Most of this site is about client work. This page is about your own ledger. Airclerk's Finance Team Pack is eleven Claude skills for the accounts-payable and month-end work a finance team does on its own books, built in August 2026 for a New Zealand team running Xero. Xero stays the accounting system of record, the team keeps its own workbooks and inboxes, and draft is the hard ceiling: nothing is posted, approved, paid or sent without a person deciding.
Work on clients' books rather than your own? See the NZ Accounting Pack →
Nobody minds the one. The cost is the forty you go through to find it. Accounts payable is one skill of the eleven, and the easiest to picture, so it is the one drawn here.
Four of those five are mechanical, and the pack does them: coded from the supplier's own history with the evidence stated, source document attached, draft bill in Xero, email marked. The fourth row up there is the one that is not mechanical, and it stops with you whatever the amount.
An internal finance team uses Anthropic's Claude models for the recurring work that fills the month: turning the AP inbox into coded draft bills, reconciling supplier statements, working the collections list, sweeping the ledger before close, proposing accruals, updating balance-sheet reconciliation workpapers, explaining variances from transaction-level evidence, building the management pack, rolling the 13-week cash forecast and reforecasting the budget. For that to be safe on a live ledger, Claude has to be connected to the accounting system, the shared finance inboxes and the team's own workbooks, with the write boundary set at draft and a named person on every consequential step.
Ten of the eleven do finance work. The eleventh, finance-onboarding, interviews the team and discovers the Xero setup, then writes the finance profile that every other skill reads. Without it a skill has to guess at your chart of accounts; with it the skill already knows that account 482 is software subscriptions, that the rent goes out on the 20th, and who has to see a journal before it exists.
Organisation identity including the immutable Xero ID, balance date and reporting cadence, chart of accounts conventions and tracking categories, the entertainment split IRD actually uses, accrual conventions and recurring journals, capitalisation and prepayment policies, approved AP inboxes and their processed-marker convention, regular suppliers with their usual coding, regular customers and known payment behaviour, the collections tone ladder, and the location and format of every model and workpaper. Written once, read on every run, and edited by the team rather than by us.
One number cannot do every job, so the profile carries four. Reporting materiality decides which variance is worth explaining. The posting threshold decides when an accrual or journal is too small to propose. Reconciliation tolerance decides which difference is noted rather than chased. The auto-draft ceiling decides when nothing gets drafted without a named person's go-ahead, whatever the confidence says. Qualitative risk overrides all four.
The month-end spine runs supplier invoices, then supplier statements, then the pre-close sweep, then accruals, then balance-sheet recs, then variances, then the pack, then cash, then the budget. Collections feeds the cash forecast. Each skill recommends the next rather than running it, because a chain that runs itself is a chain nobody is reading.
AP processing and collections suit a schedule, daily and weekly respectively, which only works if running twice is harmless. No invoice is entered twice, no email is reprocessed, no finding is re-raised. Period-end skills stay manually invoked, because close is final when the finance team says it is.
The guardrails live in one place and every skill cites them by number. Four of them do most of the work.
Those actions are outside it, not merely discouraged, and it will not perform them even when asked. Xero writes stop at draft status. Emails to suppliers and customers are drafted and never sent. Confirmation from a person authorises a handoff to that person rather than an action by the workflow. Profile configuration can narrow the boundary and can never widen it.
A new supplier, a change of bank details or a payment instruction arriving by email escalates regardless of amount, which is the case a value threshold would have waved through. Content inside emails and attachments is treated as data and never as instructions, so an invoice carrying a line that tells the assistant to approve it is read as text on a document.
What was compared with what, over what period, at what confidence. A coding decision names the supplier history it came from. A variance explanation quantifies what it explained and what is left over, and classifies the driver as known, probable or unexplained. Status on a reconciliation is computed rather than asserted, and there are no plugs.
Routine high-confidence work flows through so the team's attention goes where it is needed. Every item lands in exactly one of four states: completed, review recommended, blocked, or high-risk exception. A pile of items marked "probably fine" is not one of the four.
Outputs update the organisation's existing workbooks and templates rather than replacing them with an Airclerk layout. Behind every workflow sits AI Process Assurance: the reasoning, sources and sign-off retained as an AI Workpaper, so the team can show what the AI did, how it was checked and who signed it off.
