Most of what Indian brands call attribution today is pre-post analysis, and pre-post analysis is not marketing-mix modelling — it is a before-and-after photograph with a caption written by whoever ran the campaign.
The context makes this distinction urgent. The deprecation of third-party cookies and the tightening of privacy law — the General Data Protection Regulation (GDPR) in Europe from 2018, and India’s Digital Personal Data Protection (DPDP) Act now shaping every data conversation with legal — did not kill marketing measurement. It killed one particular family of measurement: user-level, cookie-stitched, multi-touch attribution. What it revived is the older, sturdier discipline of Marketing Mix Modelling (MMM), which needs only aggregate data — weekly sales by geography, spend by channel, pricing, promotion, seasonality — and asks the only question a Chief Financial Officer respects: what would sales have been without this activity? For India, where the indirect distribution model means consumer-level transaction data mostly does not exist and aggregate secondary sales data from distributor systems does, MMM was never a fallback. It is the natively appropriate method.
But here is what happens in practice. A brand runs a campaign in March. Sales in April are higher than sales in February. A deck is produced attributing the delta to the campaign, perhaps with a seasonality adjustment if the analyst is careful. Leadership approves the next campaign on the strength of it. Across a decade of deployments, from consumer goods to retail to telecom, this pattern repeats with remarkable consistency — and it has a name. It is pre-post campaign analysis: establish a baseline in the period before, measure the level after, attribute the difference to the intervention. It is quick, it is legible, and it is the method most enterprise reporting stacks were actually configured to deliver, whatever the sales material called it.
The flaw is structural, not cosmetic. Pre-post analysis does not control for concurrent activity. In the same weeks your campaign ran, a trade promotion was live in two regions, a competitor cut price, the festive calendar shifted, a distributor expanded coverage, and the weather changed. Without a control group or a holdout — some population or geography that saw everything except your campaign — the pre-post delta is the sum of all of these forces, and the campaign gets credit for the lot. When the delta is positive, nobody audits the method. When it is negative, the same method quietly disappears from the review deck. That asymmetry alone should tell you it is not measurement; it is narrative.
A proper mix model is different in kind, not degree. It is a statistical decomposition — regression across time and geography — that separates baseline sales from the incremental contribution of each marketing variable, controlling for the concurrent activities that pre-post analysis ignores. The output is an elasticity: what an incremental rupee in a given channel returns, net of everything else that was happening. There is a well-worn maturity progression here that I have watched large advertisers climb. First comes the traditional mix study: static, backward-looking, a once-a-year report that arrives too late to change anything. Then the “due-to” stage: a decomposition of results by activity type that tells you, line by line, what each lever contributed. Finally the forward-looking stage: scenario planning and what-if optimisation, where the model stops being an autopsy and becomes a planning instrument that lets the organisation sense and respond as conditions shift. Most Indian brands believe they are at stage three. The evidence of their decks says most are at stage zero — pre-post — with stage-three vocabulary.
The dual lens matters here. The brand voice will argue, correctly, that a campaign is more than its measurable increment — that it builds memory structures and pricing power that a quarterly regression will underweight. That argument deserves respect, and a good Bayesian model can even encode it as a prior. What the brand voice may not do is borrow the language of causal measurement — “the campaign delivered X” — while the underlying arithmetic is a before-and-after subtraction. The evidence voice has one job: to say plainly what the numbers can and cannot prove. A pre-post delta proves that time passed and sales changed.
The excuse for avoiding real MMM used to be cost: proprietary models from global research houses, priced for global budgets, refreshed annually. That excuse expired. Meta published Robyn, an open-source mix-modelling framework, in 2021; Google published Meridian in 2024. Both are free, both are Bayesian in spirit, and Bayesian methods are particularly suited to India’s data reality — limited degrees of freedom, many brands and geographies, aggregate secondary sales as the outcome variable. A competent in-house data science pair, or a modest consulting engagement, can now stand up a defensible model in a quarter. The remaining constraint is not software. It is the discipline of assembling clean spend, pricing, and distribution data — and the political will to accept an answer that may embarrass last year’s decisions.
So, Monday. Ask your analytics lead one question about your current attribution: what, specifically, controls for concurrent activity — a holdout, a geo-split, or a model term? If the answer is a pause, you are running pre-post analysis, and you should rename it as such in every internal document, because words like “attribution” and “return on investment” are doing unearned work. Then commission two things in parallel: a geo-holdout on your next major campaign, because a controlled experiment is the cheapest honest measurement available, and an open-source mix model on two years of history. Where the model and the experiment agree, reallocate with confidence. Where they disagree with the pre-post decks of the past — and they will — treat that gap as the price of having flown blind, paid once, rather than a reason to shoot the model.