AI can help an NDIS provider categorise income by funding type and reconcile claims by matching bank deposits to the claim batches behind them, then listing the claims that were not paid. It works best on de-identified exports, and it only suggests: you or your accountant approve every entry and own every figure.
Two jobs eat most of a provider's bookkeeping time. One is deciding what each incoming payment is: an NDIA-managed claim, a plan manager's remittance, or a self-managed participant paying an invoice. The other is proving that the money received matches the claims submitted. Both are pattern work, which suits AI, and both involve participants' personal details, which does not.
Why NDIS income is awkward to categorise
The scheme's guide to getting paid describes three ways a participant's budget is managed, and each reaches your bank account differently. For a self-managed participant you send an invoice, they pay it through their own portal, and you send a receipt afterwards. For a plan-managed participant you invoice their plan manager, who enters the claim into the portal. For an NDIA-managed participant you submit a payment request in the myplace provider portal, singly or as a bulk upload, and the guide says you must be recorded as one of that participant's "my providers" before you claim.
So the same service, say a two-hour support session, can land in your bank as a single-line payment from a participant, as a lump from a plan manager covering six participants, or as an agency deposit covering many claims from a bulk upload. Bank descriptions vary by payer and bank, and many carry a reference rather than a name, so a bank-feed rule that works on supplier names does little. The guide also says valid claims from providers recorded as "my providers" are usually paid within two to three business days, and payment can take about ten business days if you are not recorded as one or a claim needs checking. Your cash-flow expectations should follow that, and the bank-feed review should expect lumps.
That is why the books need two layers. The accounting software holds income by funding type and support category, plus the deposits. A claims record, in your provider software or a spreadsheet, holds claim-level detail. Reconciliation means tying the two together. For how provider software prices and what it holds, NDIS software for small providers: per-user costs and what you get is a useful companion.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowSet up the categories before you ask AI for anything
AI suggests categories from the ones you give it. If your chart of accounts has one line called "Income", the best it can do is put everything there. A small provider might use something like this, to be confirmed with your accountant:
| Account | What goes in it | How it usually arrives |
|---|---|---|
| Income: NDIA-managed claims | Payments from bulk or single payment requests | Agency deposits, often one lump per batch |
| Income: plan-managed | Payments from plan managers | Deposits that cover several invoices |
| Income: self-managed | Payments from participants | Individual transfers |
| Income: other (grants, rebates) | Anything outside participant funding | Varies |
| Claims control (clearing account) | Claims submitted but not yet paid | Journal at submission, cleared at payment |
| Refunds and rejected claims | Money returned or claims reversed | Negative lines or adjustments |
The clearing account is the quiet hero. When you submit a batch, you record it as "claimed, not yet received". When the deposit lands, the deposit clears the claim. Whatever is left in the clearing account at month-end is, by definition, claimed and unpaid. It is the list you chase, and AI is very good at helping you build it. Ask your accountant whether this structure suits how you recognise income, because that is their decision, not the AI's.
You may also want sub-lines by support category, for example core supports versus capacity building, if your reports need them. Keep the number of lines small. Each extra line is something the AI has to guess between.
Keep participant details out of the AI tool
Participant data is sensitive. A bank line from a self-managed participant carries their name. A claim record carries a participant name, a participant number, dates of support and item numbers. None of that should go into a general chatbot. The related tutorial on which AI uses are safe under the privacy principles goes deeper, and I have not given legal advice here.
The practical routine is to export only what the task needs and strip the rest, as follows:
- For categorisation, export date, amount, and the bank description, then replace any personal name with a placeholder such as "PERSON" before the file leaves your computer. A spreadsheet find-and-replace on your known participants' names does most of this, and a quick scan of the file does the rest.
- For reconciliation, export from your claims record only: claim batch ID, claim line ID, support item reference, and amount. Do not include participant names or numbers. Use your internal claim ID instead.
- Use an AI account on a business plan with training on your content switched off, and an account that belongs to the business, not a personal login.
- Keep the working files in a folder that only the people who need them can open, and delete exports when the month is closed.
Better still, use the AI built into your accounting software if it covers the task. It is already inside the system that holds the data, which is a lower-risk place than a separate chat window, though you should still check what it sends where, and what the vendor's terms say.
Step one: categorise a month of bank-feed lines
Take the month's bank feed for the business account. Export it as a CSV and strip names. Then give the AI your account list and a few examples of how you categorise. A prompt that works:
You are helping a bookkeeper for a small disability-support provider.
Categorise each bank line below using ONLY these accounts:
- Income: NDIA-managed claims
- Income: plan-managed
- Income: self-managed
- Income: other
- Not income (transfer, refund, owner funds): needs review
Rules:
- Lines from the funding agency are NDIA-managed.
- Lines from a plan manager company are plan-managed.
- A payment from a single person is self-managed unless the description says otherwise.
- If you are not sure, choose "needs review" and say why in 8 words.
Return a table: line number, account, confidence (high/medium/low), reason.
