The verdict in three sentences
For an SME receiving 1,000 to 3,000 supplier invoices a month, AI extraction cuts processing time from 5 minutes to about 1 minute per invoice and frees the equivalent of one full-time position. A market solution at 0.10 to 0.50 EUR per invoice works as long as invoices stay standard and the accounting software is natively connected. A custom pipeline at 10,000 to 25,000 EUR excl. VAT makes sense when you need purchase order matching, your own cost allocation rules and reliable ERP integration, with 95 to 98% accuracy on key fields.
E-invoicing does not remove the need for extraction
Since 1 September 2026, every French company must be able to receive e-invoices through an approved platform. Issuing only becomes mandatory for SMEs and micro-businesses on 1 September 2027. For at least twelve months, an SME in Lyon will therefore receive a mixed flow: structured invoices (Factur-X, UBL), plain PDFs, foreign invoices outside the reform, scanned expense notes and credit notes.
| Type of invoice received (Lyon SME, 2026-2027 estimate) | Share of flow | Suitable processing |
|---|---|---|
| Factur-X or UBL via approved platform | 35 to 55% | direct read of structured data |
| Native PDF without structured data | 25 to 40% | AI extraction on text |
| Scanned or photographed invoices | 5 to 15% | OCR + AI, stronger checks |
| Foreign supplier invoices | 5 to 10% | multilingual AI, intra-EU VAT |
| Credit notes and expense notes | 3 to 8% | specific sign and allocation rules |
| Multi-page invoices with detailed lines | 5 to 10% | line extraction, PO matching |
The 2026 challenge is no longer just reading a PDF but merging two flows into a single validation and allocation circuit.
Template OCR, generative AI or market solution: the comparison
Template OCR recognises recurring suppliers well but breaks as soon as a layout changes. Generative AI models and specialised document models understand an invoice's structure without templates, at the cost of mandatory consistency checks.
| Criterion | Template OCR | Market solution (Yooz, Libeo, Pennylane, Mindee) | Custom AI pipeline |
|---|---|---|---|
| Setup cost | 2,000 to 6,000 EUR excl. VAT | 0 to 3,000 EUR excl. VAT | 10,000 to 25,000 EUR excl. VAT |
| Unit cost | 0.05 to 0.15 EUR per invoice | 0.10 to 0.50 EUR per invoice | 0.02 to 0.08 EUR per invoice (API + hosting) |
| Accuracy on key fields | 80 to 90% | 92 to 97% | 95 to 98% |
| Line items and PO matching | rarely | depends on plan | yes, custom |
| Automatic cost allocation | no | simple rules | learned from history |
| ERP integration (Sage, Cegid, SAP B1) | CSV export | standard connectors | native API |
| Data hosting | varies | vendor cloud | your choice, EU |
At 1,800 invoices a month, a market solution at 0.30 EUR costs about 540 EUR a month, or 6,500 EUR a year. A custom pipeline at 18,000 EUR excl. VAT with 0.05 EUR per invoice and 150 EUR hosting costs 240 EUR a month to run: the 300 EUR monthly difference only pays back the build in 5 years. Custom is therefore justified by features, not by unit price.
The target circuit and its checks
An AI that extracts 97% of fields correctly lets 3% of errors through: on 1,800 invoices, that is 54 wrong invoices a month if no check is planned. The pipeline must cross-check extracted data with what the company already knows.
| Circuit step | Automatic check | Rate sent to a human (order of magnitude) |
|---|---|---|
| Receipt and filing | duplicate (invoice no. + company ID + amount) | 1 to 2% |
| Field extraction | confidence score per field | 5 to 10% |
| Consistency check | net + VAT = gross, valid VAT rate | 1 to 3% |
| Supplier identification | known company ID and IBAN | 2 to 4% |
| PO and goods receipt matching | price or quantity gap < 2% | 5 to 8% |
| Accounting allocation | account and cost centre suggested | 3 to 6% |
| Approval and payment release | approval chain by amount threshold | per delegation |
Overall, 75 to 85% of invoices go through untouched, and the rest land in a review queue with the disputed field highlighted.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Mini case study
Julien, CFO of a 120-employee electrical equipment distributor in Lyon, processes 1,800 invoices a month with two assistants who spend about 150 hours on them. After deploying a custom pipeline at 18,000 EUR excl. VAT, 80% of invoices pass without intervention and checking the remaining 20% takes 45 hours a month: 105 hours saved, or about 3,700 EUR a month at a loaded 35 EUR an hour. Running cost: 240 EUR a month. The build pays back in 5 to 6 months, and one of the two assistants moves to customer collections.
FAQ
What accuracy should you expect from AI invoice extraction in 2026?
Between 92 and 98% on key fields (supplier, date, number, amounts, VAT) for native PDFs. On poor-quality scans, count 85 to 92% and systematic human review.
Does the e-invoicing reform make AI useless?
No, issuing only becomes mandatory for French SMEs in September 2027 and foreign suppliers stay outside the reform. AI remains useful for 45 to 65% of the flow for at least a year, then for allocation and matching.
Can invoice data be sent to an AI hosted in the United States?
It is possible with suitable contractual clauses, but most CFOs prefer EU hosting. Major API providers offer EU regions at a 0 to 10% premium.
At what volume does a custom pipeline make sense?
Above 1,000 invoices a month with PO matching or complex cost allocation. Below that, a market solution at 0.10 to 0.50 EUR per invoice remains more cost-effective.
How long does the project take?
Count 6 to 10 weeks for a custom pipeline, including 3 weeks of calibration on 500 to 1,000 real historical invoices.
Let's scope your project. Tell us your monthly invoice volume, ERP and approval rules: we will price a solution between 10,000 and 25,000 EUR excl. VAT or point you to a market tool if that is more cost-effective. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
