Skip to main content
AI // Document intelligence

Your team didn't go to university to retype invoices.

We build extraction pipelines that read invoices, POs, delivery notes, and contracts, validate the results against your business rules, and send only the uncertain ones to a human.

inference.tracelive

{

"summary"
: "Extract, validate, route exceptions, learn from corrections.",
"documents"
: "Invoices · POs · contracts",
"output"
: "Validated structured data",
"destination"
: "Odoo or any ERP",
"reviewed_by"
: "a human"

}

The extraction pipeline

  1. Step 01

    Ingest

    Email inboxes, uploads, scanners, or API, with file type and size checks.

  2. Step 02

    Extract

    OCR plus a model that returns fields in a strict schema.

  3. Step 03

    Validate

    Business rules: totals add up, vendor exists, PO matches, dates make sense.

  4. Step 04

    Route

    Confident results flow to the ERP; uncertain ones go to a review queue.

  5. Step 05

    Learn

    Reviewer corrections feed the evaluation set for the next model update.

A schema, not a guess

schemas/invoice.pypython
class InvoiceLine(BaseModel):
    description: str
    quantity: Decimal
    unit_price: Decimal

class Invoice(BaseModel):
    vendor_vat: str | None
    invoice_number: str
    issue_date: date
    currency: Literal["USD", "EUR", "GBP", "INR", "CAD"]
    total: Decimal
    lines: list[InvoiceLine]
    confidence: float = Field(ge=0, le=1)

Try it

The AI document sandbox in our labs shows the validation and review flow on sample documents.

Why document work is still manual

Most businesses already have the data they need; it just arrives as PDFs, scans, and email attachments. Someone opens each file, reads the vendor, the dates, the lines, and the totals, and types them into the ERP. It is slow, it is boring, and every keystroke is a chance for a typo that surfaces weeks later as a payment dispute or a stock error.

Older OCR tools tried to fix this with templates: draw a box where the invoice number lives, repeat for every vendor layout. That works until a supplier changes their template or a new one arrives, which is to say, constantly. Modern language and vision models read documents the way a person does, understanding that “Inv. No.”, “Invoice #”, and “Rechnungsnummer” mean the same thing, without a template per layout.

The model is only half the system, though. A production pipeline also needs validation that catches the model's mistakes, a review queue for the uncertain cases, a clean hand-off into your ERP, and a record of what happened to every file. That surrounding engineering is what we build.

Documents we typically handle

Every document type gets its own schema and its own rules.

DocumentFields extractedTypical checks
Vendor invoicesVendor, tax IDs, dates, lines, taxes, totalsLines sum to total; vendor exists; PO and receipt match
Purchase ordersBuyer, items, quantities, prices, delivery datesItems exist in the catalog; prices within agreed ranges
Delivery notesShipment references, items, quantities, signaturesQuantities match the order; no duplicate receipts
ContractsParties, dates, renewal terms, key clausesRequired clauses present; dates consistent
Bank and card statementsTransactions, dates, amounts, referencesOpening plus movements equals closing balance

Built for real-world documents

Mixed formats

Native PDFs, scans, phone photos, and email bodies, each routed to the right extraction path.

Confidence scoring

Each field carries a confidence score, so thresholds decide what is auto-approved and what is reviewed.

Review queue

Reviewers see the document beside the extracted fields, fix what's wrong, and approve in a click.

ERP hand-off

Approved data becomes draft vendor bills, purchase matches, or records in Odoo or your ERP via API.

Data controls

Retention, redaction, and hosting choices decided up front, not discovered later.

Accuracy reporting

Per-field accuracy and auto-approval rates tracked over time, by vendor and document type.

How a project runs

  1. Step 01

    Sample and baseline

    Week 1

    Collect a representative set of your real documents and measure how long they take today.

  2. Step 02

    Schema and rules

    Week 1–2

    Agree the fields, formats, and validation rules with the people who use the data.

  3. Step 03

    Evaluation set

    Week 2

    Label a held-out set of documents so accuracy is measured, not estimated.

  4. Step 04

    Pilot

    Week 3–6

    Run the pipeline on live documents with every result reviewed, and tune thresholds.

  5. Step 05

    Scale

    Ongoing

    Raise auto-approval where accuracy proves it, add document types, and hand over monitoring.