Examples of how I categorise:
"NDIA PAYMENT 2111-xxxx" = NDIA-managed
"PLANCO PAYMENT 4471" = plan-managed
"TRANSFER FROM PERSON" = self-managed
LINES:
[pasted, de-identified lines]A realistic result for a small provider with 31 lines in a month will put most lines in the right account at high confidence, mark a handful medium, and send two or three to "needs review". Those needs-review lines are where it earns trust. A line such as "REFUND ADJUSTMENT 0093" should never be called income. A line such as "TRANSFER FROM PERSON" may be a self-managed participant paying, or a director putting personal funds in. The AI cannot know which, and says so.
What to fix in the first draft
Review in this order: low-confidence lines first, then every line the AI put in "self-managed" (the riskiest category, since it is the one most often a guess), then a random five from the high-confidence group. The usual fixes are a plan manager the AI did not recognise, a refund treated as income, and two payments from the same participant lumped together. Each fix is a candidate for a bank rule in your accounting software, so next month the rule catches it before AI is needed. After two months, most of your income lines should be handled by rules, and AI only handles the odd ones.
Step two: match deposits to claim batches
This is where the time is saved. Suppose, as an illustration, a provider submits a bulk upload of 31 claims on the first of the month, totalling $6,482.50. Two business days later the bank shows a deposit of $5,931.10 from the agency. The difference is $551.40. Where did it go?
Export your claims for that batch with an internal ID, support item reference and amount, and ask the AI to find which claims explain the gap:
Below are 31 claim lines (id, item reference, amount) submitted in one batch,
and the deposit received. Find combinations of claims whose total equals the
shortfall of 551.40. List every combination you find, show the sum for each,
and say which one looks most likely if any claim looks unusual (for example a
claim over the usual size for its item). Do not guess: if no combination sums
exactly, say so.
BATCH:
[id, item, amount]
DEPOSIT: 5931.10 BATCH TOTAL: 6482.50 SHORTFALL: 551.40Suppose it returns two combinations: claims 7 and 19 ($312.00 and $239.40), or claims 7, 12 and 24. The match is only a lead. You check the agency's response in the portal, where a rejected or held claim will show a status. The scheme's guide says a claim can be put on hold for review, and in a bulk upload only the claim under review stays open while the rest are paid as normal, which is exactly how a deposit ends up short of its batch. A rejected claim shows a "rejected" status in the myplace provider portal with an error message saying what to fix, and claim and payment enquiries go through the my NDIS provider portal. If claims 7 and 19 were rejected for a mismatched item reference, you now have two claims to fix and resubmit, and your clearing account shows $551.40 waiting. That is exactly what the account is for.
Notice that the AI did arithmetic here, and arithmetic is a weak point. Treat its sums as suggestions and verify them in a spreadsheet with a SUM formula. A faster way, if you already use a spreadsheet, is to ask the AI for a formula or a short script for the matching rather than the answer itself, then run it on your own machine. Google Sheets AI prompts has patterns for that kind of request; tidy the export columns first so the formula has clean amounts to work on.
Plan-manager remittances
Many plan managers pay several of your invoices in one transfer and send a remittance advice listing them. The format varies from one plan manager to the next and the scheme's guide does not set one, so look at what yours actually sends. The matching job is the same, with a twist: a remittance usually quotes your invoice numbers, not claim IDs. Make sure every invoice you send carries a clear, unique number and the right business number, because the scheme's guide says an invoice sent to a plan manager must include one unless you are exempt. When a remittance arrives, copy its lines into a sheet and ask the AI to match them to your open invoices by number and amount, flagging short-pays and invoices not mentioned.
Step three: month-end reports your accountant can use
Once the lines are categorised and the deposits matched, ask for three small reports. Each one can be a short AI-drafted summary, built from numbers you give it, that your accountant reads and corrects. First, income by funding type for the month, compared with the previous month. Second, the unpaid-claims list: every claim in the clearing account older than 14 days, oldest first, with the likely reason if the portal gave one. Third, a short list of anything odd, such as a deposit with no matching batch, a fall in one support category, or a claim older than the 90-day mark the guide mentions, after which extra checks apply.
The unpaid-claims list should drive action, not just reporting. A provider with 12 participants who chases rejected claims within a fortnight rather than a quarter keeps cash moving and avoids claims drifting towards the scheme's two-year limit for submitting payment requests. The scheme says that is the window, so build the check into the monthly routine.
A filled-in month-end summary
Here is what a month-end income summary can look like once the steps above are done. All figures are invented, for a provider with about a dozen participants.
| Line | This month | Last month | Note |
|---|---|---|---|
| NDIA-managed claims received | $11,240.00 | $9,870.00 | Two batches, each matched to its deposit |
| Plan-managed received | $7,415.50 | $8,102.00 | One remittance short-paid by $186.00 |
| Self-managed received | $1,960.00 | $1,960.00 | One invoice still unpaid at 21 days |
| Claimed, not yet paid (clearing) | $738.30 | $0.00 | Three claims, two rejected for item reference |
| Refunds and reversals | -$95.00 | $0.00 | One overpayment returned; own account, not netted against income |
A summary like this tells the owner three things in a minute: money is arriving, one plan manager short-paid, and three claims need fixing. Ask the AI to draft the notes column from your totals, then check each note against the source. It should never invent a reason; if the portal gave none, the note says "reason not stated" and you look it up.
