AI + OCR: Viridem's hybrid, gradual approach

At Viridem, artificial intelligence assists people, it never replaces them. OCR captures the data on your invoices, AI makes it intelligible, and someone on your team always confirms. Here is why this transparent, controlled approach is safer.

AI in the service of accounts payable automation

In accounts payable automation, AI is not a gadget: it is what turns a supplier invoice into reliable data. At Viridem, AI reads and makes the invoice intelligible, the preconfigured workflow validates it, matches it to the purchase order and records it in your ERP, under the eye of a person who confirms. The result: less manual entry, fewer errors, and an accounts payable process that is measurable end to end.

OCR + hybrid AI: read, then make an invoice intelligible

OCR (optical character recognition) turns a paper or PDF invoice into usable text. On its own, it stalls as soon as the layout changes. Hybrid AI adds the understanding layer: it locates the vendor, amounts, taxes and line items, even on unfamiliar formats. This is the principle of IDP (Intelligent Document Processing), at the heart of recent document-AI research, from LayoutLM (Microsoft Research, 2020) to LayoutLMv3, DocFormer (Amazon) and Donut (Naver Clova). By combining both, Viridem reaches 99% accuracy, where OCR alone falls short. And in the end, a person always validates.

Viridem's AI features

Concretely, Viridem's AI provides document-recognition assistance: it suggests the document classification and pre-fills the right fields, which your accountant validates at a glance. It also handles automatic data extraction and two- and three-way matching. The AI proposes, the human decides: it is AI in the service of the decision, never the other way around. The technical detail is on our Technology page.

Preconfigured workflows, not machine learning

We made a deliberate choice: not to hand our workflows over to machine learning. Machine learning works well enough, but its learning curve is long and often tedious: you have to train it, correct it, wait for it. We prefer workflows preconfigured by sector and by trade, which we have known and refined since 2013.

The image is simple: our preconfigured workflows are rails already laid down, your invoice runs straight, fast and without surprises. Machine learning, on the other hand, is clearing new ground: you open a path by trial and error, never quite knowing where it leads. For accounts payable, we prefer the rails.

This choice has another merit: it avoids the black box. Every rule is explicit, every step is traceable and verifiable. You always know why an invoice took a given path, something an opaque statistical model cannot guarantee.

Why gradual AI, not "100% AI"

A fully autonomous "100% AI" solution is appealing on paper. But it carries concrete risks: hallucination (a plausible but wrong value), runaway (cascading decisions with no checkpoint) and loss of data control. On invoices, meaning money and compliance, these risks are unacceptable. Hence our constant principle: the human confirms, the AI assists. And we roll out each AI capability in stages, once proven in the field, never all at once.

A safer approach, backed by research

The scientific literature supports this caution. Work on document extraction (DocVQA, Pix2Struct by Google) shows that accuracy depends on context and degrades on rare documents, which justifies human supervision. Landmark research in process mining (Wil van der Aalst) reminds us that a process is only reliable if it is measured and controlled. That is exactly our method. Find all our sources on our Scientific literature page.

Control and confidentiality of your data

As a Canadian publisher, Viridem handles your financial data with the greatest care, in the spirit of Canadian privacy standards: controlled access, an audit trail, and data that is never sold, rented or exploited for commercial purposes. You keep control of your information. For the detail, see our privacy policy.

A Canadian software publisher, with an AI department since 2023

Viridem has been building its platform in Canada since 2013. Since 2023, a dedicated AI department evolves the solution gradually: every advance is tested, validated, then integrated, at the pace of the field. We are listed on the Vitrine IA du Québec.