GGA RESEARCH & METHOD
Debtor Segmentation Across Two Dealer Books
A four-class model, run on two live Indian dealer ledgers with 1,575 dealers and ₹83.42 crore between them. One book came back 75.4% in recovery, the other 30.3%. A model that flatters every book measures nothing; the value is in the distance between these two answers.
The argument
Sorting a debtor book by age tells you how old the money is. Sorting it by customer behaviour tells you what to do — and the two books here need almost opposite things.
In short
- Two earlier client-classification examples are retained. A seasonal agri input book of 765 dealers and ₹57.02 crore, and a non-seasonal sports goods book of 810 dealers and ₹26.40 crore.
- The answers are almost opposite. Class A — dealers paying substantially to terms — was 11.7% of value on the first book and 49.6% on the second. Class D, dormant with money outstanding, was 75.4% against 30.3%.
- Read the class shares within each example. Their difference describes the published classifications; it does not by itself validate the current calculator or predict recovery performance.
- Class concentration differs as sharply as class size. On the agri book, 90.5% of class D value sits beyond one year. On the sports goods book, class A is 99.9% inside six months — effectively no tail at all.
- Exposure per dealer differs by two orders of magnitude. The agri book carries ₹3.33 crore per class A dealer against ₹4.03 lakh on the sports goods book. The same class label describes a different commercial relationship.
- Neither book needed purchase history. Both were segmented from a debtor ageing alone, using recent open invoicing as a proxy for continuing activity — a proxy with a stated limitation, not a substitute for a last-order date.
The four classes
These are the historical labels used to discuss the retained examples. The current calculator instead applies D for legal status, then A, B or C from the share beyond the buying window. Historical recovery class D is not current legal class D.
| Class | Behaviour | What it needs |
|---|---|---|
| A — clean | Pays substantially to terms. Little or nothing past the warning band. | A routine internal cycle. Collection effort spent here is displaced from B. |
| B — warning | Past terms but not yet aged. Still trading. | Action now. Cheapest money on the ledger and the most time-sensitive. |
| C — deep overdue, active | Heavily overdue and still buying. | Priority. Supply leverage exists while they are still ordering, and will not later. |
| D — recovery | No recent billing, with real money outstanding. | Field verification and settlement. No supply leverage remains. |
| E — legal | In legal process and still carried in the debtors ledger. | Separation from the collectable book. A provisioning decision, not a collections one. |
One of the two clients classifies its own book on rules of this shape, published inside its own file: class A where balance beyond six months is under a threshold share, class C where balance beyond one year is 20–50% of the total. The model here is not imposed on the data; it formalises what experienced credit teams already do by eye.
The two books
The published examples map earlier client classifications. Their figures are retained under those historical labels; they are not a new run of the current calculator.
| Class | Agri inputs — seasonal, 765 dealers, ₹57.02 cr | Sports goods — non-seasonal, 810 dealers, ₹26.40 cr |
|---|---|---|
| A — clean | 11.7% | 49.6% |
| B — warning | 4.3% | 14.8% |
| C — deep overdue, active | 8.6% | 5.3% |
| D — recovery | 75.4% | 30.3% |
The agri book is genuinely distressed and the score says so. 72.2% of its value sits beyond one year, and 90.5% of class D value is beyond a year on its own. A D weighting of 75.4% is an accurate reading rather than a modelling artefact.
The sports goods book has a different problem. 38.6% beyond one year overall, but its class A is 99.9% inside six months — no tail at all. What it has instead is 14.8% of value in class B, money past terms on dealers still trading. That is the cheapest recovery on any ledger and the most time-sensitive: left alone it becomes C if they keep buying and D if they stop.
What sits inside each class
Class size alone can mislead. The ageing profile within a class tells you whether it is stable or moving.
| Class | Agri inputs | Sports goods |
|---|---|---|
| A — clean | 4.3% | 0.0% |
| B — warning | 14.6% | 36.4% |
| C — deep overdue, active | 33.1% | 94.0% |
| D — recovery | 90.5% | 93.3% |
Read the B row across. On the agri book, 14.6% of the warning class is already beyond a year — it is ageing but not yet lost. On the sports goods book that figure is 36.4%, meaning more than a third of its warning class has already crossed the point where the cheap recovery stops being cheap. Same class label, different urgency.
