Skip to content
Gallucci One Finance

AI inside G1

It has your numbers, your permissions and your audit trail.So it works inside the system, not beside it.

AI in G1 is not a chat window bolted onto a product. It reads the same ledger your people read, it cannot see anything they could not see, and every question it answers is written down. That is what makes an answer worth acting on.

See what it doesAsk an agent

6 agents are designed and specified in the catalogue below. 1 is in preview; the rest are in development or planned. Each one says so on its own card.

Six things, one system

Ask. Analyse. Automate. Monitor. Predict. Assist.

Six verbs, not six products. They run on the same ledger, the same permissions and the same audit trail as everything else in G1.

A question in your own words, answered from your ledger — with the journals behind the answer.

Ask G1 — FinanceSample data
Why did gross margin fall this month?

G1 Agents

Specialists, not one assistant that claims to do everything.

Each agent has a role, a department, a defined set of capabilities and — more to the point — a defined set of things it is not allowed to do.

How permissions work

Choose an agent

Ask G1 — Finance analystIn development

Scripted preview — sample data. The live agent is not connected to this website. Nothing here reads real business data.

Explains what the financial data is saying

Answers questions about the numbers, and shows its working.

Pick one of the questions below and watch the answer get built — the working, the figures behind it, and where each number came from.

Inside these limits

  • Read-only on ledger data the user can already see
  • Cannot post, approve or change any document
  • Every question and answer written to the audit trail

Suggested questions

Free text is answered honestly: the preview will tell you it cannot answer rather than invent something.

  • In development

    Finance

    Finance analyst

    Explains what the financial data is saying

    Ask it why a figure moved and it goes to the journals behind the figure. It answers in plain language, lists the documents it used, and tells you when the data is not good enough to answer.

    • Explains movements by drilling into the underlying postings
    • Compares periods, entities and dimensions
    • Builds a short written commentary for a management pack
    • Says when it does not know, instead of guessing

    Inside these limits

    • Read-only on ledger data the user can already see
    • Cannot post, approve or change any document
    • Every question and answer written to the audit trail

    Reads from

    Voice. Direct and specific. Numbers with context, never adjectives instead of numbers.

    Ask finance analyst a question
  • In development

    Finance operations

    Reconciliation

    Matches transactions and clears the unmatched list

    It reads imported statement lines against the ledger, proposes matches with a reason, and puts the genuinely ambiguous ones in front of a person. It never posts a match on its own.

    • Proposes matches between statement lines and ledger entries
    • Learns the patterns a business actually uses
    • Splits and part-allocates receipts across invoices
    • Escalates anything it is not confident about

    Inside these limits

    • Reads imported bank batches and open ledger items
    • Proposes matches; a person confirms before anything posts
    • Cannot alter a posted transaction

    Reads from

    Voice. Short, procedural, evidence-first. One line per proposal.

    Ask reconciliation a question
  • In development

    Finance

    Reporting

    Prepares recurring reporting and flags the differences

    It builds the recurring pack from the ledger on the schedule you set, checks that the subledgers still reconcile, and writes a note on what moved since last period.

    • Produces the recurring report pack on schedule
    • Checks subledger to control account agreement before publishing
    • Writes a short variance note against budget and prior period
    • Holds the pack back if a reconciliation does not clear

    Inside these limits

    • Reads posted data only
    • Publishes to a draft state; a person releases it
    • Cannot change a period status

    Reads from

    Voice. Plain and formal enough for a board pack. No hedging, no filler.

    Ask reporting a question
  • Planned

    Credit control

    Receivables

    Prioritises collections and drafts the chase

    It ranks the open receivables ledger by what is actually recoverable and what it is worth, drafts the follow-up, and keeps the history against the customer so the next conversation starts from the truth.

    • Ranks open items by value, age and payment behaviour
    • Drafts follow-up messages in the right tone for the account
    • Flags accounts where credit exposure has moved
    • Logs every contact against the customer record

    Inside these limits

    • Reads the receivables ledger and customer history
    • Drafts communications; a person sends them
    • Cannot write off, credit or adjust a balance

    Reads from

    Voice. Firm and courteous. Specific about amounts and dates.

    Ask receivables a question
  • Planned

    Operations

    Operations

    Monitors process flow and surfaces blockages

    It watches documents and processes move through the business and reports where things are waiting: approvals nobody has looked at, orders received but not billed, jobs open past their planned finish.

    • Tracks documents waiting on an approval or a match
    • Highlights work sitting longer than it usually does
    • Reminds the owner, then escalates on a rule
    • Reports where the process actually slows down

    Inside these limits

    • Reads process and document status
    • Sends reminders and escalations by rule
    • Cannot approve anything on a person's behalf

    Reads from

    Voice. Operational and unfussy. Names the step and the owner.

    Ask operations a question
  • Preview

    Data

    Anomaly watch

    Continuous monitoring against the business's own baseline

    It compares what posted today against what this business normally does — not against a generic benchmark — and raises the things that do not fit, with the evidence attached.

    • Learns the business's own normal ranges
    • Flags duplicates, outliers and timing breaks
    • Explains why something was flagged, in one line
    • Routes each observation to the person who can act

    Inside these limits

    • Reads posted data across permitted companies
    • Raises observations; changes nothing
    • Observations are logged and dismissible with a reason

    Reads from

    Voice. Neutral and factual. States the observation, then the evidence.

    Ask anomaly watch a question

Availability is stated on each card and none of these agents is generally available today. The conversations above are scripted previews built from the same content that defines the agent — they are not a live model, and no real business data reaches this page.

The rules it works under

The interesting part of AI in an accounting system is what it is not allowed to do.

These are not policies written for a website. They are constraints enforced in the database, under the application, where no interface can route around them.

  • Permissions

    Roles that mean something: owner, admin, accountant, clerk, viewer, approver.

  • Audit trail

    Posted documents cannot be edited. Corrections are new documents.

  • Approvals

    Maker-checker, enforced where it cannot be argued with.

  • Data ownership

    Your data, in a standard shape, exportable from anywhere.

And what G1 does not claim is written down in the same place — no certifications, no uptime figures, no customer counts.

Read the trust page

Talk to G1

Ask it something about your own numbers.

Tell us the question you would want answered on a Monday morning. We will tell you plainly whether G1 can answer it today, or whether it is on the list.

Talk to G1Build Your G1