AI at work

AI that learns.From your business.

From the first scan to the accounting proposal, FINIDIA turns your confirmed work into better preparation. It learns supplier layouts, refines its instructions and checks improvements before using them again.

Your data is confidential. Protecting it is our priority.
From a document to a lasting understanding
ILLUSTRATIVE DOCUMENT

Example supplier

Supplier invoice

Invoice no.F-2026-018419.10.2026CHF
Professional services1’000.00
Net amount1’000.00
VAT · 8.1%81.00
TotalCHF 1’081.00
FINIDIA · ILLUSTRATIVE DOCUMENT01 / 01

SUPPLIER LEARNING

The next invoice has a head start.

  • Confirmed examples retained
  • Instructions refined automatically
  • Verified before reuse
Your company’s knowledge, under your control

Illustrative example · fictional data

ONE DOCUMENT. A CONNECTED JOURNEY.

From a stack of pages
to a proposal you can trust.

The original stays close at every stage. AI prepares the information; you review it before it becomes part of your accounts.

  1. 01

    Bring in the document

    Scan or import a PDF or image. FINIDIA reads the text and available QR information while preserving the original.

    Your source remains the reference.
  2. 02

    Put the pages in order

    Review page order and suggested splits. Turn a mixed scan into separate invoices, receipts or other documents.

    You control how documents are assembled.
  3. 03

    Let AI prepare

    Identify the supplier, dates, amounts and payment details. A matching verified supplier model guides extraction; otherwise, general extraction takes over.

    Every amount comes from the new document.
  4. 04

    Review and clarify

    Check the proposal beside its PDF. Correct a detail or answer a targeted question. Your confirmed corrections stay with the current proposal.

    Recurring instructions can be saved as Your rules.
  5. 05

    Save into accounting

    Approve the proposal, review the editable invoice and save it with its accounting entry. The source document stays linked.

    Posting and payment remain separate decisions.
  6. 06

    Make future work easier

    A saved invoice with valid, balanced accounting can become a learning example. FINIDIA uses confirmed evidence to build and check supplier instructions.

    Learning begins with confirmed accounting.

Original document → extracted facts → your rules → reviewed accounting. The evidence remains connected.

SELF-LEARNING, WITH VERIFICATION

Experience becomes
better preparation.

FINIDIA improves the instructions it uses for your suppliers over time. It checks a proposed improvement against confirmed examples before making it available for future invoices.

  1. 01

    Learn from confirmed work

    Use a valid reference document already recorded in accounting, together with up to two other confirmed documents from the same supplier model.

  2. 02

    Generate or refine

    Gemma uses the original evidence, your saved rules and any verification errors to build more precise instructions.

  3. 03

    Replay and compare

    Test the instructions on every selected example. Compare extracted facts and the accounting treatment with the confirmed results.

  4. 04

    Reuse what passes

    Only verified instructions become available for matching future documents. Each new proposal still goes through review.

If a check fails, refine and test again.

Gemma receives the exact differences and revises the instructions. All selected examples are checked again, within a limit of three attempts per run.

Once all examples pass, the instructions are ready to reuse.

The supplier model becomes verified. If checks still fail, general extraction remains available.

What triggers learning, and what actually changes?

Learning starts with valid examples already recorded in accounting. A new example triggers another check of an unverified model; changes to reference values or manual verification can also start a check. Verified models are not retested for every new invoice. The system improves saved supplier instructions, not the underlying AI model’s weights. Verification never rewrites your saved accounting entries or your rules.

THE DIFFERENCE IN EVERYDAY WORK

Your context.
A more useful proposal.

Three fictional examples show how source evidence, saved instructions and review work together. Select a situation to explore it.

01 / 03

The invoice date is not the service period.

A February telephone invoice covers January. Save an instruction to look for the billed service period on this supplier’s invoices.

“For this telephone supplier, fill in the service period when it is printed on the invoice.”

The service period is included when found and remains optional when it is absent.

What the document says
Invoice date
01.02.2026
Billed service
01–31.01.2026
Total
CHF 126.35
Prepared with your rule
Service period
01–31.01.2026
Expense account
Telecommunications
Payable total
CHF 126.35

The period is read again on the next invoice. January is never copied as a fixed value.

02 / 03

Read the printed VAT. Apply your accounting rule.

Your company has decided that VAT on this purchase is not recoverable. The printed figures remain evidence, while the full total becomes an expense.

“For this supplier, expense the full amount including VAT. Do not recover VAT.”

The payable total stays unchanged. Source facts and accounting treatment are checked separately.

What the document says
Net amount
EUR 47.54
Printed VAT · 22%
EUR 10.46
Total
EUR 58.00
Prepared with your rule
Expense amount
EUR 58.00
Recoverable VAT
EUR 0.00
Payable total
EUR 58.00

Illustration of a company-approved treatment, not a determination of VAT eligibility.

03 / 03

Account for what was invoiced.

The PDF includes an order confirmation and an invoice for a partial delivery. The larger order total is supporting context, not the amount currently payable.

“Use the partial-delivery invoice total. Do not use the attached order confirmation as the invoice amount.”

You review the proposal against the relevant invoice pages.

What the document says
Order confirmation
CHF 3’547.85
Partial-delivery invoice
CHF 1’081.00
Prepared with your rule
Invoice net
CHF 1’000.00
VAT · 8.1%
CHF 81.00
Payable total
CHF 1’081.00

A correction applies to this document. Saving it as a recurring supplier rule is a separate choice.

Illustrative scenarios with fictional data. Actual proposals depend on source quality, saved rules and your review.

CONFIDENTIALITY IS A PRIORITY

Your data is confidential.
Protecting it comes first.

Your documents contain sensitive business information. FINIDIA keeps access controls and company boundaries central to document preparation, learning and accounting.

Trust & control

Access for the right people

Company memberships and roles govern who can see and change information. AI-assisted actions follow the same permission checks.

Learning within your company

Your saved rules, supplier models and confirmed examples belong to your company workspace. They are not automatically shared with other customer companies.

Private records, reviewable decisions

Your documents and accounting records are not published by the learning process. Proposals remain reviewable, with source evidence and a history of changes.

AI processing is part of your deployment choices. Document content may be sent to your configured AI provider for analysis; its processing and retention terms must be agreed for your deployment.

Questions, answered

Does the system learn from everything I type?

A correction updates the current document. To keep an instruction for future use, choose Save rules explicitly. Supplier learning uses eligible examples already recorded in accounting; it does not treat every message as a permanent rule.

What happens when verification still fails?

After up to three attempts, the model remains unverified and general extraction stays available. You can inspect the differences and retry later, manually or when a new eligible invoice triggers another check.

Can AI change my accounts by itself?

Learning and verification do not change saved invoices or journal entries. AI prepares a proposal for review; saving, posting and payment keep their own authorized steps.

Does learning make my data public?

No. Supplier learning does not publish your documents or automatically share your examples with other customer companies. Configured AI providers may process document content, subject to the arrangements for your deployment.

LESS REPETITION. MORE UNDERSTANDING.

Every confirmed example can make the next step easier.

  1. 01

    Follow the document journey

  2. 02

    Explore the practical examples

  3. 03

    See the connected accounts

Contact

Let’s talk about your business.

Tell us what your team needs. We’ll help you explore how FINIDIA fits your work.

sales@finidia.com