Division 296: AI for the New Super Tax
A New Tax That Is Really a Data Problem
For accounting and self-managed super fund administration firms, the most consequential change to the 2026 financial year is not a headline rate cut or a new deduction. It is Division 296, the additional tax on very large superannuation balances, which after a long and contested passage now takes effect from 1 July 2026. The politics around it were noisy. The operational reality for the firms that service affected clients is quieter, more technical, and arriving whether or not anyone is ready.
The measure itself is straightforward to describe. From 1 July 2026, individuals whose total superannuation balance exceeds $3 million face an additional tax on the earnings attributable to the portion of their balance above that threshold. The revised version of the measure, announced by the Treasurer in October 2025, made two changes that matter enormously for how firms handle it. It removed the taxation of unrealised gains, so the tax now applies to realised earnings, and it introduced indexation of the thresholds so they rise over time rather than dragging more people in through bracket creep. The first Division 296 assessment applies at 30 June 2027, based on the position across the 2026 to 2027 year.
The tax is levied on the individual, not the fund, and the Australian Taxation Office calculates each person's liability using information reported by their super funds. That single design fact is where the work lands on accounting and SMSF firms. The ATO can only calculate correctly from data that is complete, accurate and reported on time, and for clients with balances near or above $3 million, that data is rarely simple. It spans multiple funds, an SMSF with its own asset valuations, contributions and pensions, and timing that has to reconcile across the whole year. Division 296 is, underneath the legislation, a data reconciliation and reporting problem, and that is precisely the territory where disciplined automation earns its keep. This guide is for the practice principal, SMSF manager or client-facing accountant who now has to service this obligation across a book of clients without drowning in manual work.
Division 296 in plain terms
- Applies to individuals with a total superannuation balance over $3 million.
- An additional 15% on earnings attributable to the balance between $3 million and $10 million, on top of the usual 15% in accumulation, for a total of 30% on that portion.
- The revised measure applies to realised earnings from 1 July 2026, with unrealised gains removed, and the thresholds are indexed.
- Levied on the individual, calculated by the ATO from data reported by super funds.
- First assessment applies at 30 June 2027.
Why the Data Is Harder Than the Formula
The Division 296 calculation looks tidy on paper. The ATO works out the proportion of a person's earnings attributable to the balance above $3 million by taking the total superannuation balance, subtracting $3 million, and dividing the result by the total superannuation balance to get a percentage, then applying the additional tax to that share of the year's earnings. The revised measure also sets a higher tier, with an additional 25% applying to earnings attributable to the portion of a balance above $10 million, bringing the total on that slice to 40%. The arithmetic is not the hard part.
The hard part is producing a total superannuation balance you can defend. For a client with a single APRA-regulated fund and no complications, the number is reported cleanly and there is little for a firm to do. For the clients Division 296 actually targets, the picture is messier. They frequently hold more than one fund. They often run an SMSF holding assets that require careful valuation at year end, such as property, unlisted investments or private company shares. They make contributions and draw pensions on dates that affect the balance. Getting the total superannuation balance right, and getting the fund reporting to the ATO right, means pulling accurate data from several sources, valuing assets properly, and reconciling the whole thing to a single defensible figure at a single point in time.
Where the Effort Actually Sits
| Metric | Assumed Hard Part | Real Hard Part | Improvement |
|---|---|---|---|
| The tax itself | Complex formula | ATO does the sums | Low effort |
| Balance figure | Read off a statement | Reconcile multiple funds | Data work |
| SMSF assets | Prior year value | Defensible year-end valuation | Judgement |
| Timing | Point in time | Contributions and pensions across the year | Tracking |
| Client comms | One email | Proactive planning across a book | Capacity |
This reframing changes what a firm should build. If the effort is data assembly, valuation support and client communication rather than the tax computation itself, then the productivity lever is a reliable data pipeline, not a smarter calculator. That is a very different investment, and a much more valuable one, because the same pipeline that supports Division 296 also supports every other high-balance-client conversation a firm needs to have.
There is a second reason the data deserves more attention than the formula. Because the ATO calculates the liability from what the funds report, an error in the reported balance flows straight through to a client's assessment. If a super fund reports a balance that is too high, because an asset was overvalued or a pension drawdown was recorded late, the client can be assessed on earnings they did not really have. If it is too low, the shortfall surfaces later. Neither outcome is one a firm wants to explain after the fact. The defence against both is the same boring discipline: accurate source data, defensible valuations, and reconciliation done before the reporting deadline rather than after the assessment lands. For a client sitting just above $3 million, a modest valuation difference can be the difference between being in scope and out of it, which raises the stakes on getting the number right the first time.
Where AI Genuinely Helps, and Where It Must Not Decide
It is worth being precise about what AI should and should not touch here, because superannuation tax is an area where a confident wrong answer is a professional liability, not a minor error. The right mental model is that AI handles the assembly and the drafting, and a qualified human owns the judgement and the advice.
