Five scores, five weights, one number
The structure of the calculation is deliberately simple. Each of the five pillars is converted to a sub-score on a shared scale, multiplied by its own weight and summed, and the result lands on the 0 to 1000 scale. All the difficulty lies in that first conversion: how the raw data of each pillar becomes a number comparable with the other four.
| Symbol | Pillar | How it scores |
|---|---|---|
| E1 | Repayment behaviour on the platform | Deductive — starts at full marks and falls with each negative event |
| E2 | Iranian Credit Scoring rating | Direct mapping from the official scale onto the model scale |
| E3 | Facilities and commitments at other banks | Stepped, on the ratio of debt burden to turnover |
| E4 | Account balance and turnover | Two sub-indicators combined at equal weight |
| E5 | Blocks and returned cheques | Deductive — it only subtracts, never adds |
Why we keep the equation simple
The standing temptation in scorecard design is to add interaction terms and non-linear functions for a few percentage points of statistical accuracy. We did not take that route, and the reason is operational rather than technical. A score a credit officer cannot explain to a customer in one meeting gets worked around in practice; and a score an auditor cannot trace gets challenged at the first inspection. The simplicity of the equation is the price we pay for defensibility, and we pay it knowingly. The complexity sits where it can be tested and reviewed: in the rules that convert raw data, each of which is calibrated and monitored independently.
Repayment behaviour: three things measured at once
This is the heaviest part of the model, because it is the only input that measures a company's actual conduct rather than its own assertions or someone else's assessment. Every company enters the observation window on full marks, and each negative event subtracts from that. But "how much it subtracts" has no fixed answer; it depends on three things.
One — severity of the event
Negative events fall into five classes, and the penalty for each class is meaningfully larger than the one before it. In order, from lightest to heaviest:
| Event class | Definition | Position on the severity scale |
|---|---|---|
| Short delay | Entry into arrears and return within a few days | The lightest recordable event |
| Moderate delay | Remaining in arrears for up to about a month | Several times a short delay |
| Long delay | Remaining in arrears up to the sixty-day boundary | Near the ceiling of the delay class |
| Entry into doubtful | Crossing sixty days and changing accounting treatment | A sharp jump — not a continuation of the class before |
| Partial settlement | Settling for less than the instrument amount, regardless of delay | An independent event, cumulative with any delay penalty |
Two — age of the event
A three-month-old delay and a twenty-month-old delay mean two different things. The first says the company is under liquidity pressure right now; the second only says it once was. The model holds events in a multi-year window and reduces their effect in steps: a current event carries full force, a mid-period event loses part of it, and an old event leaves only a shadow. Outside the observation window an event drops out of the calculation but is not deleted from the history — and the difference between those two matters in an audit.
The practical consequence for a customer is clear: a poor record can be repaired. A company that corrects its behavioural pattern recovers its score within a few quarters without anything being erased from its file. This is deliberate; a model with no route back removes the incentive to improve and, in practice, sends the customer to another platform.
Three — recurrence of the pattern
If several events of the same kind recur within a close window, the total penalty for that class is multiplied by a factor greater than one. The factor grows with each repetition and is capped, so that one short crisis does not disable a company's file permanently.
Further down this page you will find two sample cases in which this rule is decisive — and the difference it makes is larger than the difference between the two companies' official ratings.
Three external windows and one restraining filter
The first pillar says how a company has behaved with us. The next three say what is happening outside the platform, and the fifth looks only for restraining signals. All four use conversion rules that are declared in advance and identical for every company.
Pillar two — the Iranian Credit Scoring rating
A direct, unmodified mapping from the official scale onto the model scale. No judgement is added here; all the model does is put the rating on a comparable footing so it can sit beside the other four pillars. If the rating is unavailable — common for newly incorporated companies — a neutral value is substituted and the score is issued flagged as "incomplete data". We assume neither the best case nor the worst; we report an absence of data as an absence of data.
Pillar three — facilities and commitments at other banks
The question here is simple: how much of this company's repayment capacity is already promised to someone else? The basis is the ratio of total outstanding facilities and commitments to verified annual turnover, and the scoring is stepped rather than continuous. The reason is that in practice the difference between two companies in different steps is meaningful, while their difference within one step is measurement noise. An active overdue facility reduces this pillar independently of the debt burden ratio.
Pillar four — account balance and turnover
This is the only pillar that measures capacity rather than track record, and it is built from two sub-indicators at equal weight. The first is cash coverage: average balance against upcoming monthly obligations. The second is turnover stability: whether monthly turnover moves around a steady trend or is erratic and patternless. The second is often overlooked, yet two companies with equal annual turnover and different patterns represent two entirely different risks.
Pillar five — blocks and returned cheques
This pillar starts at full marks and only falls; no company earns extra points for "having no returned cheques", because that is the normal state rather than an achievement. Four signals are read: a remedied returned cheque, an unresolved returned cheque, an active judicial block, and non-performing debt in the banking network. The same ageing factor as pillar one applies here too, and two of these four signals also trigger a knock-out rule in addition to deducting points.
