when a file calls for a phone call rather than an email. The French government lists better compliance with payment terms among the expected benefits of the reform.

6. Checking issued invoices and out-of-scope flows. Before submission to the platform, the agent checks that the new mandatory mentions are present (customer SIREN, transaction category, delivery address) and consistent with the order. It also handles what the reform does not cover: invoices from foreign suppliers, which still arrive as PDFs, and transactions subject to e-reporting.

For a broader view of which agents to deploy first, across all functions, see the 15 AI agents to deploy first in your company.
‍

Case study: how Pan Distribution applies 300+ payments a week with no manual entry

Pan Distribution is an audiovisual production and distribution company. Its accounting and distribution teams spent several hours a day applying payments by hand: one by one, comparing bank statements with the invoices available on Cinego, the industry's business platform. More than 300 invoices a week went through this process, with a constant risk of keying errors and duplicates.

Qolaig built an automatic cash application tool connected to the existing environment, without changing the team's tools. The user imports the bank statement export, the only action required. The tool queries Cinego, identifies the invoices matching each line and applies the payments. Lines that cannot be matched automatically are sent back to the team for review.

Results: more than 300 invoices processed automatically every week, several hours saved every day, zero duplicates or keying errors on automatically processed payments, and a team refocused on complex cases. Read the full case study.

The lesson applies elsewhere: value does not come from the most sophisticated tool, but from starting with a repetitive, measured task, with clear rules and explicit handling of exceptions.
‍

Built-in platform AI or a custom AI agent: how to choose

Many approved platforms and accounting software now include AI features: invoice reading, coding suggestions, duplicate detection. For a standard scope, that is often enough. A custom AI agent becomes relevant when at least one of these criteria applies:

• the process spans several systems: ERP, approved platform, bank, industry-specific tool, email;
• your rules are specific: tolerances by purchase category, approval workflows, negotiated terms;
• the volume of exceptions is high and handling them, not data entry, is what consumes time;
• the data you need lives in a tool that software vendors do not connect to, like Cinego at Pan Distribution.

Connecting an agent to these systems increasingly relies on open standards such as MCP, which we cover in our guide to the Model Context Protocol.
‍

Deploying an AI agent on your invoices: a 4-step method

1. Map the flows. Monthly volumes of invoices received and issued, channels, systems involved, number of people involved, share of invoices still outside the electronic format.
2. Measure the baseline. Average processing time per invoice, exception rate, average time from receipt to payment approval, collection time. Without this baseline, no gain can be demonstrated.
3. Design the agent with a human in the loop. Define what the agent handles alone (below an amount threshold, within a discrepancy tolerance), what it proposes for validation and what it escalates. Start with a single use case, often matching or cash application.
4. Measure ROI over time. Compare the step 2 indicators after three months in production, then track running costs. Our method for calculating the ROI of an AI agent details the formula and the indicators.
   ‍

Governance: the safeguards to set from day one

An agent that touches financial flows must follow the same internal control principles as an employee. It never triggers a payment on its own: it prepares, a human authorises. Every proposal and every decision is logged, with the data that justifies it, so that it remains auditable. The agent's access rights are limited to what is strictly necessary, and supplier and customer data are hosted in line with your confidentiality requirements. These rules are easier to set during scoping than to add afterwards.
‍

FAQ

Is e-invoicing mandatory for mid-sized companies in France in 2026? Yes. Since 1 September 2026, ETIs must receive and issue their invoices electronically through an approved platform, and transmit their e-reporting data. SMEs and micro-businesses will have to issue e-invoices from 1 September 2027.

Does the approved platform automate invoice processing? It automates transmission, format checks and status tracking. Matching with purchase orders, coding, dispute handling and cash application remain the company's responsibility, and that is where an AI agent comes in.

What is the difference between OCR and an AI agent for invoices? OCR reads a document to extract data. An AI agent reasons about that data: it compares the invoice with the order, suggests coding, classifies a discrepancy and decides whether to escalate it. With structured e-invoices, the reading step becomes secondary.

Can an AI agent pay an invoice on its own? Technically yes, but it is not recommended. Best practice is to let the agent prepare the payment approval and to keep payment authorisation with a human, in line with segregation of duties.

What does a non-compliant company risk? The government has confirmed that no penalties will be applied before the end of 2026, as part of a lenient start-up phase. Companies therefore have a few weeks to make their processes reliable before the fines provided for by law apply.

Which use case should you start with? With the most repetitive and most measurable task in your chain: usually supplier invoice matching or cash application. A first agent in production on a narrow scope is worth more than a company-wide project that never leaves the pilot stage.

At Qolaig, our Workshop → Build → Run methodology frames this type of project before any technical investment: mapping your invoice chain, estimating recoverable time and ROI, then deploying a first agent in production. Let's talk about your project.