Evidence is the agent's job; the estimate is a person's. It ages the receivables ledger, calculates historical loss rates by ageing band, applies the matrix, lists customers that need a specific provision with the reasons, and drafts the workpaper and journal. The finance manager sets the forward-looking adjustment, decides each specific provision and signs the workpaper before anything is posted.
What is the bad debt provision, and who owns it?
Trade receivables are carried at what you expect to collect. Most accounting frameworks require a forward-looking estimate of credit losses, and many companies meet it for trade debtors with a provision matrix: a loss rate for each ageing band, built from history and adjusted for current and expected conditions. On top of that sit specific provisions for customers known to be in trouble: in insolvency, in dispute or simply silent.
The credit controller knows the customers. The financial controller or finance manager owns the estimate and its method. Auditors test it every year, because it is a judgement that moves profit.
The same discipline, at far greater scale and under banking supervision, applies to lenders; see preparing loan loss provision workpapers.
What are the steps, and at what depth?
- Agent, depth 1: extracts and ages the ledger. Open items by customer and invoice date, reconciled to the receivables control account in the general ledger.
- Agent, depth 1: calculates historical loss rates. For each ageing band, the share of balances that were eventually written off, over the look-back period the method sets.
- Agent, depth 2: applies the matrix. Loss rates by band, giving a collective provision, with the calculation laid out.
- Agent, depth 2: proposes specific provisions. Customers with insolvency notices, disputes, broken payment plans or no contact, each with evidence: collector notes, correspondence, public notices.
- Person, finance manager: sets the forward-looking adjustment. Whether current conditions, such as a sector downturn among your customers, justify a higher rate than history. This is the core judgement.
- Person, finance manager with credit control: decides specific provisions. Customer by customer, with the amount and reason.
- Agent, depth 2: drafts the workpaper and journal. The movement from last period, split into new provisions, releases and utilisation against write-offs. The journal is parked.
- Person, financial controller: reviews, signs and approves the journal.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Age ledger | Extracts and reconciles | 1 | Nothing |
| Loss rates | Calculates from history | 1 | Approving the method |
| Collective provision | Applies the matrix | 2 | Nothing new |
| Specific candidates | Lists with evidence | 2 | Deciding each one |
| Forward-looking view | Nothing | none | Finance manager |
| Workpaper and journal | Drafts and parks | 2 | Controller signs and posts |
What does a worked loss rate look like?
Take an illustrative distributor whose receivables in the 61 to 90 day band stood at around 500,000 at each quarter end over the look-back period. Tracking those balances forward, about 20,000 of each quarter's band was never collected and was written off. The historical loss rate for the band is therefore 4 in every 100. The agent applies that rate to this period's 61 to 90 day balance, shows the arithmetic, and repeats it for every band.
The workpaper should let a reviewer follow one band from raw ledger data to provision in a few lines. If the reviewer cannot, the method is too opaque for an auditor as well.
Which controls matter, and what evidence does each leave?
Ledger ties to the control account. The aged listing total equals the general ledger balance. Evidence: a tie-out at the top of the workpaper.
Method version. Bands, look-back period and the rule for what counts as a loss are written down and versioned. Evidence: method reference on every calculation.
Specific provisions decided by name. Each customer decision shows who made it and why. Evidence: decision log with evidence links.
No double counting. Balances with a specific provision are removed from the collective calculation. Evidence: an exclusion list.
Movement analysis. Opening, new, released, used, closing. Evidence: the roll-forward, reconciled to the ledger.
Write-offs are separate. Provisioning a debt and writing it off are different decisions. The agent proposes neither as a write-off here; whether an agent can write off a bad debt is a separate capability with its own approval.
What should you measure?
- Provision against actual losses, looking back a year, by band.
- Specific provisions proposed by the agent and rejected, with the reason.
- Customers moved into specific provision late, after a loss was already likely.
- Audit adjustments on the provision.
Where does it usually go wrong?
Unapplied cash. Customer payments sitting unallocated make balances look older than they are. The agent should list unapplied cash by customer and show the provision with and without it.
Credit notes not yet raised. A disputed invoice that will be credited is not a credit loss. Treat it as a revenue adjustment, not a provision.
History that no longer fits. A loss rate built from years when the customer base was different. The forward-looking adjustment is where a person corrects this, and the reason must be written down.
The agent's evidence is external. An insolvency notice found on a web page or forwarded in an email is external until confirmed from an official source. Fine for proposing; a person confirms before the decision.
Tax treatment ignored. Many tax regimes treat general and specific provisions differently. Pass the split to whoever prepares corporate tax provision workpapers.
Where Pryme fits
In Pryme, calculating the matrix and proposing specific provisions are separate capabilities, certified through Attest on your own past provisioning periods. Ledger data arrives sealed and collector notes declared, so the workpaper shows which proposals rest on what. The journal is held for the controller and every decision is filed. More on finance bounds is in AI agents for finance.
Questions people ask
How often should the provision be prepared?
At least at each reporting date. Many teams run it monthly because the agent makes it cheap, and the trend is useful for credit control.
Can the agent decide the look-back period?
No. It can show results under different periods so the finance manager can choose, but the method is a decision a person signs.
What about receivables from related companies?
Exclude intercompany balances from the trade provision. Their recoverability is a separate assessment.