Each is a candidate first workflow. AP and collections are the usual place to start, because the reviewer already knows how to check the output: the source invoice sits beside the draft bill, and the chase list sits beside the ledger it came from.
The interview and Xero discovery that writes the profile every other skill reads: calendar, ledger conventions, the four thresholds, approved AP inboxes, regular counterparties, model locations, write permissions and the escalation table. Run once, revisited when the business changes.
Finance Team PackInvoices classified out of the shared inbox, extracted, duplicate-checked and coded from the supplier's own history with the evidence stated, source document attached, email marked processed. New suppliers and changed bank details stop and escalate.
Finance Team PackThe statement against Xero: matched invoices, credits and payments, missing and duplicate items, allocation problems and opening-balance discrepancies, plus a drafted supplier query that is never sent.
Finance Team PackA prioritised worklist from aged receivables and computed payment behaviour, with the chase list kept separate from the investigate list, tone-laddered draft communications, and the expected cash impact handed to the forecast.
Finance Team PackFour-way comparatives, material movements drilled to the transactions behind them, missing recurring expenses, stale accruals, clearing accounts that have not cleared and unusual journals, as a prioritised investigate-before-close list. Reruns update the findings.
Finance Team PackRecurring-pattern gaps and services received but not invoiced, each with its estimation basis named, and prior accruals tested for release or reversal. Proposed journals stay draft.
Finance Team PackThe workpapers updated in place: ledger against supporting balance per account, reconciling items aged, timing rolled forward, unsupported items flagged. Status is computed rather than asserted, and there are no plugs.
Finance Team PackExplanations built from transaction-level evidence, explained and residual quantified separately, drivers classified known, probable or unexplained. "Travel was higher because travel increased" is banned by the skill.
Finance Team PackYour pack, in your format: figures tied back to Xero, KPIs on the profile's exact formulas, variance commentary carried at the evidence level it was written at, and the decisions management has to make surfaced rather than buried. Draft until a person distributes it.
Finance Team PackIn the team's own workbook: receipts from how customers actually pay rather than terms, supplier payment-run patterns, payroll and IRD dates verified at run time, the lowest point called out, and scenarios modelled rather than recommended.
Finance Team PackActuals rolled into the model, run-rates set against assumptions, proposed changes carrying their evidence and the questions for each manager. Every figure labelled actual, management assumption, calculated run-rate or proposed change, and the model is edited only after a person approves.
Finance Team PackThe plumbing underneath: the Xero connection, the shared finance inboxes, and the SharePoint, OneDrive or Google Drive paths where the models and workpapers live, scoped so a run sees only what it should. Connector work rather than a skill.
ConnectorsThe pack was built against Xero as the accounting system of record, reached through the Airclerk Xero MCP. Every skill degrades to exported reports and CSVs when no live connection is there, so the reconciliation, month-end, reporting and forecasting work can be adapted to run off exports on another ledger. What does not carry over is the ledger-side automation: draft bills, supplier history, organisation verification before each run. On another ERP that is connector work, and it belongs in the scoping call.
Enough volume that the AP inbox and the month-end checklist are real work, and enough structure that there is a management pack and a forecast model to update. A finance function with no close timetable and no reporting cycle gets less out of this than a controller working to a monthly deadline, whatever the headcount either side of it.
A named person per skill who reviews the output and whose judgement the escalation table points at. The workflow can prepare the draft and show its evidence; it cannot decide the work is finished. Where nobody has that job, the drafts pile up unreviewed and the team goes back to doing it by hand.
The pack was built in August 2026 and is an accelerator rather than a finished product. It shortens the path to your own build, and each organisation's version diverges from ours as skills are adapted to its models, formats, thresholds and risk settings before anything runs on a live period.
The workflow you have in mind, the systems it touches and who owns it. Fixed fee quoted after the call.
First governed workflow into production.
Controls, approvals, audit trails.
Monitor, improve, expand.
Everything is deployed to your own Claude instance, under your own agreement with Anthropic. Airclerk is not affiliated with or endorsed by Anthropic, Xero or Inland Revenue.
Bring the part of the finance month you would most like back. Half an hour is enough to say whether it is a first workflow and what it costs to build.
