Skip to content
Giallo

AI for auditors: more time for judgment, less time chasing paperwork.

How does AI help an auditor?

AI for auditors works on gathering and first-pass review of evidence: it follows up document requests to the client, prepares and reconciles external confirmations, matches supporting documents to accounting entries, and drafts working papers and opinions. Giallo Studio builds these agents with Claude, GPT and Gemini; professional judgment and signature stay with the auditor.

Frequently asked questions, straight answers.

By starting from what it does not do. It does not choose the sample, assess risks or form the opinion, which stay with the auditor. Within these limits it gathers evidence, follows up external confirmations and prepares draft working papers for review.

Mainly, prepare the groundwork for opinions. The audit body issues opinions on the budget, budget changes and other acts: the agent reads proposals and attachments, checks the consistency of the schedules and prepares a draft with the points to verify.

Yes, as a documented tool. Drafts produced by the agent enter the audit file only after the auditor's review, and every step leaves a record of what was read and proposed.

The agents see only the engagement file, and every access is logged in the file itself.

For long documents and reasoning about consistency across schedules, we usually use Claude Opus; for scanned PDFs and complex tables, Gemini; for extracting data from confirmations, GPT. None of the three draws conclusions on the financial statements: they flag inconsistencies between schedules and notes, which the auditor then investigates.

No. The agents work alongside the software you use, on exported files and engagement folders, and deliver sheets and drafts in your formats.

Sources of answers: What can AI do for local government auditors?

The AI agents we design, one by one.

The external confirmation agent at work: requests leave the audit file for the bank, customer, supplier and lawyer and come back with the reply; each confirmed balance is set beside the ledger account, checked off if it matches and circled if there is a difference the auditor must investigate.

Document request agent

Manages the client request list, recognizes the documents received and prepares reminders.

From acceptance of the engagement to the end of fieldwork.

Connects to: Folder shared with the client · Office email · Request list in Excel · The firm's audit software

  1. Starts from the request list the auditor set for that engagement.
  2. Recognizes each file received: trial balance, general ledger accounts, bank statements, contracts, minutes.
  3. Matches it to the list item and flags when two versions of the same document don't agree.
  4. Updates the status of the list and prepares targeted reminders with the exact list of what's missing.
Claude Haiku
Quickly classifies many incoming files and messages, recognizing the document type from its content and the text that comes with it.
Gemini Flash
We evaluate it during testing when the exchange with the client runs through Google Drive, to work directly on the shared folders.

Who approves. The auditor decides the request list; reminders go out after the auditor reads them or with a text the auditor has already approved.

External confirmation agent

Prepares the confirmation requests, tracks the replies and proposes the reconciliation with the books.

After the auditor has selected the sample of balances to confirm.

Connects to: Ledger accounts and accounting extracts · Certified email (PEC) · Reconciliation sheets · Engagement file

  1. Fills in the request letters for banks, customers, suppliers and lawyers from the sample the auditor specified.
  2. Logs send dates, reminders and replies, so the audit file stays complete.
  3. Extracts balances, open items and additional information, such as guarantees or litigation, from the replies.
  4. Compares the confirmed balances with the ledger accounts and proposes the reconciliation, flagging the differences to investigate.
GPT
Replies from banks and counterparties come in very different formats: this calls for structured extraction with output checked field by field.
Codex
Writes and maintains the scripts that read the ledger accounts exported from the client's accounting software and prepare the reconciliation sheets; it does not read the confirmations.

Who approves. The auditor chooses the sample, and the letters go out under the auditor's control. The auditor investigates the differences and decides whether they are errors or normal reconciling items.

Supporting documents testing agent

Matches orders, delivery notes, invoices and accounting entries in the sample and flags what doesn't add up.

During work on financial statement items, when the auditor performs tests of details.

Connects to: E-invoices and delivery notes · Extracts from the client's accounting software · Test sheet · The firm's audit software

  1. Receives from the auditor the sample of entries to test.
  2. Reads the supporting documents provided by the client and extracts date, counterparty, amounts and references.
  3. Matches each entry to its documents and checks dates, amounts and period.
  4. Prepares the test sheet with the outcome of each item and the exceptions highlighted.
GPT
Extracts comparable data from mixed documents and lets you check every field before it enters the test sheet.
Claude Sonnet
Describes exceptions clearly and in measured terms, without labeling them as errors or irregularities: that assessment belongs to the auditor.

Who approves. The auditor reviews every exception and signs the test sheet. The agent does not decide the sample size or the test outcome.

Audit body opinion agent

Reads the proposal and attachments, checks the consistency of the schedules and prepares a draft opinion with the points to verify.

When the local authority sends the audit body a draft resolution that needs an opinion.

