Evolution of marketing mix modelling in Indian FMCG
The evolution of marketing mix modeling in the Indian FMCG industry reflects a shift from explaining historical sales with broad media inputs to estimating how marketing investment creates incremental demand across a complex marketplace. Indian consumer goods companies now operate across television, digital video, social media, retail media, e-commerce, quick commerce and owned channels, often with different regional footprints, promotional calendars and distribution strengths.
For Australian marketing teams, this shift offers useful lessons about combining disciplined econometrics with contemporary data. The two markets differ in retail structure, media consumption and consumer geography, so Indian findings cannot simply be transferred. What can be transferred is a more rigorous approach to incrementality, experimentation, price elasticity and the measurement of campaigns that receive little coverage.
| Modelling era |
Core analytical question |
Typical inputs |
Principal contribution |
Persistent limitation |
| Legacy econometrics |
Which variables are associated with sales? |
Monthly sales, advertising, price, promotion and distribution |
Established regression-based MMM |
Limited digital detail and weak causal interpretation |
| Integrated MMM |
How do multiple media channels work together? |
TV, digital, search, social, price, promotion and availability |
Introduced saturation, carry-over and cross-channel effects |
Dependence on consistent, granular data |
| Data-rich MMM |
Where do shoppers actually see and buy products? |
Retail panels, household data, e-commerce and first-party signals |
Connected exposure with purchase behaviour |
Visibility gaps across informal and offline channels |
| Causal MMM |
What sales would not have occurred without a campaign? |
Experiments, media data, outcomes and contextual variables |
Prioritises incremental effects over claimed attribution |
Experiments are costly and available only for selected campaigns |
| Adaptive MMM |
Which decisions should be made next? |
Near-real-time signals, calibrated causal models and scenario inputs |
Supports continuous budget and activation decisions |
Requires governance, speed and organisational trust |
From national regressions to market decisions
Early marketing mix models for Indian FMCG businesses were built around monthly or quarterly brand sales. Analysts regressed sales on advertising expenditure, price changes, promotions and distribution measures, often using television and print as the main media variables. Long-term advertising effects were represented through adstock functions, while logistic curves captured diminishing returns. These techniques remain relevant, but the assumptions behind early models were much narrower.
The first limitation was the dominance of national data. A national model could identify that a television campaign coincided with higher sales, yet it could not explain why two similar markets responded differently. Regional distribution gaps, retailer relationships, local competitors and language-specific advertising were treated as background factors rather than modelled drivers. Consequently, many recommendations concentrated on moving money between broad media categories rather than changing how campaigns were targeted or sequenced.
Integrated media becomes the centre of analysis
The rapid expansion of mobile advertising, vernacular content and platform-based buying fragmented the media landscape. Indian FMCG companies began connecting television and print exposure with YouTube, connected TV, Instagram, Facebook, search, e-commerce sponsorship and retailer-funded advertising. JioHotstar, for instance, is valuable for premium video alongside connected channels. Analytics evolved from measuring average media weight to estimating saturation, lagged response and interaction between channels.
Integrated media mix modelling also changes the unit of analysis. A product may be discovered through connected television, researched through social content and purchased through a quick-commerce platform. A single exposure variable can no longer represent that journey. Modern models introduce path-specific effects, such as search activity following video exposure, and calculate marginal returns after the strongest channels have already saturated. Offline and online signals need to be aligned to common weeks, markets and campaign periods to make this analysis credible.
Australia has followed a similar path, with Amazon Ads, Google, Meta, TikTok and retailer media forming part of many FMCG campaigns. Retail-media environments associated with Woolworths and Coles provide another source of purchase-linked signals. The operational lesson is common to both countries: media performance should be assessed across the full path to purchase, not through separate reports for each platform.
Better data changes what can be measured
The data foundation has improved alongside the media mix. Retail scanner panels and household panels now offer more frequent observations on volumes, prices, household penetration and repeat purchase. E-commerce platforms add browsing and basket information, while digital campaigns expose impressions, frequency and conversion events. These sources allow analysts to connect consumer behaviour with product availability instead of relying only on shipments recorded by a manufacturer.
The distinction between sell-in and sell-out remains critical. Shipment data can rise because retailers stocked more inventory, even when shoppers bought no additional units. In Indian FMCG, a model must therefore separate consumer demand from loading at the factory, orders placed by distributors and stock pushed into modern trade. Stock-outs, delayed distributor payments and differences between primary and secondary sales can otherwise be misread as genuine changes in demand.
Geography adds another layer. Traditional retail is distributed across kirana stores, supermarkets, organised retail chains and regional distributors, while e-commerce and quick commerce create additional availability patterns. Data may be strong in metropolitan areas but thin in smaller cities and rural districts. A model that treats all observed retail channels as equally representative risks learning the preferences of shoppers who are easiest to measure. Hierarchical models and carefully designed overlays can reduce this problem, although they cannot eliminate incomplete market coverage.