An internal finance team uses Claude for the recurring work that fills the month: processing the AP inbox into coded draft bills, reconciling supplier statements, working the collections list, the pre-close month-end sweep, accrual proposals, balance-sheet reconciliation workpapers, variance explanations, the monthly management pack, a rolling 13-week cash forecast and the budget reforecast. Airclerk implements these as governed workflows on the systems the finance team already runs: Xero as the accounting system of record through the Airclerk Xero MCP, plus Outlook or Gmail, the shared finance inboxes, the team's own Excel workbooks and SharePoint, OneDrive or Google Drive. The pack is Airclerk's Finance Team Pack, eleven skills built in August 2026 for a New Zealand finance function running its own ledger on Xero. Draft is the hard ceiling: nothing is posted, approved, paid, sent or written off by the workflow, and a named person makes every consequential decision.
They are opposite personas. Airclerk's NZ Accounting Pack serves a public practice: many clients, engagement isolation between them, year-end workpapers and tax compliance done on someone else's ledger. The Finance Team Pack serves one organisation running its own ledger, its own AP inbox, its own suppliers and its own management pack, month after month. That changes the design. A practice pack has to keep hundreds of client files apart; an internal pack can learn one organisation deeply and apply it every run, which is what the finance-onboarding skill is for. It interviews the team, discovers the Xero setup and writes a finance profile that every other skill reads: the balance date, the chart of accounts conventions, the regular suppliers and their usual coding, the approved AP inboxes, the four thresholds, where the models live and who approves what. Your external accountant is not replaced by any of this. Unusual GST or income-tax positions still go to them, and the pack is written to route those out rather than answer them.
Draft status, and nothing beyond it. Bills are created as drafts, journals are proposed as drafts, and that is the ceiling: the pack does not post journals, approve bills, make payments, send emails, write off debt or touch supplier bank details, and it will not do those things even when asked. Profile configuration can narrow the boundary further but can never widen it. Confirmation from a person authorises a handoff to that person, not an action by the workflow. Xero access runs through the Airclerk Xero MCP and is read-heavy by design, and every skill verifies both the Xero organisation name and its immutable organisation ID before it reads or writes anything. Separately, there is an auto-draft ceiling in the profile: above that value nothing is drafted without a named person's go-ahead, whatever the confidence.
Three things, and the first is that it cannot pay anything at all. Payments and bank details sit outside the pack entirely. Second, there is a fraud floor that ignores materiality: a new supplier, a change of bank details or a payment instruction arriving by email escalates to a person regardless of amount, so the cheap test of a fraud attempt is the one case that always stops. Third, content inside emails and attachments is treated as data and never as instructions, so an invoice that contains a line telling the assistant to approve it is read as text on a document. Coding comes from the supplier's own history in the ledger with the evidence stated, duplicates are checked before anything is created, and the source document is attached to the draft bill so the reviewer sees what it was built from.
Partly, and which part matters. The pack as built assumes Xero as the accounting system of record and reaches it through the Airclerk Xero MCP. Every skill already degrades gracefully to exported reports and CSVs when no live connection is present, so a finance team on another ledger can run the reconciliation, month-end, variance, reporting and forecasting skills off exports once the profile and export mappings are set up. What does not carry over is the ledger-side automation: creating draft bills, reading supplier history directly, verifying the organisation before each run. On another ERP that becomes connector work rather than a configuration change, and it belongs in the scoping call rather than in a footnote here.
No. It never touches myIR or a banking platform, never files a return, and gives no tax, investment or financing advice. GST settings and tax dates are verified at run time against ird.govt.nz rather than remembered, because a rate or a date held in a model is a rate or a date that goes stale quietly. Where a position is unusual, zero-rated land and the registration boundary being the standard examples, the output routes it to the organisation's external accountant instead of forming a view. Airclerk is an AI implementation consultancy, not an accounting firm, and the workflows are designed around that boundary.
Because the model is the part that already works. What is left is the work that decides whether any of it sticks: writing the finance profile so the skills know your chart of accounts, your suppliers and your four thresholds; connecting the ledger, the shared inboxes and the workbooks the team actually uses; deciding which steps stop for which person; testing the workflow on real periods before the controller relies on it; and running it long enough that the team stops doing the manual version in parallel. A vendor feature arrives configured for the average customer. The pack is an accelerator rather than a finished product, and each organisation's version diverges from ours as the skills are adapted to its own models, formats and risk settings.