Take two statements about the same food-delivery company, both true in the same quarter: "revenue grew 62% and adjusted EBITDA improved ₹180 cr" and "the company loses ₹9 on every incremental order once you count the delivery subsidy and the voucher." A consolidated P&L cannot arbitrate between them, because a consolidated P&L is a sum: old cohorts and new, mature stores and launches, profitable cities and land-grabs, all mashed into one number. Unit economics is the derivative where the P&L is the integral: it asks what the next unit contributes, because the next unit is what growth is made of.
Why does unit economics matter?
You've met this logic before, in the master rule of capital allocation: growth destroys value whenever ROIC < WACC. Unit economics is the same rule at the atomic scale. A company is worth, roughly, unit value × the number of units it can eventually run. So if one unit, fully costed, is worth less than it costs to create, every rupee spent on growth is a rupee spent manufacturing losses. The faster that company grows, the more value it burns. That's how the same growth rate can thrill you about one company and terrify you about another.
Advertisement
It's also why "we'll make it up in volume" is either a profound truth or the oldest joke in business. It's true when fixed costs amortise over more units, which you'll see in operating leverage at the unit level. It's a joke when contribution is negative, because negative contribution scales negatively.
From this week on, your job in every analysis is to decide which it is, with arithmetic.
Three questions organise everything that follows:
- What is the unit, and what does one unit earn after all variable costs? Choosing the unit and costing its contribution answer that one.
- What does a unit cost to create, and how long until it repays that? CAC and churn come first here. LTV and payback follow.
- Is the machine improving or decaying as it scales, and does the story the company tells reconcile with the statements it files? Cohorts, the two end-to-end builds, leverage and the games take that one on.
Choosing the unit: the three tests, and one worked identification per archetype
Everything downstream depends on choosing the right unit, and you have to decide what it is. Nobody hands it to you. There are three tests, and all of them must pass:
- Test 1: it repeats. The unit recurs in comparable form, so you can compare instances across time and geography. "A customer" repeats. "The Jio launch" doesn't.
- Test 2: it carries a full revenue-and-cost loop. Both the revenue and the costs of serving attach naturally to it. An "app download" fails, because no revenue attaches to it. "A transacting customer" passes.
- Test 3: it matches the growth decision. Growth capital is spent in multiples of it. DMart grows by the store and Netflix by the subscriber. An insurer grows by the policy, an IT-services firm by the hirable hour. If management's next ₹500 cr of spend buys more of X, then X is the unit, because judging that spend is what unit economics is for.
Here's the identification run once for each archetype. Keep these six as your permanent pattern library:
| Archetype (M4.01) | The unit | Why it passes the tests | The unit's own KPI |
|---|---|---|---|
| Transactional retail (DMart, Costco) | The store | An order is too small (it carries no capex); the store is the repeatable object the next ₹45 cr buys, with revenue, staff, power, and inventory all attaching to it | Sales per store, store EBITDA %, payback years |
| Subscription (Netflix, SaaS) | The subscriber/account | Revenue (ARPU) and serving cost attach per subscriber; growth spend (CAC, content) is spent to add and keep subscribers | Contribution per subscriber, churn, CAC payback |
| E-commerce / delivery marketplace (Nykaa, Eternal) | The order, with the customer as the second layer | The order carries the full variable loop (product margin, shipping, payment, support); but CAC is spent per customer, so orders answer "does serving work?" and customers answer "does acquiring work?" | CM2 per order; orders/customer/year; CAC payback |
| Insurance | The policy | Premium, expected claims, commission, and admin all attach per policy written; the book is a stack of policy-vintages (persistency = retention) | Combined ratio per policy; VNB |
| Capacity services (hotels, airlines, IT services, hospitals) | The seat-hour / room-night / seat-km / billed hour / bed-day | Capacity is perishable: an unsold room-night expires at midnight; revenue and variable cost attach per occupied unit while fixed cost attaches per available unit, which is why utilization is everything (see break-even point) | Utilization/occupancy/load factor vs its breakeven |
| Lending | The loan (or ₹1 of AUM) | Interest spread, credit cost, and opex attach per rupee lent; growth is disbursement | NIM − credit cost − opex per rupee of assets (banks and NBFC playbooks) |
Common mistakes when choosing the unit
Two rules that save you from the classic mis-picks:
- The two-layer rule. Flow businesses need a flow unit for contribution (the order, the ride, the room-night) and a stock unit for acquisition and capital (the customer, the driver, the store). Collapse the layers and you get nonsense. "CAC per order" punishes loyal customers. "Contribution per customer" hides a broken order-level cost structure behind frequency.
- Treat management's unit as a claim. A quick-commerce company that presents only per-order economics is inviting you to overlook the dark store. That's the layer where the rent and the idle riders live, and it's the unit that decides profitability. Practice problem P4 makes you compute it. Choose your own unit first, then read theirs as a disclosure decision.