Automated transaction coding is the process of assigning account codes and GST treatments to bank transactions automatically — without a bookkeeper manually reviewing and entering each line. Instead of opening a transaction, selecting an account from a dropdown, choosing a GST code, and saving, the software makes those decisions itself. The bookkeeper's job shifts from data entry to reviewing a pre-coded ledger and approving or correcting it.
For Australian bookkeeping practices, this is the most time-intensive part of the reconciliation workflow. Every transaction in every client's bank feed needs an account code and a GST treatment before it can feed into the BAS worksheet. When a practice manages 15, 20, or 40 clients — each with hundreds of monthly transactions — the manual coding burden multiplies fast. Automated transaction coding is how practices break that relationship between client volume and time-per-period.
This page explains what automated transaction coding is, how Reconlink's three-layer approach works, who it suits, and what a practice can realistically expect in terms of time savings and accuracy.
What automated transaction coding does (and what it does not do)
Automated transaction coding software assigns account codes and GST codes to bank transactions on your behalf. It does this by applying a combination of business rules, machine learning, and language model inference to each raw bank description — and returning a coded, confidence-scored suggestion that either auto-commits to the ledger or enters your review queue.
What it does not do is replace the bookkeeper's judgement. The software handles the mechanical, predictable part of coding — the part where a bookkeeper looks at "WESTPAC INSURANCE PREM" and codes it to Insurance Expense at 10% GST for the fifteenth consecutive month. The software handles that. The bookkeeper focuses on the transactions that are genuinely ambiguous, new, or require a professional decision.
This is a critical distinction for practices responsible for their clients' ATO compliance. Automated coding is a tool to remove repetitive work, not a tool that removes accountability. The bookkeeper reviews, approves, and signs off. The software is fast; the bookkeeper is responsible.
For context on why getting this right matters before BAS lodgement, see how to automate bank reconciliation.
How automated transaction coding works: three layers
Reconlink processes every incoming transaction through three layers in sequence. Each layer is optimised for a different class of transaction. The result is high accuracy on the transactions the system is confident about, and an intelligent review queue for everything else.
Step 1 — Deterministic coding rules
The first layer applies rule-based matching. If the bank transaction description matches a saved rule pattern, the account code and GST treatment are assigned immediately. No machine learning, no inference — just a direct match.
Rules handle the high-frequency, predictable vendors that appear across your client base: utility providers, insurance companies, payroll processors, software subscriptions, ATO direct debits, loan repayments. These are the transactions that have always been obvious — they just required a bookkeeper to click through them one by one.
Because rules are deterministic, the error rate for in-scope transactions is effectively zero. A practice with a well-maintained rule library auto-codes 50–70% of its total transaction volume at this layer before the more computationally expensive layers run.
Critically, Reconlink maintains rules at the practice level, not the client level. A rule you build for one client applies across every client in your practice. This is a fundamental design difference from Xero's bank rules, which are scoped to the individual organisation. For a practice managing 30 clients, a practice-level rule library compounds in value with every new client added.
To learn how to build and maintain your practice rule library, see the guide to setting up coding rules in Reconlink.
Step 2 — Per-client machine learning model
Transactions that do not match any saved rule pass to a machine learning model trained on each individual client's coding history. This model learns the patterns specific to that client — how their particular set of vendors, transaction descriptions, and account structures behave over time.
If your bookkeeper has coded "ALDI SUPERMARKET BONDI JUNCTION" as Meals and Entertainment (FRE) for a specific hospitality client six quarters in a row, the model learns that pattern. It will apply the same coding automatically from the seventh quarter onward, without requiring a rule to be written.
Every ML suggestion carries a confidence score between 0 and 100. You set the auto-commit threshold at the practice level. Transactions scored above the threshold are written to the ledger without queuing. Transactions scored below the threshold are held in the review queue — pre-coded with the best available suggestion and the confidence score shown, so the bookkeeper is reviewing a suggestion in context, not a blank input.
This threshold is adjustable. New practices typically set it conservatively (higher threshold, more in the review queue) until they are confident in the model's accuracy for their clients. Mature configurations run at higher auto-commit rates as the models accumulate training data.