On the assembly side, the gains are real and low-risk. Pulling data from fund statements, SMSF accounting systems and platform reports into a single view is exactly the kind of repetitive extraction and reconciliation that AI-assisted tooling does well and humans do slowly. Flagging which clients across a book sit above or near the $3 million threshold, so the firm can prioritise conversations, is a screening task that a well-built system can run continuously rather than once a year in a panic. Drafting the plain-English explanation a client needs, so an accountant edits rather than writes from scratch, saves hours per client without ceding any judgement. For firms already modernising their engine room, this sits naturally alongside the Xero and MYOB automation trends we mapped for accounting firms in 2026.
On the judgement side, the line is firm. An asset valuation that underpins a total superannuation balance is a professional judgement with real consequences if it is wrong, and it is not something to delegate to a model. The tax advice itself, whether a client should change contribution patterns, restructure, or simply pay the tax, is regulated advice that a qualified professional must own. And every number that flows to the ATO carries the firm's name behind it. AI can prepare the ground for all of these, but it cannot be the one standing behind them.
AI or Human for Each Division 296 Task?
The firms that get the most from AI here are the ones that draw this line deliberately and put it in writing. When everyone on the team knows which steps are machine-assisted and which are human-owned, the tooling accelerates the practice without quietly eroding the professional standard that the practice sells. When the line is fuzzy, an unreviewed model output eventually reaches a client or the ATO, and the productivity gain turns into a professional-standards problem. Our note on AI agent governance and human override sets out how to make that override real rather than theoretical.
A Practical Pipeline for a Firm's Whole Book
The mistake most firms will make with Division 296 is to treat it client by client, reactively, as returns come in. The better approach is to build one repeatable pipeline that runs across the whole book, so the affected clients surface early and the manual effort per client falls sharply. The building blocks are not exotic.
A Division 296 Readiness Pipeline
The value of building it as a pipeline rather than a checklist is that it turns an annual scramble into a standing capability. Once the ingest and screening steps run continuously, the firm always knows which clients are in scope, which are trending toward the threshold, and where the data is incomplete, months before any deadline. That converts Division 296 from a compliance cost into a client-service advantage, because the firm can start the conversation with an affected client from a position of preparation rather than reaction. Multi-entity and multi-fund consolidation is a problem our own reporting work has focused on for years, and the same consolidation discipline that powers tools like ReportingMate is what makes a book-wide view practical rather than aspirational.
Where the Time Goes, Reactive vs Pipeline
None of this removes the professional work, and it should not. What it removes is the low-value manual assembly that currently sits between an accountant and the advice their client is actually paying for. The reconciliation still has to be right and the valuations still have to be defensible. The point of the pipeline is to give the human more time for exactly those judgements by taking the data drudgery off their desk.
What to Do Before 30 June 2027
The first Division 296 assessment applies at 30 June 2027, which sounds comfortably distant in mid-2026. It is not, because the position it assesses builds across the whole 2026 to 2027 year that has already started. Decisions clients make now, about contributions, pensions and asset holdings, affect the balance that will be measured. A firm that waits until returns season to think about Division 296 will be reconstructing a year that has already happened, which is exactly the wrong time to discover a data gap.
Getting Ready Across the Firm
A related point is worth making about client expectations. The clients Division 296 affects are, by definition, among a firm's most valuable, and they will not appreciate learning about a new tax on their superannuation from a return rather than from their adviser. Reaching out early, with a clear explanation of what the tax is, when it applies and what if anything they might consider, is both good service and good risk management. It positions the firm as on top of the change, and it heads off the difficult conversation that follows a client discovering an unexpected liability. A book-wide screening capability is what makes that outreach practical, because it tells the firm exactly who needs the conversation and roughly where they sit relative to the threshold, without a partner manually trawling the client list.
It is also worth resisting the urge to over-engineer the response. Division 296 does not require a firm to rebuild its entire technology stack. It requires clean data ingestion, reliable screening against a threshold, and a disciplined reconciliation at year end, wrapped around the professional judgement that was always going to be human. A firm that already has reasonable data hygiene may need only to formalise and automate what it does informally today. A firm running on spreadsheets and memory has a bigger job, but it is a job worth doing for reasons that extend well beyond this one tax.
The strategic read for a practice principal is that Division 296 is a capability test disguised as a tax change. The firms that come through it well will not be the ones that computed the tax most cleverly, because the ATO does the computation. They will be the ones whose data was in order, whose affected clients heard from them early, and whose staff spent their hours on advice rather than assembly. Building that capability is a modernisation project the practice needed anyway, and it is precisely the kind of work covered in our broader guide to AI automation for accounting firms and tax compliance. Division 296 is simply the deadline that makes the case for getting on with it.
Related Reading
- AI automation for accounting firms and tax compliance
- AI for accounting firms: Xero and MYOB automation in 2026
- The AI Xero reconciliation agent for bookkeepers
- AI cash flow forecasting: a guide for growing businesses
- AI agent governance, data access and human override
This article is general information, not tax, financial or legal advice. Division 296 outcomes depend on individual circumstances. Clients should seek advice from a qualified tax or financial adviser about their own position.