Connects to: Draft resolutions and attachments in PDF · The authority's financial schedules · The auditor's opinion templates · Certified email (PEC)

  1. Recognizes the type of act: budget, budget change, recognition of off-budget debts, new borrowing.
  2. Reads the proposal, the attachments and the report of the head of financial services.
  3. Checks that totals, balances and cross-references between schedules are consistent with each other.
  4. Prepares the draft opinion following the template chosen by the board or the sole auditor, with the points to verify highlighted.
  5. Lists the clarifications to request from the offices before finalizing the opinion.
Claude Opus
Holds very long proposals and attachments together and reasons about consistency across different schedules: this is the task that needs the most capable model, at low frequency.
Gemini
Reads scanned PDFs and tables with complex layouts well, which are common in the attachments sent by local authorities.

Who approves. The opinion is an act of the audit body: the auditor writes it, edits it and signs it. The agent prepares the groundwork.Source

Your engagements, the first agent

Tell us what engagements you handle, companies or local authorities, and where the work piles up between requests, confirmations and opinions: we'll suggest the first agent to add alongside your procedures.

Prefer the full form? Go to the contact page.

No spam or sales pitches. Response within 1 business day.

We use this data to contact you and, if we work together, to manage the project. We store it in Giallo's internal management system, on servers in the European Union, where only our team can see it; by email we only receive a notice, without your data. If you do not become a client, we delete it 3 months after our last contact. Details and your rights: privacy policy.

Do you prefer to write or call us?

Where time gets lost today.

These are the points we hear about most often. Each one is a possible first agent.

The client request list never closes

At the start of the engagement, the client gets a list of the documents needed. Then come scattered files, different versions of the same trial balance, partial answers, and someone keeps a spreadsheet of what's missing up to date by hand.

An agent receives the files, matches them to the list items, flags inconsistent versions and prepares targeted reminders. The auditor sees at any moment what has arrived and what is holding up the work.

External confirmations are prepared and reconciled by hand

Letters to banks, customers, suppliers and lawyers have to be prepared, tracked and, once they come back, compared with the book balances. Replies arrive in different formats, and the line-by-line comparison takes days.

The agent prepares the letters from the sample chosen by the auditor, logs requests and replies, extracts the confirmed balances and proposes the reconciliation with differences highlighted.

The opinion on the budget change comes with only a few days to spare

In local government, the draft resolution arrives with bulky attachments and the council's deadlines are tight. The audit body has to read everything, check the balances referred to and write a reasoned opinion.

The agent reads the proposal and attachments, rebuilds the changes by mission and program, checks the internal consistency of the schedules and prepares a draft opinion with the points the auditor must verify.

Which AI model we use, and for which task.

We don't tie you to a single provider: we work with Claude by Anthropic, GPT and Codex by OpenAI, and Gemini by Google, and we choose the version during testing, on your own cases.

Claude Haiku

  • Document request agent. Quickly classifies many incoming files and messages, recognizing the document type from its content and the text that comes with it.

Gemini Flash

  • Document request agent. We evaluate it during testing when the exchange with the client runs through Google Drive, to work directly on the shared folders.

GPT

  • External confirmation agent. Replies from banks and counterparties come in very different formats: this calls for structured extraction with output checked field by field.
  • Supporting documents testing agent. Extracts comparable data from mixed documents and lets you check every field before it enters the test sheet.

Codex

  • External confirmation agent. Writes and maintains the scripts that read the ledger accounts exported from the client's accounting software and prepare the reconciliation sheets; it does not read the confirmations.

Claude Sonnet

  • Supporting documents testing agent. Describes exceptions clearly and in measured terms, without labeling them as errors or irregularities: that assessment belongs to the auditor.

Claude Opus

  • Audit body opinion agent. Holds very long proposals and attachments together and reasons about consistency across different schedules: this is the task that needs the most capable model, at low frequency.

Gemini

  • Audit body opinion agent. Reads scanned PDFs and tables with complex layouts well, which are common in the attachments sent by local authorities.

How we choose between models: the best AI model for agents.

How we work with you.

  1. Redo last year's confirmation round

    We take an engagement that's already closed and give the agents the same sample of customers, suppliers and banks the auditor had chosen. The agent's letters, reply log and reconciliations are set beside the archived ones: every difference becomes a rule to fix before the new engagement.

  2. Your procedures, not ours

    The agents follow the lists, templates and working paper models the firm already uses: we don't introduce a different audit methodology.

  3. Traceable evidence

    Every step the agent takes leaves a record of what it read and what it proposed, so the audit file also documents the tool's contribution.

AI agents for auditors and public-sector audit bodies

Which phase of the engagement do you want to lighten?

Tell us what engagements you handle, companies or local authorities, and where the work piles up between requests, confirmations and opinions: we'll suggest the first agent to add alongside your procedures.