How we work

From engagement to disposal, every step leaves a record.

A process built for work that may be challenged: chain of custody on intake, human review in the middle and disclosed coverage at the end.

The Lastroo method: seven steps and two loops.

A missing document comes back as a request before the audit, not as an assumption inside the report.

10% The question
  1. Engagement.

    Scope, period, type of work and sponsor. NDA and data processing agreement.

  2. Intake.

    A folder we control. Every file gets a hash on arrival.

  3. Cataloging and triage.

    Every document identified. What's missing becomes a list.

    ↺ Loop 1 A gap found during triage becomes a new request. ↑ back to step 2
  4. Calibration.

    We configure the agents for the type of case or around the client's own criteria. This is where the team's experience comes in.

80% The reading
  1. Reading and cross-checking.

    Every document read, then compared with the others: dates, amounts, names and users.

10% The verification
  1. Human review.

    Every finding checked against the source before the report.

    ↺ Loop 2 A finding that depends on a document not received becomes a request, not a conclusion. ↑ back to step 2
  2. Delivery and disposal.

    Report, walkthrough meeting and disposal of the data, with a record.

What a finding looks like.

Fact, cross-referenced documents with pages, criterion and next step.

G-07 High
Illustrative example

Supplier registered and approved by the same user

Fact
[Supplier A] was registered by user purchasing.03 on 12 March. The same user approved the next three orders to that supplier, between 14 March and 2 April.
Documents cross-referenced
DOC-0412 · p. 3DOC-0518 · p. 1DOC-0519 · p. 1DOC-0520 · p. 1
Criterion
Purchasing Policy, item 4.2: supplier registration and order approval by different people.
Next step
Inquiry with the purchasing team and a preventive block on new orders until it is resolved.

Coverage statement.

Generated by the system from each document's record, not from the model's text. The report doesn't say “we analyzed everything”. It says what was read, what wasn't, and why.

Coverage statement Illustrative example
Documents received 3,215
  • Read in full 3,162
  • Read by computer vision model 38
  • Out of scope 11
  • Illegible, resubmission requested 4
Total 3,215 of 3,215 with a final status

Client documents are handled as evidence.

  • Data processing agreement
  • Documents never train models
  • Chain of custody by hash
  • Sensitive data masked
  • Mandatory human review
  • Disposal with a record

FAQ

What does cross-checking find that reading doesn't?
Patterns across documents: the same person in roles that should be separate, dates that don't add up, repeated amounts, shared shareholders.
Does the AI decide what's irregular?
No. It reads, organizes and flags with references. The responsible professional, on your team or ours, qualifies and signs off.
How do I know everything was read?
Every document ends with a status (read, out of scope or illegible), and the coverage statement is built from that record.
Do you sell software?
No. We deliver finished work. We charge per case, or a setup fee plus a monthly fee when the volume is recurring. Pricing depends on the size of the document set and the goal.
What if the document set is too large?
We process it in batches. The report says what was read and what wasn't. Illegible files go through image reading or are read by hand; if they stay illegible, we ask the client for another copy.
What formats do you accept?
PDFs, scans, spreadsheets, electronic invoices (XML), system exports. Native formats are preferred.
Do you need access to our systems?
Not necessarily. The minimum is an export to a controlled folder.
Does this replace an independent audit?
No. It complements it, giving a complete and traceable reading to whoever issues the opinion.

Start with step zero.

Send us the scope and type of work. We'll reply with an initial document list and what to expect from the document review.

Discuss a case