The headline finding
On 6 April 2021, HMRC confirmed (Agent Update 83; reflected in its Insurance Policyholder Taxation Manual at IPTM3820) that two allowances inside the top-slicing relief computation — the Personal Savings Allowance and the savings starting-rate band — must be recalculatedat the notional income level used inside the relief, not carried through from the taxpayer's actual-income position. The five steps of the relief did not change; one input running through two of them did.
ParaplanAI's analysis of HMRC's own worked methodology puts a number on it. On a single, unremarkable offshore bond — a £60,000 gain built over six complete years, set against £35,000 of other income — the correct post-2021 method gives £10,046 of relief. The pre-2021 method, which carries the Step-3 Personal Savings Allowance of £500 into Step 4 rather than recomputing it to £1,000, gives £9,446. Same bond, same gain, same income: a £600difference that turns entirely on which allowance figure is used at one step. This is the engine's load-bearing regression case, pinned to zero-pence tolerance against HMRC's method on every change.
£600 on a small case. Larger gains and longer policy terms move more. And because the error produces a plausible, positive, right-order-of-magnitude number, it does not trip a sense-check — which is exactly why it persists.
Three error patterns we document
Across HMRC's published worked examples and the cases our engine is tested against, the same structural mistakes recur. None is exotic; each understates or overstates a client's tax in a way a spreadsheet will not flag.
1. The lagging top-slicing method.Carrying the Step-3 Personal Savings Allowance into Step 4 (the pre-2021 method) understates relief on every case where the full gain crosses a rate boundary the annual-equivalent slice does not. HMRC's own online calculator returned pre-change figures for a period after April 2021, and a long tail of downloaded adviser spreadsheets still does.
2. The missing CPI step in defined-benefit inputs. A defined-benefit pension input amount is the increase in the capitalised value of benefits — but the opening value must first be uprated by September CPI (Finance Act 2004, s.235) before it is subtracted. Skip that step and the input amount, the annual-allowance excess and the charge are all overstated — manufacturing a charge that is not due. In a high-inflation year (CPI was 10.1% for 2023/24) the gap is thousands of pounds on a single member.
3. The money-purchase-allowance folklore.“The MPAA wipes carry forward” is widely repeated and wrong (PTM055100 says the opposite). Carry-forward survives an MPAA trigger; it simply can never be set against the £10,000 money-purchase cap, while remaining available for the alternative annual allowance. A tool that lets carry-forward absorb a money-purchase excess understates the charge — the worst direction for compliance.
Each of these is unpacked, with the worked arithmetic and the governing HMRC reference, in our technical library.
The 2026 Paraplanner Tax-Calculation Survey
The analysis above is about the rules. The other half of this report is about the profession: how paraplanners and advisers actually run these calculations, what they use, and how confident they are in the figures they sign off. To measure that, we are running the Paraplanner Tax-Calculation Confidence Survey — a short, anonymous study of UK paraplanners and advisers.
It asks what tools the profession relies on for top-slicing, annual-allowance and defined-benefit calculations; how confident respondents are in those outputs; whether they were aware of the 2021 top-slicing change; and how a wrong figure would be caught. The findings will be published here, in aggregate and anonymised, as the quantitative half of this report — and shared with any trade title that wants to cover them.
Take part
The survey opens shortly. Paraplanners and advisers who want to be notified — and to see the results first — can register interest at info@paraplanai.co.uk. It takes about five minutes and is fully anonymous.
Methodology & data note
We hold a deliberately high bar on what we will and won't claim. The calculation findingsin this report are grounded in primary sources: HMRC manuals and legislation, and HMRC's own published worked examples, reproduced by ParaplanAI's deterministic calculation engine to the penny. Every figure — the £10,046, the £600, the CPI gap — traces to a specific HMRC reference or the engine's test corpus. None is an estimate.
The practitioner findings — how the profession works and how confident it is — come from the survey above, and are not yet published; this edition asserts no statistics about practitioner behaviour until that data is collected. ParaplanAI is a new tool, and we will not substitute invented numbers for a real sample. When the survey closes, the aggregate, anonymised results will be added here and dated.
For journalists & editors
This analysis is free to cite as “ParaplanAI's analysis” with a link to this page. The underlying worked examples, the engine's HMRC test corpus, and comment are available on request: info@paraplanai.co.uk. Run the headline case yourself on the free top-slicing relief calculator, or read how every figure is verified on the methodology page.
For planning and illustration purposes only. Verify all inputs against source documents. This report does not constitute financial or tax advice.