A customer is threatening to sue. A supplier dispute has been running for months. There is a warranty on work you completed last year, an environmental obligation attached to a site, a guarantee given for a subsidiary.
Each of these is either a provision, a contingent liability or nothing at all, and the difference is a recognised liability on the balance sheet, a note at the back, or silence. Companies get it wrong in both directions, and auditors spend a disproportionate amount of time here because the judgement is genuinely difficult and the incentive to reach a convenient answer is obvious.
Three tests, in order
A provision is recognised only when all three are met.
First, there is a present obligation arising from a past event. Something has already happened. The work was done, the contract was signed, the damage occurred. An intention to do something, however firm, is not an obligation.
Second, an outflow is probable. The usual reading is more likely than not.
Third, a reliable estimate can be made. This fails far less often than people claim.
Meet all three and you provide. Meet the first but not the second, and you have a contingent liability to disclose. If the possibility is genuinely remote, you do nothing.
The order matters. Teams tend to start with the amount, find it uncertain, and conclude there is nothing to recognise. That is working backwards. Establish whether an obligation exists first, because if it does, uncertainty about the amount is a measurement problem rather than a recognition one.
The obligation can be constructive
An obligation does not have to be legal. If your established practice has created a valid expectation in others that you will act, that expectation is itself an obligation.
The standard illustration is a retailer with a published refund policy exceeding its legal duty. Customers expect the refund, the retailer always gives it, and the obligation is real whatever the contract says.
This catches more businesses than expected. Routine goodwill repairs outside warranty. A redundancy practice more generous than statute. Any pattern consistent enough that stopping it would be a visible change of behaviour.
Measurement, briefly
The amount is the best estimate of what it would take to settle at the balance sheet date.
For a single obligation, such as one lawsuit, that is usually the most likely outcome, adjusted if the range of possibilities is skewed. For a large population of similar items, such as product warranties, it is the expected value across the population, which is why warranty provisions are built from historical claim rates rather than guessed.
Discount where the time value is material and settlement is distant, which matters for decommissioning and long-tail obligations. Do not deduct an expected insurance recovery from the provision. If recovery is virtually certain, recognise it separately as an asset, capped at the provision.
Restructuring, which has its own rules
Restructuring provisions are tightly controlled because they were historically abused, used to load costs into a bad year so later years looked better.
You need a detailed formal plan identifying the business, locations and approximate numbers affected, and you need to have raised a valid expectation in those affected by announcing it or starting it. A board resolution alone does not do it.
Only direct costs of restructuring go in. Not retraining, not relocating continuing staff, and never future operating losses.
What to write down, and when
The conclusion matters less than the record of how you reached it. For each item, note at the time: what the past event was, your assessment of likelihood with the reasoning, the estimate and how it was built, and what external evidence you have, usually correspondence from a lawyer.
Do it when the facts are current. These items reappear in the same form for years, and the file note written when the dispute arose is worth a great deal more than a reconstruction eighteen months later, both to your auditor and to whoever inherits your job.