Provision matrix segmentation: how should receivables be grouped?
IFRS 9 allows a provision matrix as a practical expedient, but expects the loss rates to reflect groups of receivables that share credit risk characteristics. Useful dividing lines are the ones where loss experience genuinely differs:
- Customer type: large retailers, independent shops, government bodies, individuals.
- Industry or geography: customers in a sector or region that moves differently in a downturn.
- Size and credit rating: a few large, rated customers may be better assessed individually.
- Credit enhancements: receivables covered by credit insurance or letters of credit lose less.
- Product or channel: online sales to consumers behave differently from trade credit to businesses.
An example: why a blended matrix fails when the mix changes
A distributor sells to retail chains and to independent shops. Historically, chains made up most of the receivables, but the company has grown its independent shop business, which now holds 458,000 of the 1,180,000 total, up from a much smaller share.
| Allowance | |
|---|---|
| Retail chains, own loss rates | 4,900 |
| Independent shops, own loss rates | 24,810 |
| Segmented total | 29,710 |
| One blended matrix, rates reflecting the historical mix | 25,261 |
| Understatement from blending | 4,449 |
The blended rates were right for the old mix but are too low now, because more of the balance sits with riskier customers. The Segments sheet of the Loss rate builder (Excel) runs this comparison.
How many segments are enough?
Enough to separate the groups whose loss experience clearly differs, but not so many that each segment has too few write-offs to give reliable rates. A segment with one or two write-offs a year will produce volatile, unreliable loss rates. Most companies end up with two to five segments, sometimes with a few large customers assessed individually, and review whether the split still works each year.
When should a customer be assessed individually?
When it is large enough to matter on its own, has its own credit rating, or shows specific signs of trouble, such as missed payments, a restructuring or news of financial difficulty. A customer assessed individually is removed from the matrix, so its balance is not counted twice, and its specific provision is documented separately.
What data is needed to segment?
A customer master file that records each customer's type, industry and region consistently, so that ageing reports and write-off histories can be split the same way. Many companies find that their customer data is not clean enough at first, and segmentation starts with tidying it up.
How is credit insurance reflected?
Receivables covered by credit insurance that is part of the trade terms, or by letters of credit, have lower losses. They can form their own segment with lower loss rates, or the matrix can be applied to the uninsured part only. Insurance bought separately is usually a separate asset, not part of the matrix calculation itself.
How often should segments be reviewed?
At least annually, as part of preparing the year-end allowance, and whenever the business changes: a new customer type, a new market, or a change in credit policy. Testing whether loss rates differ significantly between candidate segments is a good discipline before adding or merging any segments.
See historical loss rates, forward-looking adjustments and the IFRS 9 provision matrix.
Need help applying the standards?
Our Chartered Accountants help finance teams and students apply IFRS and US GAAP to real transactions.
Questions people ask
Why segment a provision matrix?
Because customers with different credit risk have different loss rates, and a single blended matrix becomes wrong when the customer mix changes.
How do you choose segments for a provision matrix?
By shared credit risk characteristics where loss experience differs: customer type, industry, geography, size, credit insurance or product.
How many segments should a provision matrix have?
Enough to separate groups with clearly different losses, but each needs enough data for reliable rates; two to five is common.
When should a customer be taken out of the matrix?
When it is large, rated, or shows specific signs of financial difficulty; it is then assessed individually.
Sources
Every fee, date and rule on this page was taken from these official and primary sources.
Rules and fees change. If you are reading this long after October 4, 2026, confirm the figures with the source before you rely on them.
Related guides
More in ECL
This guide is general information. It is not tax or legal advice for your situation.