Step 3 — LLM fallback for new and ambiguous vendors
For transactions that do not match any rule and that the ML model cannot classify with sufficient confidence — typically new vendors the client has never transacted with before — Reconlink routes the transaction to a large language model (LLM).
LLMs interpret ambiguous bank descriptions using broad knowledge of business operations and Australian commerce. "STRIPE TECHNOLOGY AUST PTY LTD" is a payment processor fee. "MEDIBANK PRIVATE DD" is a health insurance direct debit. "CANVA PTY LTD" is a software subscription. An LLM can make these interpretations on first encounter, without prior transaction history.
Because LLM inference has a higher per-transaction cost than rules or ML, the architecture minimises how many transactions reach this layer. In a well-configured practice, fewer than 15% of transactions need the LLM. That proportion falls further as the rule library and ML models mature across periods. The system is designed to get more efficient over time, not to stay constant.
Confidence scores and the review queue
The confidence score system is what makes automated transaction coding usable in a compliance-sensitive environment. The bookkeeper does not blindly trust the software's output — they review the low-confidence transactions with full context: the raw bank description, the suggested account code, the GST code, the confidence percentage, and any comparable transactions from the client's history.
This is a qualitatively different task from manual coding. Instead of reading a blank transaction and deciding from scratch, the bookkeeper is evaluating a suggestion. Research consistently shows that review tasks are faster and more accurate than entry tasks. The cognitive load is lower, and the speed is higher.
The review queue shows only what needs attention. High-confidence transactions are already coded. The bookkeeper works through the uncertain subset, which in a mature configuration may represent 10–15% of total transaction volume.
Cross-client rule library: why it matters for practices
The single largest efficiency driver in Reconlink — beyond the AI coding itself — is the practice-level rule library.
Every bookkeeping practice accumulates knowledge about how to code common transaction types. The major supermarket chains, the big four banks' fee descriptions, the common insurance providers, the standard payroll processor references. This knowledge currently lives in the heads of experienced bookkeepers, in client-specific bank rules inside Xero or MYOB, or in documented procedures that have to be applied manually.
Reconlink externalises that knowledge into a shared, searchable rule library at the practice level. Build a rule once, and it applies to every current and future client. When you onboard a new client, the practice's full rule library is immediately active for their transactions. The new client's setup is not starting from zero — it is starting from the collective intelligence of your entire practice history.
For practices growing their client base, this is compounding leverage. The hundredth client benefits from rules built for the first ten.
Who automated transaction coding is built for
Reconlink's automated coding delivers the most value for:
- Practices managing 5 or more BAS-active clients where manual coding is the primary time cost across the quarterly cycle
- Registered BAS agents who need an auditable, ATO-compliant coding record behind every return they lodge
- Bookkeeping practices scaling headcount that need to increase client capacity without hiring proportionally
- CA firms with bookkeeping divisions that want practice-level oversight of coding progress across all clients before each BAS period closes
- Practices currently using Xero or MYOB who find that per-organisation bank rules do not scale across a multi-client practice
If you are a sole trader or business owner managing your own books, Reconlink is likely more than you need. A general-purpose accounting platform with basic bank rules will serve a single-entity workflow adequately. Reconlink is designed for the practice managing many clients — not the client managing themselves.
Feature comparison: manual coding vs. Reconlink automated coding
| Manual transaction coding | Reconlink automated coding | |
|---|---|---|
| Coding speed | 2–4 minutes per transaction (open, code, GST, save) | Rules and ML auto-code in under 1 second per transaction |
| Review volume | Every transaction | Only below-threshold transactions (typically 10–20% of volume) |
| GST code consistency | Dependent on individual bookkeeper attention per line | ATO GST codes (GST, FRE, INP, N-T, CAP) applied per saved rule and ML model |
| Audit trail | Manual documentation or none | Full transaction-level log: coding method, confidence score, coding history |
| BAS prep time | 2–6 hours per client per quarter | 30–60 minutes per client per quarter (internal benchmark, Jan–Apr 2026) |
| Multi-client management | Context-switching between separate client files | Single dashboard showing coding completion status across all clients |
| Rule reuse across clients | Rules re-entered manually per client | Practice-level rule library active for every current and future client |
| New vendor handling | Bookkeeper looks up and decides each time | LLM interprets on first encounter; rule or ML takes over from second occurrence |
| Error correction | Backtrack through transaction history manually | Correction updates ML model; propagated coding suggestions adjust accordingly |
| Confidence visibility | No signal — every decision looks equally certain | Confidence score shown per transaction; review queue sorted by uncertainty |
Pricing
Reconlink is priced per practice, based on the number of clients under management and the monthly transaction volume.
