SARS record keeping when an old assessment reopens
A tax assessment you thought was closed can be reopened after three years, and the High Court has just confirmed the conditions on which it happens. The cost of that is not the tax. It is whether you can reproduce a five-year-old transaction, with the documents behind it, in an afternoon or in three weeks of somebody's life.
Your record keeping is the variable there, not the rule. Two businesses with identical exposure will pay very different amounts to answer the same query, and the difference is entirely about retrieval.
What was actually decided, and what this piece will not tell you
The story is being carried as a SARS warning. It is not one. BusinessTech reports on a Western Cape High Court judgment, Commissioner for SARS v Meiring Citrus, handed down on 26 June 2026, in which the court allowed SARS to reopen a 2017 assessment that had prescribed.
The condition matters more than the headline. An assessment normally becomes final after three years, under section 99(1)(a) of the Tax Administration Act. SARS reopened this one under section 99(2)(a)(ii), which requires it to show that its failure to assess the full amount was due to fraud, misrepresentation or non-disclosure of material facts by the taxpayer. The court found material non-disclosure on the facts. It did not hold that SARS may reopen anything it likes at any time, and any summary saying so has dropped the load-bearing half of the sentence.
Beyond that, this piece asserts nothing. Not how the exception applies to your circumstances, not how long you are required to keep anything. Those are questions for the judgment itself and for your tax practitioner, and getting them slightly wrong in an article is how a reader ends up confidently mistaken.
What is worth writing about is the part practitioners do not cover, because it is not a tax question. When a query arrives about a year you had stopped thinking about, what does it actually cost you to answer it.
A reassessment is a retrieval problem
Picture the request landing. It concerns a specific year and usually a specific set of transactions. To answer it you need four things for each one. The invoice, the supporting documents that justify the deduction or the input claim, the proof of payment, and the entry in the ledger that ties them together.
In a business with good retrieval that is a search. Someone opens the accounting system, finds the transaction, and follows a reference to the document. An afternoon, maybe two.
In a business with poor retrieval it is an excavation. The invoice is in the accounting package. The supporting document is in an email folder belonging to someone who left in 2024. The proof of payment is in the bank's portal, which only goes back so far. The connection between them existed only in the head of the bookkeeper who captured it.
The tax outcome may be identical. The cost of arriving at it is not.
What the difference actually costs
Three weeks of a bookkeeper's time is the visible number, and it is the smallest one.
The larger costs are the ones nobody invoices for. A practitioner charging by the hour to reconstruct what you should have been able to hand over. Deductions abandoned because the supporting document could not be found, so the deduction gets conceded rather than defended. An owner and a financial manager pulled off the work of the business for a month. And the quiet compounding effect, which is that a business that answered badly last time answers badly next time.
Then there is the interest and penalty exposure on anything conceded for want of paperwork. A claim you were entitled to and cannot prove is, practically speaking, a claim you did not have.
Why the records exist but cannot be found
Almost no business loses its documents. What it loses is the link between them.
The invoice number in the accounting package does not appear on the file name of the scanned document. The purchase order that justified the spend lives in a different system with a different numbering scheme. Payment references were free text, so the same supplier is paid under four different descriptions. The email that carries the delivery confirmation is filed by sender, not by transaction.
None of that matters while the transaction is recent, because a person remembers. It matters entirely at the five-year mark, when the person has gone and the memory with them. Retention and retrieval are different problems, and the compliance conversation is almost always about the first one.
The direction of travel makes this sharper
This gets more pointed rather than less. SARS has put its VAT modernisation paper out for comment, with comments closing on 16 October 2026. The model it describes has the revenue service holding structured, invoice level data from both sides of a transaction.
Think about what that does to a query five years out. SARS will hold a clean, machine-readable record of what happened. The question becomes whether your version can be matched to theirs, transaction by transaction, rather than whether either of you can find the paperwork.
A business whose invoicing system and ledger already agree at that level of detail is in a completely different position from one that reconciles the two with a monthly journal. This is exactly the sort of joining-up work Zorah does inside the accounting and finance function. The reason to do it now is that it takes years rather than months, and it pays for itself long before any deadline.
What to do on Monday
Pick a transaction from three financial years ago. Something over R50 000, ideally with a supplier you no longer use. Start a stopwatch.
Find the invoice, the document that justifies it, the proof of payment and the ledger entry. Stop the clock when you have all four in one place and can show how they connect.
However long that took, multiply it by the number of transactions a query would cover. That is your real exposure, it is measurable this week, and it is the only part of this you control.