Four conditions the overall score does not get to overrule
In some situations a high overall score should not authorise credit. These rules run outside the equation and cap the grade, even when the number points to a higher one.
| Condition | Effect | Why it sits outside the equation |
|---|---|---|
| Unresolved returned cheque | Grade capped at C — issuance only with committee approval | A payment obligation has been dishonoured and not yet remedied. It should not be possible to mask that with good turnover or a high official rating; if it enters the equation, that is exactly what happens. |
| Active judicial block | New credit issuance halted | When a company's assets are not at its own disposal, measuring repayment capacity is meaningless. The question here is not the size of the risk but its feasibility. |
| Entry into doubtful in recent months | Grade capped at B2 | This is current conduct, not history. The ageing factor has not yet discounted it and it should not be neutralised by good marks on the external pillars. |
| Written-off receivable within the control window | Grade D | The most severe event possible in an instrument life cycle. The logic of gradual point deduction was not designed for this condition and applying it gives a misleading result. |
Company A — a good payer with two small slips
A company with two years of history and moderate activity on the platform. Its file is clean but not flawless — which is what makes it a good example, since most healthy companies look exactly like this.
Case profile
| Dimension | Status |
|---|---|
| Behaviour on the platform | |
| Activity volume | Several dozen instruments in the observation window — enough history for the full weight of pillar one |
| On-time settlement | The large majority of instruments settled with no event at all |
| Negative events | Two short delays and one partial settlement, since remedied |
| Distribution in time | Events spread across two years rather than clustered in one window |
| Entry into doubtful | None on record |
| External data | |
| Iranian Credit Scoring rating | In the upper half of the official scale |
| Debt burden | Light — a small share of annual turnover committed elsewhere |
| Cash coverage | Balance slightly above upcoming monthly obligations |
| Turnover stability | Very steady |
| Returned cheques and blocks | None |
What happens in the calculation
The repayment pillar closes with a small deduction from full marks, and three factors hold that deduction down. First, all three events fall into the light classes. Second, two of them are old enough that the ageing factor has reduced their effect to a fraction of the original. And third — the most important — the recurrence factor does not activate, because only one of the two same-class delays falls inside the close window. On the external side, all three pillars land in the upper steps and the fifth pillar is untouched.
The basis for the ceiling is the company's average verified monthly turnover on the chain, and the grade multiple is applied to that — not to a figure the company has declared. Grade A1 means approval without officer involvement and a semi-annual review.
Company B — same rules, an entirely different outcome
A company with lower activity and a recurring delay pattern. This case shows what actually drives a score down — and it is usually not what people guess.
Case profile
| Dimension | Status |
|---|---|
| Behaviour on the platform | |
| Activity volume | About half that of Company A |
| Negative events | Three delays of differing severity, all three within the close window |
| Distribution in time | Clustered — all in one compressed period |
| Entry into doubtful | One event, in the second half of the observation window, since settled |
| External data | |
| Iranian Credit Scoring rating | Around the middle of the official scale |
| Debt burden | Heavy — close to the whole of annual turnover committed |
| Cash coverage | Below upcoming monthly obligations |
| Turnover stability | Considerable volatility |
| Returned cheques | One event, remedied and old |
| Blocks | None |
What happens in the calculation
The repayment pillar loses close to half its marks and that, at a forty percent weight, decides the outcome of the file. But the bulk of that deduction does not come from the delays themselves; it comes from the recurrence factor, which fires because all three same-class events fall within the close window and enlarges the total penalty for that class appreciably. The entry into doubtful also has an effect, but being old, the ageing factor has discounted most of it. On the external side, the heavy debt burden and insufficient cash coverage both land in the lower steps and offer no compensation.
No knock-out rule fired: the entry into doubtful was not in the recent window and the returned cheque has been remedied. The grade therefore comes straight from the score. Review is monthly and every new instrument goes through the committee.
What pulled the score down, and what could not rescue it
Two sensitivity analyses on Company B's file, each changing one input and holding the rest constant. These two comparisons show the logic of the model better than any explanation.
If its official rating had been good
Replacing Company B's Iranian Credit Scoring rating with Company A's good one and holding everything else constant moves the file up by one grade and leaves it still in the lower half of the table. The ceiling on what this pillar can do is bounded by its twenty percent weight, and in practice only part of that shifts. A good external rating does not buy off a bad behavioural pattern.
If the same delay had come in one piece
Had Company B's total days of delay occurred as a single long event rather than three separate ones — that is, roughly the same number of days late — the file would have landed two grades higher. The whole of that difference comes from the recurrence factor, and it is larger than the effect of changing the official rating in the scenario beside it. For this model, three short slips are worse than one long one.
When history is thin, the model does not pretend to know
The first pillar only means something when enough settled instruments stand behind it. For a newly onboarded company, rather than producing an unsupported number, the model redistributes the weights and flags the score.
| Condition | Model behaviour | Flag on the score |
|---|---|---|
| Very little history on the platform | The weight of pillar one is reduced appreciably and redistributed across the external pillars | Insufficient history |
| Limited history | The weight of pillar one is reduced to a degree | Limited history |
| Sufficient history | Standard weights apply | — |
| No official credit rating available | A neutral value is substituted | Incomplete data |
The flag is not merely a label; it appears on the enquiry report and in the file as well, so the decision-maker knows how much evidence the number rests on. A score of 700 with full history and a score of 700 flagged "insufficient history" are two different things and should not be read alike.
Recalculation is event-driven
The score does not wait for a period to close. Instrument maturity, settlement (full or partial), a change in the accounting class of a receivable, a fresh bureau enquiry, the recording of a returned cheque or block, and the release of a new version of the weights — each triggers a full recalculation. The previous version is not deleted; it remains beside the new one with its time and cause, so that any historical score can be reproduced under the weights of its own day. That is the precondition for answering the question "why was this number issued on that date".