And read the C row. 94.0% of the sports goods C class is beyond a year, against 33.1% on the agri book. Both are "deeply overdue and still buying", but one is a recent deterioration and the other is a long-standing arrangement the dealer has settled into.
Exposure per dealer differs by two orders of magnitude
The agri book carries ₹3.33 crore per class A dealer — two dealers holding ₹6.66 crore between them. The sports goods book carries ₹4.03 lakh per class A dealer across 325 of them.
That changes what the class means operationally. A concentrated A class is a credit exposure whatever its payment record; a fragmented one is an administrative population. The same two letters describe a relationship-managed account in one book and a routine ledger entry in the other.
Why the agri book is aged in seasons, not days
The agri ledger does not age in days at all. Its buckets are named for crop seasons — Kharif-26, Rabi-25, Kharif-25, Rabi-24 — with day ranges attached underneath: 0–121, 122–180, 181–303, 304–364, and so on.
That is the correct clock for the business, and it is why a standard dormancy window misreads it. A three-month test applied to a dealer between crop cycles classes a perfectly normal customer as dormant. The seasonal windows used here — recent 0–6 months, warning 6–12, old beyond 12 — exist because the underlying commerce runs on two selling seasons a year, not twelve.
The declaration comes before the scoring, not after it. Non-seasonal businesses — footwear, FMCG, lubricants, electricals, building materials, sports goods — use recent 0–3 months, warning 3–6, old beyond 6. Seasonal ones — agri inputs, woollens, cooling, fans, festival goods — shift each band outward. It is a single judgement and it sets all three bands.
Reading the output
| Signal | What it usually means |
|---|---|
| B is a large share of value | Review terms, disputes and aged balances. The class alone does not identify the reason for delay. |
| C is larger than B | Review the account history. The class alone does not show how long follow-up has been absent. |
| Most of C still buying | Confirm recent orders before considering supply leverage. Ageing alone does not confirm activity. |
| D dominates by value | Historical class D calls for a documented account review. Choose recovery steps from current facts. |
| A is large but concentrated | Review concentration and payment history together. |
| E still inside the debtors file | Review legal status and the applicable accounting treatment. |
Two diagnostics worth running on the scored list
Ageing within class. A warning class with a third of its value already beyond a year, as on the sports goods book, is not an early-warning population any more.
Receipt re-application. Take the twenty largest persistent balances and match each receipt to the invoice its remittance advice referenced, rather than to the oldest open item. Where the profile shifts materially, first-in-first-out application has been concealing a dispute rather than resolving one.
Method and limitations
The earlier study reports an agri input book of 765 dealers and ₹57.02 crore across 20 states, and a sports goods book of 810 dealers and ₹26.40 crore. It describes mapping the clients’ existing classifications to the labels shown here, including sports goods B2 to C and C to D. These are retained published results, not newly recalculated outputs.
Recent open invoices are an activity proxy, not a purchase record. A dealer with no recent outstanding may have bought and already paid. Confirm activity from dated invoices or orders where available. The earlier study reports no last-order dates in the two example ledgers.
The earlier study reports no accounts under its historical legal label E. An unpopulated or absent legal field does not establish that no legal cases exist. The current calculator uses an affirmative legal flag for class D.
The band thresholds are operating rules, not externally validated cut-offs. The seasonal and non-seasonal windows are calibrated to billing rhythms observed across GGA trade channel engagements. Where a business bills on an unusual cycle, the bands should be set to that cycle.
Two figures differ slightly between sheets in one source file — a summary tab shows 25 and 13 dealers in two classes where the detail tab shows 27 and 11. Both reconcile to the same 810 total and the same value. The detail tab is used throughout.
