Build the intuition on the chain before trusting any single link. A D2C brand spends ₹10 lakh on ads: 50 lakh impressions, 1% CTR → 50,000 sessions, 2% conversion → 1,000 orders, of which 700 are new customers. CAC = ₹10,00,000 ÷ 700 ≈ ₹1,429. Average order ₹2,000 → revenue ₹20 lakh → ROAS 2.0x. Looks alive, until margin enters: at 35% contribution, those orders earn ₹7 lakh against ₹10 lakh of spend. This '2x ROAS' business loses ₹3 on every ad rupee until repeat purchases arrive, which is why CAC's only meaningful reading is against contribution margin and repeat behaviour (payback period and CLV), never against revenue. The chain forces the discipline: every ratio is a numerator and denominator you must be able to name, and most metric arguments are secretly denominator arguments.
CAC's honest versions come from splitting. Paid CAC (paid spend ÷ customers attributed to paid) versus blended CAC (all spend ÷ all new customers): the gap between them is your organic engine, and watching paid CAC rise while blended holds is the classic sign you're buying customers the brand was already earning. New-versus-returning splits stop the flattering error of counting a loyalist's fifth order as an acquisition win. And the honest ceiling on all of it is incrementality: the customers who would have arrived anyway (they searched your name, and the discount intercepted them at checkout) make any attributed CAC look better than the causal one. The gold-standard check is the holdout: pause or geo-split the spend and watch what actually changes. Companies that run this test regularly are routinely humbled by how much 'performance' was reallocation of the inevitable.
ROAS needs three corrections before it can be trusted with a budget. Margin: convert to profit-on-ad-spend or set ROAS targets per category margin, since a fixed 4x target across a catalogue where margins run 15% to 60% systematically overfunds the wrong products. Attribution: platform-reported ROAS is a claim by an interested party. Click-based models over-credit the last touch, and every platform counts the same conversion. MER (total revenue ÷ total marketing spend) is the blunt but unbluffable supervisor: when platform ROAS rises while MER falls, the platforms are taking credit, not creating customers. Diminishing returns: ROAS is an average, but budgets move at the margin. The first lakh targets your eager buyers, the tenth chases strangers, so the question for more budget is always marginal ROAS, which no dashboard shows by default and every scaling decision depends on.
Ratios also need volume and guardrails to mean anything. A CAC of ₹300 at a thousand customers a month and the same CAC at fifty thousand are different businesses: efficiency metrics divorced from scale reward shrinking to excellence ('our best-ever ROAS' on a tenth the spend is usually a smaller company, not a better one). The working setup pairs every optimisation metric with a guardrail: conversion rate with average margin per order (kill the discount-spike illusion), CAC with new-customer volume and payback months, ROAS with MER and contribution margin. And a share of judgment stays reserved for what these metrics structurally can't see: the brand effects that make next year's clicks cheaper, which short-term dashboards record as nothing happening.