AI prepares. Humans approve.
Every agent drafts the work. A qualified human reviews and signs off. Nothing posts to your books without approval. This is the only honest way to run AI on accounting data.
What we do · Automation layer
Six named AI agents. Each one does a specific job. Humans own the sign-off.
We don't sell “AI-powered accounting”. We design, configure, and run named agents on your stack — and every piece of work passes through a qualified human before it touches your books.
The agents
Each agent has a name, a job, a reviewer, and visible proof.
Adding more is a configuration decision, not a marketing one. We ship a new agent when the workflow is well-defined enough that a reviewer can audit it.
Agent
The Receipt Catcher
Routes documents from WhatsApp, email, photo, or PDF.
What it does
Reads incoming receipts and invoices from any channel — operator's phone camera, supplier email, photo dropped in a WhatsApp group — classifies them by category and outlet, and attaches them to the right Xero transaction.
Reviewer
Process manager confirms the classification on any low-confidence match.
What you can see
Source channel · destination Xero record · timestamp · confidence score.
Agent
The Vendor Chaser
Follows up on missing invoices and approvals without anyone having to remember.
What it does
Tracks outstanding supplier invoices, expected delivery notes, and pending approval flows. Sends the chase message to the right person — usually the manager who placed the order — without the bookkeeper having to remind anyone.
Reviewer
Process manager sees the chase log; sets escalation rules per vendor.
What you can see
Vendor · invoice expected · days outstanding · chase history.
Agent
The Close Captain
Runs the monthly close checklist so it doesn't live in someone's head.
What it does
Owns the month-end checklist — bank recs, sales recs, supplier recs, payroll recs, accruals, COGS, GST. Knows what's done, what's blocked, what's waiting on a human. Surfaces blockers in the daily Slack/email digest.
Reviewer
Qualified accountant signs off on each line. The Captain doesn't tick anything itself.
What you can see
Live checklist · per-line owner · blocker reason · ETA to close.
Agent
The Outlier
Flags transactions that don't look right before they become a problem.
What it does
Watches for anomalies: a duplicate supplier payment, an unusually large refund, a category mis-coded against historical pattern, a delivery platform payout that doesn't reconcile within tolerance. Surfaces it for review before the books close.
Reviewer
Qualified accountant reviews each flag; decides post / hold / investigate.
What you can see
Flag · transaction context · historical comparison · resolution log.
Agent
The Categoriser
Drafts expense classification from historical patterns — humans approve.
What it does
Suggests the right account, the right outlet, and the right tax code for new transactions based on how similar transactions were categorised in your books. Drafts only. Nothing posts without approval.
Reviewer
Bookkeeper approves or corrects; corrections feed back into the model.
What you can see
Draft category · top 3 alternatives · confidence · approver.
Agent
The Compass
Drafts the weekly management summary so owners read, not assemble.
What it does
At end of each week and month, produces a plain-language summary of what changed: sales by outlet, cash position, cost trends, exceptions worth knowing about. Drafts the narrative. Owners get a 90-second read.
Reviewer
Account lead reviews the narrative for accuracy before send.
What you can see
Draft summary · linked numbers · revision history.
How a piece of work flows
AI prepares. Humans approve. Nothing posts without sign-off.
Every entry that ends up in your books has been touched by two humans after the agent drafts it — the process manager who designed the workflow, and the qualified accountant who owns the sign-off.
This is the only honest way to run AI on accounting data. It is also the slowest part of the story to fake — which is the point.
How a piece of work flows
machine → human → qualified human
01 · Capture
AI agent runs the work
Routes receipts, drafts categorisation, runs the close checklist, chases missing data, surfaces exceptions.
02 · Review
Process manager checks
Workflow design owner reviews the agent's work. Catches edge cases the agent miscoded. Owns the QA layer.
03 · Sign-off
Qualified accountant approves
Final review and accountability sit with a qualified human who understands the rules and your business. Nothing posts without sign-off.
AI agents prepare the work. Humans own the review and the sign-off. Nothing posts without approval.
How we use AI honestly
Three rules we don't break.
Named agents, specific jobs
Not 'AI-powered accounting'. Six named agents, each doing one job, each with a defined reviewer. You can audit every one.
Built for your stack, not sold as a product
We design and configure these agents around your actual POS, banks, vendors, and team — and we own the operation, not the licence. The agents are part of an engagement, not a product you subscribe to.
What AI is and isn't good at
The honest version.
What AI handles well
Repetitive, pattern-matching admin: classification drafting, chasing reminders, exception detection, summary writing, checklist tracking. It's good at the boring high-volume part.
What AI still struggles with
Edge cases, ambiguous coding decisions, anything requiring judgment about your business or tax position. We don't ask it to do those.
What we never do
Let an agent post to your books without a human in the loop. Claim AI 'does accounting' on its own. Replace qualified accountants. Make outcome claims we can't back up.
See it, don't just read about it
The proof, in three artifacts.
A live owner dashboard, a worked stack teardown, and published case studies. If you want to evaluate the AI claim seriously, these are the artifacts to read.
The stack underneath
The agents run on top of Xero, Aspire, and the rest of the stack we design around. See the full list of platforms we configure on the partners page.
Less admin
Which agent would save your team the most time?
Tell us where the admin is heaviest — receipts, vendor chasing, month-end close, exceptions, categorisation. We'll show you what configuring that agent on your stack actually looks like.