Platform

DACH Bank Detail Verification

German bank-detail change forms, typed or handwritten, become verified CSV rows. A value is auto-accepted only when two independent readings agree.

Try the live demo Talk to us

Runs in your cloud or on your servers · People stay in control

Bank Detail VerificationLive preview

Platform

Read this handwritten Form A from Cedar GmbH.

✓ Detected the formForm A · 38 fields · 2 sites

✓ Checked the IBANChecksum valid · OCR and the AI model agree

36 fields verified. Please confirm the terminal-site town: the handwriting is unclear.

Why teams bring it in

The work it takes off your plate.

Payment providers in Germany, Austria and Switzerland receive bank-detail changes as signed forms: 'Änderungen der Stammdaten und/oder Bankverbindungen' and 'Antrag auf Änderung der Bankverbindung'. The platform detects the form, reads every field with OCR, has the AI model read the cropped cell again without seeing the OCR text, checks IBAN, BIC, PLZ and dates, and shows a reviewer only the values that need a look. Approved forms export to CSV or Excel in your column order.

What it handles

4 jobs, done properly.

Bank detail changes

IBAN, BIC, account holder and bank read and checked.

No payments to wrong accounts

Master-data updates

Company, contact and site details captured from Form A.

Clean records

Terminal sites

'Identisch' tick boxes fill the terminal side from the main site.

Fewer blank fields

Bulk back-office queues

Up to 30 PDFs per batch, two read at a time, one export.

Hours back each week

How it works

4 steps, start to finish.

Each step is visible to your team, and nothing important happens without a person's say-so.

✉ EmailPDFXLS

Upload

Drop in up to 30 typed, scanned or handwritten PDFs.

Detect and read

Form A or B is detected from its header; every field is read by OCR.

Details✓

Rules✓

Sources✓

Read again and verify

The AI model reads each cell image independently; IBAN, BIC, PLZ and date rules check the result.

Approved ✓

Review and export

Press J for the next issue, confirm or correct, then approve to CSV or Excel.

independent readings per value2

fields on Form A, detected automatically38

auto-accepted wrong values in the sample eval0

Try it yourself

The business workflow

What happens, and where people decide.

◎What starts it

A batch of signed forms or contracts is uploaded: typed, scanned or photographed, up to 30 PDFs at a time.

⇣What goes in

Bank detail change forms (two German layouts, 38 and 9 fields), with typed, scanned and handwritten entries and tick boxes.

✕Manual work it replaces

Reading each form, retyping every field and checking bank details digit by digit.

✦What it produces

A fixed set of columns per form, every value checked, exported to CSV or Excel for import.

☺Where a person decides

A reviewer sees the page with a box on every field. Only flagged values need attention. Press J for the next issue, then confirm or correct, and approve the export.

Before

By handevery field retyped

Riskyone wrong digit misdirects a payment

Every fieldneeds a person's eyes

→

After

Secondsto read a form

Checkedevery IBAN and BIC verified

14%of fields sent to a person in our test

High-level architecture

From input to outcome.

Full architecture pack

Inputs

Scanned, typed and handwritten PDFs
Bulk upload, up to 30 at once

Ingestion

Pages rendered to images
Cloud OCR and text extraction OCR: forms, tables, tick boxes
Cache by file fingerprint, each PDF read once

AI layer

A large language model on a managed cloud AI service reads cropped field images independently

Rules

IBAN checksum and OCR-confusion repair
BIC, postcode, date and email checks
Cross-field consensus rules

Human review

Page view with a box per field
Only flagged values need a decision

Outcome

Approved CSV or Excel export
Per-document run record

IntegrationsExport to CSV or Excel for the customer's core system importProduction integration to the target system to be scoped

Design principle

Two independent readings before trust. A value is auto-accepted only when two independent readings agree and the rules pass. Everything else goes to a person.

Technology stack

What it runs on, and why.

OCRCloud OCR and text extraction (FORMS and TABLES)

Reads every field and table on each rendered page, plus a tick-box ink check.

AI modelA large language model on a managed cloud AI service

Second, independent reading of uncertain cells from the cropped image only. Can be switched off, leaving OCR plus rules.

ValidationDeterministic validators in

IBAN checksum, BIC, German postcodes and dates, email. IBAN repair for common handwriting confusions (7 read as Z, 8 as S).

ConsensusCross-field rules

'Same as above' boxes, BIC agreement per bank, repeated names and towns must agree.

Backendweb server and pipeline

Upload, extraction, review state and export APIs.

Review UIBrowser review app

Clickable field boxes on the page image, next-issue navigation, approve and export.

Data layerLocal run store with a fingerprint cache

Each PDF goes to OCR once; repeat uploads cost nothing. No cloud object storage in the prototype.

QualityAutomated tests and a field-level evaluation harness

Tracks the two errors a reviewer would never see: wrong values auto-accepted, and values shown empty.

Business value and evidence

Every number, labelled.

Prototype. Working build on sample or public data. Not yet in customer production.

FigureWhat it measuresLabelSourceLimits

128 of 128fields read correctly across 5 test formsMeasuredOur evaluation harness on 5 fictional sample forms (typed, scanned, handwritten), OCR plus rules.Fictional samples, small set. A customer pilot measures on real forms.

0wrong values auto-acceptedMeasuredSame evaluation.The key safety metric. Must stay at 0 on customer data.

84%of fields auto-accepted, 14% sent to a personMeasuredSame evaluation.Higher on clean typed forms, lower on poor handwriting.

−85%handling time per formTargetedDesign goal for a European payments provider back office (name withheld).Projected from 15 minutes by hand to about 2 to check. Not yet measured in production.

Proven on a hard case

A handwritten German bank detail change form with an empty 'terminal' section filled by a 'same as above' tick box. Read, the tick box honoured, handwriting confirmed by the AI model in the demo, and all 34 fields correct in our OCR-plus-rules test, with no wrong value auto-accepted.

Book a guided demo Talk to us

Try it yourself

See it work in two minutes.

Open the demoIt opens in a new tab, no setup needed.
Ask it something realTry: “Read this handwritten Form A from Cedar GmbH.”
Talk to usWe'll show it on your own data.

Try the live demo

Bank Detail VerificationLive preview

Platform

Read this handwritten Form A from Cedar GmbH.

✓ Detected the formForm A · 38 fields · 2 sites

✓ Checked the IBANChecksum valid · OCR and the AI model agree

36 fields verified. Please confirm the terminal-site town: the handwriting is unclear.

Common questions

Can it read handwriting?

Yes. OCR and the AI model read each value separately. If they disagree, the value is flagged for a person instead of guessed.

Which forms does it handle?

Two German bank-detail change layouts today, detected from their printed headers. New layouts are added as a schema file.

Where does it run?

In your AWS account (Cloud OCR and text extraction and a managed cloud AI service), with no data stored in cloud object storage. Each PDF is sent to OCR once and cached by its hash.

Can we try it?

Yes. We run a live demo on five fictional sample forms, or on your own anonymised forms.