Causal evidence replaces assumed attribution
Correlation-based attribution assigns credit according to a model’s assumptions, but it cannot reveal what sales would have happened without an investment. Randomised experiments and geographic holdouts provide a stronger benchmark because they create a credible counterfactual. Indian FMCG teams have increasingly used matched-market tests, city-level pilots and conversion experiments to estimate the incremental lift of digital video, retailer promotions or new product launches.
Causal MMM combines those experimental results with broader observational data. A geo experiment might estimate the effect of a regional campaign, while a calibrated model applies that learning to markets or channels that could not be tested. Synthetic control, difference-in-differences, Bayesian causal models and machine-learning methods are all used in practice. Their value depends less on technical complexity than on the quality of the assumptions, control variables and validation design.
This is especially important when channels compete for the same attention. If search demand rises after a television burst, a naïve model may allocate all the incremental sales to television. If a promotion coincides with a competitor’s stock-out, part of the apparent lift may belong to a competitor. Experimentation, stock data and competitor monitoring help distinguish genuine incrementality from coincidence. Australian teams can apply the same discipline to campaigns around Woolworths, Coles, Aldi and marketplace environments, where promotions and retail visibility often move together.
Price, promotion and distribution join the conversation
Marketing mix modelling in FMCG is no longer treated as a media-only exercise. Price, promotional depth, distribution and availability can affect sales as strongly as advertising, and media effects may change when a product is easier to find or more affordable. Modern models estimate own-price and cross-price elasticity, the incremental effect of discounts and the contribution of weighted versus numerical distribution. They also account for campaigns in stock-out markets, where greater availability rather than greater demand may explain unusual sales.
The analysis must distinguish push from pull. A trade discount may increase retailer inventory without generating additional consumer purchases, while a consumer promotion in a poorly distributed region may produce strong response per exposed household but limited total volume. Brand-level effects also interact with category effects. A Nescafé campaign might shift shoppers from another instant coffee brand, while a packaged-food promotion could expand the category or merely move purchases between brands owned by the same company.
These considerations make MMM more useful for portfolio decisions. Instead of optimising each campaign in isolation, teams can compare the return on advertising with the effect of price investment, display space, retailer funding and distribution expansion. The outcome is a more balanced view of incremental profit, particularly in markets where temporary discounts and trade margins are substantial.
Localisation moves beyond national averages
India’s linguistic, cultural and retail diversity makes localisation a core modelling requirement. A single national coefficient can conceal major differences between Hindi, Tamil, Telugu, Bengali and other language markets, as well as between metro areas, tier-two cities and rural districts. Campaigns tied to Diwali, Holi, Eid, the monsoon, cricket or a major regional festival can create short periods of intense demand that require event variables and flexible time effects. Product availability must be modelled at a level that reflects how consumers shop locally.
Digital data creates its own bias. A national media campaign may overrepresent connected consumers in Mumbai, Delhi, Bengaluru and Hyderabad, while exposure among offline-heavy shoppers in eastern India or smaller towns remains difficult to observe. Modern MMM teams address this through state-level models, market clusters, channel-specific observation units and experiments designed in less digitally saturated locations. A stronger estimate can come from combining imperfect national signals with granular experiments rather than pretending national impressions represent the entire population.
Australian teams face a smaller-scale version of the same problem. Melbourne and Sydney are major media buying and retail centres, but demand also differs across regional Victoria, Queensland and Western Australia. Seasonal campaigns around Easter, Christmas and Boxing Day alter gifting demand, while the formats and value propositions of Aldi and Costco differ from those of conventional supermarkets. A model that captures Australian behaviour should represent those differences rather than rely on a single national response curve.
Lessons for Australian marketing teams
The most transferable element of Indian MMM development is the operating model, not any particular coefficient. Australian teams can combine weekly or monthly performance data with retailer sales, search behaviour, promotion calendars, availability and competitor activity. A category-level model can then inform brand allocation, while focused experiments establish the causal benchmark used to calibrate that model. This structure gives finance, marketing, sales and retail teams a common language for discussing incremental return.
Retail concentration makes that integration particularly valuable. Major supermarkets such as Woolworths and Coles sit at the centre of Australian grocery buying, while Aldi and Costco serve different missions, basket sizes and price perceptions. Media placed through retailer networks can influence consideration inside those environments, but its commercial value depends on product availability and category economics. MMM links those elements and reveals whether incremental media demand is worth the associated trade and promotional costs.
Good governance determines whether sophistication produces better decisions. Model documentation should explain data definitions, channel transformations, elasticity assumptions and uncertainty ranges. Experiments should be designed to challenge the model rather than merely confirm it, and calibration decisions should be visible to commercial teams. India-focused communities such as Passionate in Analytics can support this progress by sharing applied case studies that connect statistical methods to distribution, promotion and consumer behaviour.
The evolution of Indian FMCG modelling has ultimately moved the discipline from retrospective sales explanation toward continuous, evidence-based allocation. Its strongest lesson for Australia is equally practical: trustworthy incrementality emerges when media, price, promotion and distribution are treated as interacting parts of one system, with causal experiments used to challenge and improve the model continuously.
