An enquiry arrives on an ordinary Tuesday, about a return filed last year, concerning a judgement somebody made three years before that. The law is rarely the difficulty. The difficulty is that the person who made the judgement has left, the client remembers the decision differently, and the file contains the figure without the reasoning behind it.

That gap is where enquiry cost lives. Not in argument about the treatment, but in reconstruction: the fortnight of digging, the awkward client calls, the best estimate assembled long after the facts have gone cold.

The good news is that the fix is cheap and the work is clerical. It simply has to happen years before anyone needs it.

• Most of the cost of an enquiry is evidential rather than legal. Reconstruction, not argument, consumes the hours.
• A file survives scrutiny when each material figure can be traced to a source document, a dated judgement note and a reconciliation.
• Contemporaneous beats comprehensive. A one-page note written in the year of the decision outperforms a thick folder assembled afterwards.
• The test worth applying: could a colleague who was never involved answer the question without calling the client?
• Build the habit into the annual cycle, on a small number of high-risk areas, rather than attempting a retrospective sweep of the whole book.
• Client-facing framing matters. Records are cheaper insurance than argument, and clients accept that far more readily than a lecture on compliance.

Establishing what the law says about a treatment is generally a contained piece of work. The provision exists, the guidance exists, and where there is genuine doubt the doubt is identifiable and can be stated.

Establishing what actually happened is the expensive part. How was the business-use proportion arrived at? What did the parties intend when the property was let? Which of the three versions of the spreadsheet produced the figure on the return?

Those questions have answers on the day the decision is made and vaguer answers every year afterwards.

Directors misremember. Staff leave. Systems get replaced and the old exports become unreadable. A file that depends on anybody's recollection has a single point of failure, and the failure is scheduled rather than possible.

⚠️ Watch: An explanation reconstructed after an enquiry opens is worth much less than the same explanation written at the time, and an experienced officer can tell the difference. The date on a note is part of its evidential weight.

Not the schedule that totals the figure, the artefact that generated it. A bank statement line, an invoice, a completion statement, a lease, a mileage log. For every material figure, one of these should be reachable in under a minute.

The path from the source documents to the number on the return. Where a figure is a total, the working that produces it. Where it is an apportionment, the calculation and the inputs.

This layer is usually present in some form, and usually the easiest to improve, since it is mostly a matter of saving the working rather than only the result.

The layer almost always missing, and the one that does the most work.

Wherever a treatment involved a decision rather than an arithmetic, a short note records what was decided, on what facts, on what authority, and when. One page. Often one paragraph.

"The count is easy. The judgement behind it is what an enquiry actually asks about."

Where a judgement rests on facts only the client can know, the client's confirmation of those facts, in writing, dated. A short email exchange is enough and is worth more than any amount of file memo written on the client's behalf.

Five things, in roughly this order: the question, the facts relied on, the conclusion, the authority, and the date. A worked template runs to about six lines.

✅ In practice: A usable judgement note reads like this. "Business use of the van set at 80%. Basis: mileage log Apr to Jul 2024 showing 1,940 business of 2,410 total miles, plus the client's confirmation by email 12.08.2024 that the pattern is representative. Client informed the log should be refreshed if the pattern changes. Recorded 14.08.2024."

At the moment of decision, in the working papers, before the file closes. Attempting to identify judgements retrospectively across a whole client base is the most common way this initiative dies.

Three recurring failures, all of them fixable:

1. Recording the conclusion without the basis. "80% business use" is a figure, not a judgement note.
2. Filing it somewhere the next person will not look. The note belongs with the year's working papers, not in a partner's email.
3. Writing it for the client rather than for a stranger. The audience is a colleague in five years who was never involved, or an officer who was not there.

Two clients. Same trade, same turnover of £180,000, same £14,000 of motor and travel costs, both entirely compliant. Two different practices hold the files, and an enquiry opens on the same point.

Neither client did anything wrong. One of them has an expensive fortnight.

The difference is a paragraph somebody wrote in August 2024, which took four minutes.

Nobody has capacity for a retrospective review of every file, and any plan that requires one will not survive January.

1. Pick three high-risk areas. Usually the ones where treatment rests on facts rather than arithmetic: business-use apportionments, property that changes character, connected-party transactions, anything with a relief attached.
2. Add one field to the working papers. A judgement-note box that is either completed or explicitly marked as not applicable. Visible absence beats silent absence.
3. Make it a review point. Whoever reviews the file checks that judgements carry notes, in the same pass as everything else.
4. Do not backfill. Start with the current year and let coverage build. A note written today about a decision made in 2022 is a reconstruction, and labelling it as anything else creates a new problem.
5. Tell clients what you are doing and why. Framed as cheaper insurance rather than as more paperwork, it lands well, particularly with clients who have been through an enquiry before.

The legal half of an enquiry response is where a research tool earns its place: finding the provision, the guidance and the decided case that settle the treatment, with citations that resolve to the source. It does nothing for layers one to four, which are the layers most enquiries actually turn on, and any tool sold as a substitute for the file is being sold for the wrong problem. Our limitations and responsible use page is direct about that boundary.

For the wider context on where compliance attention is currently concentrated, see our analysis of what the tax gap composition says about enquiry risk. To test the legal half against your own open questions, create a free account.

An enquiry tests two things: whether the treatment was right, and whether the practice can show how it was arrived at. The first is usually straightforward. The second decides whether the episode costs two hours or two weeks.

The file that answers first time carries one extra layer, written at the moment the decision was made, addressed to a stranger, and dated. Thickness has nothing to do with it.

Three high-risk areas, one box in the working papers, one review point, starting with the current year. That is the whole programme, and it pays for itself the first time a letter arrives.

For assessment criteria on the research tools that support the legal half of this work, see how to choose AI tax research software in the UK. Firms rolling a process like this across several offices can book a call.

Four layers: the source document behind each material figure, the reconciliation from those documents to the return, a dated note recording any judgement and its basis, and the client's written confirmation of facts only they can know. A file with all four can be defended by someone who was never involved in preparing it.

The question decided, the facts relied on, the conclusion reached, the authority for it, and the date. Six lines is usually sufficient. The date carries real evidential weight, which is why the note has to be written when the decision is made rather than assembled later.

No. A note written now about a decision made three years ago is a reconstruction, and presenting it as anything else creates a fresh problem. Start with the current year, build coverage forward, and handle historic positions on their own merits if they are ever queried.

Retention periods vary by taxpayer type and circumstance, and by whether an enquiry or a discovery position is in play, so the specific period should be checked against current HMRC guidance for the client in question. As a practical matter, judgement notes are small, cheap to store and worth keeping for as long as the underlying position could be reopened.

Those where treatment rests on facts rather than arithmetic: business-use apportionments, properties that change character or use, connected-party transactions, and anything carrying a relief with conditions. These generate the questions an enquiry officer cannot answer from the return alone.

It addresses the legal half, by locating the provision, guidance or decided case that settles a treatment, with citations that can be checked. It does nothing about the evidential half, which is where most enquiry time is spent. Treat it as a research accelerator rather than as a substitute for records.