Workflow comparison

Fulla Ledger vs. building statement extraction yourself

A vision API supplies model output. A usable statement workflow also needs review, revisions, checks, export and data-lifetime controls. Decide which responsibilities your team needs to own.

Read the comparison

What you would build.

A model response is one part of a usable statement workflow.

Your own stack

PDF handling
Extraction and validation
Source-linked review UI
Revision and approval controls
Output mapping
Retention and expiry

With Fulla

Your PDF intake
Fulla extraction and review workflow
Your reviewer checks and approves
Approved conversion export
Your downstream mapping and archive

A responsibility-by-responsibility comparison.

Fulla Ledger and Building your own extraction pipeline on a vision model API
Requirement	Fulla Ledger	Building your own extraction pipeline on a vision model API
Model selection	A promoted model policy chosen by evaluation, with the eligible routes and their retention posture published.	Your own evaluation, cost control, and provider due diligence.
Deterministic checks	Confidence scoring, amount-sign reconciliation, and balance-equation checks run before a draft is accepted.	You design the checks and the failure thresholds.
Review workflow	Row evidence, correction, split, merge, undo, and approval are part of the product.	You build and test the UI.
Retention	24-hour content lifetime for sources, rows, review events, and exports.	Your storage design and your deletion guarantees.
Privacy handling	Extracted account, card, routing, ABA and IBAN-like values are reduced or masked before scoring, persistence and telemetry. Rendered source pages still reach the selected upstream provider.	Your policy for source images, extracted identifiers, persistence and telemetry.
Operations	Queues, retries, leases, dead letters, and expiry are already built.	Your infrastructure and on-call.

Model selection

Fulla Ledger: A promoted model policy chosen by evaluation, with the eligible routes and their retention posture published.
Building your own extraction pipeline on a vision model API: Your own evaluation, cost control, and provider due diligence.

Deterministic checks

Fulla Ledger: Confidence scoring, amount-sign reconciliation, and balance-equation checks run before a draft is accepted.
Building your own extraction pipeline on a vision model API: You design the checks and the failure thresholds.

Review workflow

Fulla Ledger: Row evidence, correction, split, merge, undo, and approval are part of the product.
Building your own extraction pipeline on a vision model API: You build and test the UI.

Retention

Fulla Ledger: 24-hour content lifetime for sources, rows, review events, and exports.
Building your own extraction pipeline on a vision model API: Your storage design and your deletion guarantees.

Privacy handling

Fulla Ledger: Extracted account, card, routing, ABA and IBAN-like values are reduced or masked before scoring, persistence and telemetry. Rendered source pages still reach the selected upstream provider.
Building your own extraction pipeline on a vision model API: Your policy for source images, extracted identifiers, persistence and telemetry.

Operations

Fulla Ledger: Queues, retries, leases, dead letters, and expiry are already built.
Building your own extraction pipeline on a vision model API: Your infrastructure and on-call.

A choice that fits your work.

Fulla may fit when

Extraction is a means to an end and you would rather not run a model evaluation loop.
You need reconciliation, review, approval, and export behavior on day one.
You would rather not store customer statements at all beyond a short window.

The alternative may fit when

Extraction accuracy is your differentiator and you want to own the model loop.
You already run provider agreements and evaluation harnesses.
Your inputs are not bank statements.

The review decision stays with you.

A draft result is not an approved conversion export. Check source coverage, save corrections and approve the revision you intend to use.

Statements are deleted 24 hours after the upload starts. Save the output you need before then.

Read the API contract

Questions about this choice.

Can I use Fulla Ledger from my own product?

Yes. Workspace API keys with documented rate limits, idempotency requirements, and an OpenAPI document cover upload, status, review, approval, and convert.

Do API results differ from the website?

Website exports and API conversion exports use the same approved persisted revision. GET /api/v1/statements/{id}/result returns the latest saved review revision and does not establish approval.

Which providers can see statement pages?

Only the extraction service named in the privacy notice: Mistral through OpenRouter with zero data retention. No other service receives rendered pages.