Blended CAC Is Lying to You: The Case for Channel-Level Attribution

Most founders we meet can quote one number for their cost of acquisition. One figure, blended across every channel, trending in roughly the right direction. It feels like control. It is the opposite. A single blended CAC is an average, and an average is exactly where a failing channel goes to hide. As long as the blend looks healthy, nobody asks the harder question. Which rupee of spend actually brought the customer, and which one just rode along for free.

This is the most common way we see good Indian brands waste real money. Not on an obviously broken channel, but on a broken channel propped up by a great one, where the blend stays calm and the budget keeps flowing to the part that does not work.

What blended CAC actually averages away

Blended CAC is total acquisition spend divided by total new customers. That is its whole definition, and its whole problem. It does not ask where a customer came from. It treats the brand-search click that was always going to convert and the cold prospecting impression that built genuine new demand as the same unit of work, judged by the same average.

So picture two channels. One is brand search and retargeting, cheap, converting buyers who already wanted you, posting a tiny CAC. The other is cold prospecting, expensive, missing badly, posting a CAC that would horrify you alone. Blend them and the cheap channel’s efficiency drags the ugly one’s waste into a number that looks fine. You are not running two channels. You are running one channel funding another’s failure, and the blend is the trick that stops you noticing.

Blended CAC tells you what you paid on average. It never tells you what the next rupee will buy, and the next rupee is the only one you still control.

The honest number is marginal, and it is per channel

The number that matters is not the average. It is the marginal. What does one more rupee into this specific channel, today, at this spend level, actually return. Average CAC looks backward at money already spent. Marginal CAC looks at the decision in front of you, which is the only money you can still move.

This distinction is where most media-mix thinking quietly falls apart. Channels do not hold a flat efficiency as you scale them. The first rupees into a channel are usually cheap, harvesting the warmest, most ready demand. Push harder and you reach colder audiences, and the cost to convert each new buyer climbs. A channel can post a beautiful average CAC and a brutal marginal one at the same time, which means you are already past the point where adding budget makes sense, even though the average still flatters you.

The questions a blended number cannot answer

• If I add one lakh to this channel, how many incremental customers do I get? Not total, incremental. The ones who would not have bought anyway.
• Which channel is at its efficient ceiling and which still has cheap room to grow? Two channels at the same average CAC can be in completely different places on this curve.
• How much of my best channel’s volume is demand other channels created? Brand search converts cheaply because something upstream made people search.
• If I cut my worst channel tomorrow, what actually happens to total sales? Sometimes very little. Sometimes the whole funnel sags. The blend hides which.

Why the comfortable average survives so long

Be fair about why this persists. Blended CAC is easy. It needs no attribution model, no incrementality testing, no arguments about which touch deserves credit. It is one clean line for a board slide. And on a good month it tells a flattering story, so there is little pressure to look underneath.

Whoever owns the spend has every reason to keep the conversation on the blend. Channel-level truth implicates specific decisions and budgets. The blend implicates nobody. This is the same incentive trap we keep returning to. The metric that protects the people reporting it is rarely the metric that protects the business. Our Performance Marketing work starts by refusing the comfortable average, because you cannot fix a channel you have agreed not to look at.

The retention blind spot inside the blend

There is a second lie folded into the first. Blended CAC treats every acquired customer as one customer, full stop. But channels do not just differ on what they cost to acquire. They differ on what they bring. The cheap channel that posts a flattering CAC often buys discount-driven, one-and-done buyers who never return. The expensive channel you keep wanting to cut sometimes brings the patient, high-retention customers who actually build the business.

Judge those two on CAC alone and you will defund the channel that builds your future to protect the one that flatters your present. The fix is to stop reading acquisition in isolation. A channel is only as good as the cohort it brings, which is why we argue that retention cohorts are the one growth metric that survives a budget cut. Cut your CAC against the value a channel actually delivers over time, not against a headcount of first orders, and the rankings often flip.

This matters even more in India, where the cheapest acquisition is frequently the most disloyal. A channel that fills the funnel on deep discounts can post a wonderful blended contribution and a terrible repeat rate. The channels that earn durable relationships, including owned ones like WhatsApp used properly as a retention channel, change the real economics in ways a one-touch average never shows.

How to find the truth without rebuilding everything

You do not need a perfect attribution model. Perfect attribution is a multi-year project and a religious war. You need to break the average apart far enough to act, and a few disciplines get you most of the way.

1. Report CAC by channel, every month, never just the blend. One line per channel, side by side. The first time you do this, the subsidy usually becomes obvious in a single glance.
2. Watch the marginal, not the average. As you add budget to a channel, track what each new tranche of spend returns. When marginal CAC climbs sharply, that channel is near its ceiling regardless of how pretty its average looks.
3. Test incrementality by switching things off. The cleanest read you can get cheaply is a holdout. Turn a channel down or off in a region for a fixed window and watch total sales, not just that channel’s attributed sales. If nothing moves, you were paying for customers you already had.
4. Carry retention into the CAC view. Tag each channel’s cohorts and follow repeat behaviour, so an expensive channel that brings loyal buyers is not cut to protect a cheap one that brings churn.

None of this requires heroic tooling. It requires the willingness to look at the part of the picture the blend was hiding, and to let the spend follow that truth.

Where this bites hardest

The cost of the comfortable average is highest exactly when the stakes are. Early in a brand’s life, when every rupee is scarce, a blended number can mask the fact that your entire growth is one cheap channel harvesting demand while three expensive ones quietly fail. We have written about this in the context of the first ninety days of launching a D2C brand in India, because the habits you set in that window decide whether you spend the next year scaling a channel that works or defending an average that does not.

What changed recently

The case for channel-level truth got sharper in 2025, because the channels themselves stopped behaving like each other. Meta CPMs in India have risen sharply since 2023 while a new channel matured fast, and the gap between them is exactly the kind of thing a blend erases.

Quick commerce has become a genuine performance channel, and a different-shaped one. Reporting from Inc42 describes ads on Blinkit, Zepto and Instamart converting at roughly 3 to 8 percent against 1.5 to 3 percent on Meta and Google, with early campaigns posting about 1.5 to 2 times higher ROAS, measured as direct cart conversion at the point of intent rather than impressions. That same report is candid that the edge fades once the newness wears off and in-app keyword competition rises, which is the marginal-versus-average distinction playing out in real time. A channel can look spectacular on its first month’s average and ordinary on this month’s next rupee.

The money following this is not small. Per a Datum Intelligence forecast cited by Storyboard18, Blinkit, Zepto and Instamart together are projected to draw roughly Rs 4,900 crore in ad revenue in 2026. When a channel that books revenue at SKU and city level sits inside the same blend as a Meta line whose unit costs are climbing, the average is no longer averaging like-for-like. It is hiding which channel is buying incremental demand and which is harvesting demand the other one created. The brands that win the quick-commerce shift are the ones reading those channels apart, the same way you would read quick-commerce unit economics after platform fees rather than trusting a tidy contribution line.

The short version

Blended CAC is not wrong. It is incomplete in a way that happens to comfort the people reporting it. It tells you what you paid on average and stays silent on the only two things that change a decision. What the next rupee will buy, and which channel is paying for which. Break the average into channel-level, marginal truth, weigh it against the customers each channel actually keeps, and the loser stops hiding behind the winner.

If you can quote one CAC for your whole business and have never once seen it split by channel, that is the report to demand this week. Our Performance Marketing teams are measured on marginal efficiency and the retained value behind it, not on a blended figure that flatters a slide. The number your media mix is not showing you almost always explains more than the one it is.