| Plan | Price | Clients | Transactions/month |
|---|---|---|---|
| Starter | $89/mo | Up to 10 clients | Up to 5,000 transactions |
| Growth | $229/mo | Up to 30 clients | Up to 25,000 transactions |
| Scale | $549/mo | Up to 100 clients | Up to 100,000 transactions |
All plans include the full three-layer coding engine, practice-level rule library, confidence scoring, review queue, and CDR bank feed integration. See the full pricing page for add-ons and annual billing options.
Frequently asked questions
What does "automated transaction coding" mean in plain terms? It means software reads each raw bank transaction description and assigns the correct account code and GST treatment automatically — without you doing it manually. Instead of clicking through hundreds of transactions per client per month, you review a pre-coded ledger and approve or correct the software's decisions. The software handles the repetitive, high-confidence codings; you handle the ambiguous or professionally sensitive ones.
How accurate is automated transaction coding? In practices with a mature Reconlink configuration — a well-built rule library and at least two to three periods of ML training data — auto-coding accuracy runs at 88–94% on committed transactions. The confidence scoring system is designed to keep the auto-commit rate inside your accuracy tolerance: if you set a high threshold, fewer transactions auto-commit, but those that do are coded with very high reliability. You control the trade-off between automation rate and review burden.
What happens when a coding is wrong? You correct it in the review queue or ledger view. The correction is logged in the audit trail and feeds back into the ML model as training data, so the same error is less likely to recur for that client. If the error was a systematic one — a rule that was too broad, for example — you can edit the rule directly in the practice rule library, and the correction applies across all affected clients from the next run.
Does Reconlink work if my clients are on MYOB or QuickBooks, not Xero? Reconlink is a standalone transaction coding and reconciliation layer that connects to bank feeds directly via CDR, not through your clients' accounting platform. The coded output can be exported or synced to the accounting platform the client uses for their general ledger. You are not required to migrate clients away from their existing accounting software to use Reconlink for the bank reconciliation and coding workflow.
Does automated transaction coding replace the bookkeeper? No. Automated coding removes the data-entry portion of transaction coding — the repetitive, mechanical part. The bookkeeper's professional role — reviewing accuracy, applying judgement on ambiguous transactions, understanding client-specific context, and signing off on BAS-ready records — remains entirely with the bookkeeper. Reconlink is a tool that makes bookkeepers faster and more scalable, not a substitute for their expertise or their compliance responsibilities.
Is the coded output ATO-compliant? Reconlink applies the ATO's GST treatment codes — GST (taxable supply), FRE (GST-free), INP (input-taxed), N-T (not reportable), and CAP (capital acquisition) — per the coding rules and ML models you configure. The accuracy of the GST classification depends on the rules being set up correctly. The bookkeeper reviewing the output remains the responsible party for ensuring each transaction is coded correctly before BAS lodgement. Reconlink provides the tools and the audit trail; the registered BAS agent provides the sign-off.
How quickly can I get a practice set up on Reconlink? Most practices complete initial onboarding in a single session: connecting bank feeds via CDR, importing client transaction history, and configuring the initial rule library. The ML models begin training from the first period's reviewed codings and improve across each subsequent reconciliation cycle. Practices with an existing set of well-documented coding procedures tend to onboard faster, since those procedures translate directly into rules.
Ready to stop coding transactions manually?
Reconlink is the automated transaction coding platform built specifically for Australian bookkeeping practices. Connect your clients' CDR bank feeds, let the three-layer AI handle the bulk of coding decisions, and spend your time reviewing rather than entering.
Book a free demo — see the automated coding engine working against your actual client data, with no commitment required.
This page was last reviewed on 23 May 2026. ATO GST treatment codes and CDR bank coverage evolve — confirm current obligations at ato.gov.au. This is general guidance, not specific tax or legal advice.