Scope at a glance

Included

  • Extraction schemas and validation rules per document type
  • Review interface and exception queues
  • ERP integration and duplicate detection
  • Accuracy evaluation and monitoring dashboard

Not included

  • Guaranteed accuracy before seeing your documents
  • Replacing your accounting approval policy
  • Handwritten forms without a pilot to test them

Document AI myths

Myth: AI extraction means no human ever looks again.

Reality: The goal is fewer, faster reviews. Uncertain documents and new vendors still get human eyes, by design.

Myth: You need thousands of labeled documents to start.

Reality: Modern models work from a schema and examples. A few hundred real documents are enough to evaluate and tune a pilot.

Myth: A high average accuracy is good enough.

Reality: Averages hide problems. What matters is accuracy on the fields that cost money when wrong, such as totals, tax, and bank details.

Common pitfalls we help you avoid

The first pitfall is judging the project on a handful of clean sample documents. Real inboxes contain skewed scans, multi-page invoices, credit notes, and emails with three attachments, so we evaluate on a representative sample from the start.

The second is automating approval before the numbers justify it. We begin with every result reviewed, then raise the auto-approval threshold per document type only when measured accuracy supports it.

The third is forgetting the people who use the data. Finance, purchasing, and operations agree the fields and rules up front, so the output fits how they already work and nothing has to be reformatted downstream.

What we measure

MeasureWhy it matters
Per-field accuracyShows exactly which fields need better rules or review
Auto-approval rateHow much work no longer needs a person
Review time per documentWhether reviewers are actually faster
Exceptions by reasonWhich vendors, formats, or rules cause the most rework

Who does what

We handle

  • Pipeline, models, and validation rules
  • Review interface and ERP integration
  • Evaluation, monitoring, and tuning

Your team handles

  • Sample documents and field definitions
  • Reviewers during the pilot
  • Approval policy and final sign-off

What confidence scores do and don't tell you

Microsoft's Document Intelligence documentation describes field confidence as an estimated probability that a prediction is correct, with 0.95 meaning it's likely correct 19 out of 20 times, and recommends using confidence to decide between automatic acceptance and human review where accuracy is criticalSource [1]. At thousands of documents, even high confidence leaves errors to catch.

AWS's integration of human review with Textract shows the standard trigger patterns: review when specific fields fall within a confidence range, when required fields are missing, and on a random sample regardless of scoresSource [2]. We implement the same patterns in your own system, so the routing rules survive a change of model or provider.

Frequently asked questions

What accuracy can we expect?
It depends on document quality and variety. We measure per-field accuracy on a sample of your real documents before committing to targets.
Do you store our documents?
Retention is your decision and is designed in. Pipelines can process and discard files, or archive them in your storage.
Does it handle handwriting and scans?
Printed scans work well; handwriting varies. The evaluation phase tells you where the limits are.
Can it post directly to Odoo?
Yes: vendor bills, purchase matching, and attachments created through Odoo's API.
What about multiple languages?
Modern models handle many languages. We test each language you receive.
How do you handle duplicate invoices?
By checking vendor, invoice number, amount, and date against existing records before anything is created, and flagging near-matches for review.
Can we start with one document type?
Yes, and we recommend it. Vendor invoices are a common first choice because volume is high and the rules are clear.
What happens when a vendor changes their layout?
Nothing breaks, because there are no per-layout templates. Accuracy is monitored per vendor, so a drop is visible quickly.
Can extracted data be audited later?
Yes. Each record keeps a link to its source file, the extracted values, any reviewer changes, and who approved it.
Which models do you use?
A mix of OCR, vision, and language models chosen per document type and per your hosting and data policy, including private deployments.
How long does processing take per document?
Typically seconds to tens of seconds, depending on page count and model choice. Batch volumes are processed in parallel.
Can it split a PDF that contains several documents?
Yes. Page classification separates combined scans into individual documents before extraction.
What does it cost to run?
Running costs scale with volume, page count, and model choice. We estimate cost per document during the pilot so you can compare it with manual effort.
Why not auto-approve anything above a high confidence score?
Because high confidence still means some errors at volumeSource [1]. Combining confidence with business rules and random sampling catches what scores missSource [2].
Do you lock us into one AI provider?
No. Routing rules, review screens, and audit trails live in your system, so models can be swapped as they improve.