Turning your corrections into bank rules
The best use of AI in this job is to make itself less necessary. Every correction you make in month one is a rule waiting to be written. Most accounting packages let you create a rule such as "if the description contains PLANCO and the amount is paid in, code to plan-managed income". After the first month, list your corrections in a table: what the line said, what the AI chose, what it should have been. Write a rule for any pattern that appears twice. By month three, rules should handle most routine lines, and the AI gets only the leftovers.
Two cautions. Keep rules narrow, matching on a reference prefix, not on a single word that could appear in an unrelated payment. And review the rules every quarter, because plan managers change their bank descriptions and an outdated rule can mis-code income silently for months. Some accounting packages also offer their own automatic reconciliation features that suggest matches; if yours has one, test it on one month first and compare its suggestions with your own, as with any AI.
Five mistakes that look like AI errors but start earlier
- Two batches, one deposit. The bank combined two submissions. The AI cannot match one batch to the deposit, and the fix is to ask it to try both together.
- A claim edited after the export. The claim amount in your record is now different from the one in the portal. Export again the day you reconcile.
- Self-managed receipts mislabelled. A participant paid two invoices in one transfer and the bank line carries only one number. Check against your invoice list.
- A refund coded as negative income. The AI put a returned payment into income, shrinking the month. Refunds need their own account.
- Rounding. A hundred lines of $0.01 differences make a deposit look unmatched. Tell the prompt to allow a one-cent tolerance per line and report the total tolerance used.
What your accountant should review, and how to hand it over
Agree the split of work with your accountant before month one. A sensible division: you or your bookkeeper run the categorisation and matching with AI help, then send a one-page pack containing the categorised lines, the matched batches, the unpaid-claims list, the totals by account and a note of any rules added. The accountant reviews the pack, signs off the treatment of anything unusual, and decides how revenue is recognised and what reaches your tax return. Never let an AI chatbot's answer on tax or funding rules stand in for that advice.
Keep an audit trail of your own: a folder per month with the original bank export, the de-identified file you used, the AI's output and your corrections. If an auditor or participant asks how a claim was handled, you can show the evidence. The sibling tutorial on checking claims against shift notes with AI covers the step before this one, and how plan managers check invoices against budgets shows what the other side of your invoice looks like.
A quick sum on whether it is worth it
Suppose a provider spends about six hours a month categorising bank lines and chasing which claims were paid. With de-identified exports, a saved prompt and bank rules built from the first two months, that falls to perhaps two hours including review. Four hours saved a month at a bookkeeper's hourly rate is the value; set it against the AI subscription, the time spent building rules, and the cost of any extra accountant review. For a very small provider the saving may be modest, and the main benefit is catching unpaid claims sooner. Time yourself for two months before deciding. If you want to see where AI would help beyond the books, the NDIS automation assessment checklist ranks admin jobs for you.
Questions providers ask about AI in their NDIS books
Can I paste a bank statement into ChatGPT?
Not as it comes. Bank lines from participants and plan managers can carry names and reference numbers. Strip or replace those first, work from an export with only date, amount and a claim batch reference, and use an account on a business plan with training switched off. Check your accountant and privacy obligations before using any AI tool.
Should I create an income account for every participant?
Usually not. Track income by funding type and support category in the books, and keep participant-level detail in your claims or care system. Participant-level accounts multiply your chart of accounts, expose names in reports and add work. Ask your accountant what level of detail your reporting needs.
What should AI never do in my accounts?
It should never post entries without review, decide how revenue is recognised, file anything, or change a claim. AI suggests categories and match candidates; you or your accountant approve them. Anything involving tax treatment, funding rules or a disputed claim belongs with a qualified adviser.
How do I know the AI matched a payment correctly?
Check the total. Matched claims must add up to the deposit to the cent, and each claim must appear once. A second check is to compare the count of claims paid against the count in your batch. Any difference is a rejected, held or missing claim that needs chasing.
Further reads
- NDIS Audit Checklist: Build the Evidence Folder With AI — Keep evidence ready for an audit.
- NDIS Service Agreement Template: Plain-English Drafting With AI — Plain-English service agreements that support clean invoicing.
- NDIS Provider Policies and Procedures: Tailoring Templates With AI — Tailor policies without pasting participant data.
- How to Write NDIS Progress Notes: An AI-Assisted Audit Checklist — Progress notes that back up every claim.
- How to Get Clients as an NDIS Provider: AI for Enquiries, Referrers — Grow the income you are about to reconcile.
- NDIS Occupational Therapy Report Template: A First Draft With AI — A report template for therapy providers.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: NDIS Guide to getting paid, How to make a payment request and How to troubleshoot claims and payments (ndis.gov.au, all read 4 November 2026), and the help pages of your own accounting software.