This study makes no carrying-cost claim. Neither ledger carries billing or revenue by class, so the cost of financing each class cannot be computed from these books. An earlier version of this paper published such a table; it rested on a formula that applied days-sales-outstanding twice and has been withdrawn rather than restated.
No dealer, customer or individual is identified. GGA does not publish the payment behaviour of any named company.
Definitions
| Term | Definition |
|---|---|
| Dormancy window | The period of no billing after which a dealer is treated as inactive — three months for non-seasonal businesses, six for seasonal. Declared before scoring; it also fixes the warning and old bands. |
| Warning band | The ageing band immediately after normal terms have run out. Money past due but not yet aged, on a dealer still trading. |
| Supply leverage | The ability to influence payment because the customer still needs supply. Available only while they are buying, which is what separates class C from class D. |
| Activity proxy | Using the presence of recent open invoicing to infer continuing commercial activity, in the absence of a last-order date. A proxy with a stated failure mode, not a purchase record. |
| Concentration | Value per dealer within a class. Two books can share a class distribution and carry entirely different risk if one class is held by a handful of accounts. |
| Seasonal window | Ageing bands set to a business's selling seasons rather than to calendar months. On the agri ledger examined, the buckets are named for crop seasons. |
Frequently asked questions
How do you segment a dealer or distributor book?
By payment behaviour rather than by size or age alone, using four classes tested in order against each dealer's own balance. A pays substantially to terms; B is past terms but not yet aged and still trading; C is heavily overdue and still buying, so supply leverage exists; D has stopped buying with real money outstanding. A fifth class, E, separates accounts in legal process from the collectable book. Every threshold is a share of that dealer's own balance, never an absolute rupee figure.
Does debtor segmentation need purchase history?
Not necessarily, but the substitute has a limitation worth stating. Where no last-order date exists, recent open invoicing can be used as a proxy for continuing activity. It fails in one direction: a dealer who bought recently and has already paid shows no recent balance and will be scored worse than they deserve. Where a last-invoice or last-order date is available it should override the proxy.
How do you know a segmentation model is measuring anything?
Validate a model against known account facts and subsequent outcomes. The earlier examples report clean-class shares of 11.7% and 49.6%, and recovery-class shares of 75.4% and 30.3%, for 765 agri and 810 sports goods dealers. Those historical classifications alone do not validate the current calculator.
Why does seasonality change how a debtor book is segmented?
Because a gap in purchasing means different things in different businesses. One agri ledger examined does not age in days at all — its buckets are named for crop seasons, Kharif and Rabi, with day ranges attached underneath. A three-month dormancy test applied to a dealer between crop cycles classes a normal customer as dormant. Seasonal businesses use recent 0–6 months, warning 6–12, old beyond 12; non-seasonal ones use 0–3, 3–6 and beyond 6.
What does the ageing profile inside a class tell you?
Whether the class is stable or moving. On one book 14.6% of the warning class was already beyond a year; on the other it was 36.4%, meaning more than a third had crossed the point where cheap recovery stops being cheap. Same class label, different urgency. Similarly 94.0% of one book's deeply-overdue-but-active class sat beyond a year against 33.1% on the other — one a recent deterioration, the other a long-standing arrangement.
Can two books with similar class distributions carry different risk?
Yes, and concentration is why. One book examined carries ₹3.33 crore per clean-class dealer across two dealers; the other carries ₹4.03 lakh across 325. The same class label describes a relationship-managed credit exposure in one book and a routine administrative population in the other. A large clean class held by very few accounts is a credit exposure wearing a good payment record.
Should institutional customers be segmented with dealers?
No. Dealers and distributors buy repeatedly, which is what makes a gap in purchasing meaningful. Institutional buyers, government departments, armed forces and e-commerce marketplaces purchase on tender cycles, contract renewals or platform terms, so the activity test carries no information about them. One of the ledgers examined contains all of these types and they were excluded before scoring.
Evidence boundary
This publication distinguishes reported examples, modelled illustrations and operating observations. The examples do not validate a calculator or predict the outcome of an individual receivable, counterparty or company.
Public law, policy terms and procedures can change. Current primary material and the facts of the live case govern.