Your “Best” Campaign Might Be Your Worst: Incrementality, Explained

Here's a thought experiment that has ended a thousand marketing arguments. A shopper puts a product in their cart, gets distracted, and closes the tab. Tonight they'd have come back anyway — but this afternoon your retargeting ad catches them, they click it, and they buy. Your dashboard proudly credits the ad with the sale. Question: did that ad actually do anything?

The honest answer is no. The sale would have happened without it. You paid for a click that changed nothing — and your dashboard will never, ever tell you that. This is the problem of incrementality: not which ads get credit, but which ads cause sales that wouldn't have happened otherwise. It's the most important idea in performance marketing, and most marketers go years without confronting it.

Why dashboards can't answer the question

Attribution systems answer “what did the buyer touch?” They cannot answer “what would have happened in a world without the ad?” — because that world never ran. To measure causation you need a comparison: some people who saw the campaign, some who deliberately didn't, and the difference between them.

This is why retargeting looks unbeatable on every dashboard. It targets people already deep in your funnel — visitors, cart-abandoners, past buyers. Many were coming back regardless. The ads harvest credit for those returns, report a glorious ROAS, and hide the fact that a good chunk of that revenue was never theirs to claim.

Three ways to actually measure it

1. The hold-out test. Exclude a random 5–10% of your audience from a campaign and compare their behaviour to the exposed group. Uncomfortable — you're deliberately not advertising to people — and priceless, because the gap is the truth.
2. The geo test. Run the campaign in some cities and not in comparable ones, then compare revenue. No user-level tracking needed, which is why it survives every privacy change. This is the most practical option for mid-sized brands.
3. The pause test. The bluntest instrument: switch a channel off for two weeks and watch what actually happens to total sales. Terrifying to propose, wildly informative. Brands that pause brand-keyword search ads routinely discover most of that traffic simply arrives via the free organic listing below.

What this changes in practice

Once you think incrementally, budget conversations transform. Spend stops flowing to whatever claims the most conversions and starts flowing to what creates the most. Typically that means trimming bloated retargeting and brand-search budgets — the great credit-thieves — and reinvesting in prospecting, the unglamorous work of reaching people who've never heard of you, which dashboards chronically undervalue.

You don't need a data-science team to start. One honest hold-out, one geo split, one brave pause — any of them will teach you more about your marketing than a year of staring at dashboards. And being the person in the room who asks the incrementality question, calmly and with a test design ready, is one of the fastest ways to be taken seriously in this industry.

How to sell an incrementality test internally

The hardest part of incrementality is rarely the statistics — it is convincing someone to deliberately not advertise to a group of people. Three framings tend to work. Frame it as insurance: a small hold-out is cheap protection against scaling a channel that creates nothing. Bound the cost: a 5% hold-out on one channel for three weeks has a knowable, modest maximum downside, and you can say that number out loud in the meeting.

Pre-commit to the decision. Agree in advance what you will do at each result — “if lift is under X, we move 30% of this budget to prospecting” — so the test cannot be argued away afterwards. Tests that lack a pre-agreed decision rule tend to produce interesting charts and no change at all.

Putting a rupee value on the lift

The idea stays abstract until you attach money to it, so let’s run one. Say you spend ₹2,00,000 a month on retargeting and the dashboard credits it with ₹10,00,000 in revenue — a glorious 5x ROAS that nobody in the room would dream of touching. You quietly hold out a random 10% of the audience for four weeks. Same creative, same offer; the only difference is that one slice never sees the ads at all.

The numbers come back and they sting a little. The exposed group converts at 5.0%; the held-out group, seeing nothing, still converts at 4.25%. That gap — three-quarters of a percentage point — is all your money actually bought. It means roughly 15% of the credited ₹10,00,000 was genuinely incremental: about ₹1,50,000. Divide that by your ₹2,00,000 spend and your real, causal ROAS is 0.75, not 5. On the margin, you were paying ₹1 to make 75 paise.

Now for the part people get wrong: this does not mean you kill retargeting tomorrow. A 0.75 incremental ROAS on the whole budget is a signal that the last rupees are wasted, not the first. The move is to shrink the audience to the genuinely wobbly buyers — recent cart-abandoners, not everyone who blinked at your homepage — and re-run the test at the smaller size. Often the tighter campaign clears break-even comfortably while the bloated one never could.

One honest caution before you present any of this: a hold-out on a few hundred conversions is noisy, and a 0.75-point swing can be luck rather than truth. Run it long enough to gather a few thousand conversions per group, or the geo version if your volumes are thin, before you move real budget. The goal is a number you’d defend under questioning — not a scary chart you flash once and quietly forget.