Media mix modelling (MMM) is a statistical technique that uses historical, aggregate-level data — advertising spend, GRPs, promotional activity, seasonality, competitor actions, and sales outcomes — to estimate the contribution of each marketing channel to overall business results. MMM is privacy-safe, does not require individual user tracking, and provides a long-term view of channel effectiveness that complements shorter-term campaign measurement. For CTV, MMM is particularly relevant because it sidesteps the identity and attribution challenges that undermine MTA, and because it naturally handles the long consideration cycles that CTV drives.
This article explains how MMM works, what data inputs are required for CTV, how to represent CTV correctly in the model, which vendors have India CTV capabilities, and the limitations you need to understand before commissioning an MMM study.
How media mix modelling works
MMM builds a statistical regression model that explains variation in a business outcome (revenue, sales volume, sign-ups) as a function of marketing and non-marketing variables. For each input variable — TV spend, digital spend, price promotions, seasonality — the model estimates a coefficient: how much does a unit increase in this variable correspond to an increase in the outcome?
The model uses historical data, typically 2–3 years of weekly or monthly observations. Each row in the dataset is a time period. The model learns the relationship between inputs and outputs over time. Once trained, the model can estimate what would have happened to sales if the marketing mix had been different — higher digital spend, lower TV spend, no promotion in a given quarter — which is how it informs budget allocation decisions.
Data inputs required for CTV in an MMM
Including CTV in an MMM requires consistent historical data across multiple variables:
Media spend data
Weekly or monthly CTV spend is the primary input. This needs to be consistent and complete — gaps in the spend data create gaps in the model. For campaigns that ran through multiple DSPs or buying platforms, the spend data needs to be aggregated to a single weekly number.
Impression or GRP equivalent data
Spend alone is an incomplete proxy for media exposure. If CPMs varied significantly over the period (which they do in programmatic buying, where CPMs fluctuate with auction dynamics and seasonality), the same spend level can produce very different impression volumes. Including impression data or a GRP equivalent gives the model a more accurate measure of the actual advertising pressure delivered in each period.
Reach and frequency data
For channels where saturation effects are important — TV spend tends to show diminishing returns as frequency increases — reach and frequency inputs help the model estimate adstock (the carry-over effect of advertising over time) more accurately.
Outcome data
The business outcome being modelled — weekly revenue, monthly sales units, app downloads — must be available at the same time granularity as the media data. If CTV data is available weekly but sales data is only available monthly, the model's precision for short-term effects is limited.
How to represent CTV in an MMM when data is siloed
The practical challenge for most advertisers is that CTV impression and performance data is siloed across platforms. JioCinema data is not in the same system as YouTube CTV data, which is not in the same system as programmatic DSP data. For MMM, this data needs to be consolidated into a single weekly spend and impression series for the CTV channel as a whole, or broken into sub-channels (premium publishers, mid-tier publishers, programmatic open exchange) if the model has enough data to distinguish between them.
Steps to consolidate CTV data for MMM:
- Pull weekly spend and impressions from each CTV platform or DSP where campaigns ran.
- Aggregate to a single weekly total if treating CTV as one channel, or maintain sub-channel breakdowns if treating premium and programmatic CTV separately.
- Normalise to consistent units — impressions, not clicks or completions — to ensure comparability across platforms with different metric conventions.
- Flag periods with non-standard activity (ad-free weeks, major creative changes, sporting event sponsorships) so the model can control for these outliers.
The adstock problem: how CTV carry-over effects work in MMM
MMM models use an adstock transformation to capture the fact that advertising effects do not disappear immediately — they carry over into future periods. A CTV campaign run in week 1 may still be influencing purchases in weeks 3 and 4. The adstock rate determines how quickly the effect decays. For TV advertising, adstock rates are typically higher (slower decay) than for digital channels, because TV builds brand memory and consideration that persists after the campaign ends.
Getting the adstock rate right for CTV is important. If the model assumes the effect decays too quickly, it will underestimate CTV's long-term contribution. TV advertising in category research suggests adstock half-lives of 2–8 weeks depending on category — consumer durables and financial products at the longer end, FMCG at the shorter end. The correct rate for your category and market needs to be tested within the model calibration process.
MMM vendors with India CTV capabilities
Building an MMM requires statistical expertise, marketing data integration, and model calibration experience specific to the Indian market. The following vendors have worked with Indian advertisers on MMM studies that include TV (linear and/or streaming):
- Analytic Partners: One of the most sophisticated MMM-plus-MTA (unified measurement) vendors globally. Has India engagements, typically with large multinational brands. Integrates CTV data where available.
- Nielsen Marketing Mix: Long-established in India for linear TV MMM. CTV integration is a newer capability being added to their India model infrastructure.
- IRI (now part of Circana): Strong in FMCG sectors. India CTV integration is limited compared to their US capability.
- Local analytics firms: Several India-based analytics consultancies (Hansa Research, Madison's analytics arm, etc.) offer MMM services calibrated to the Indian market, including CTV where data is available. These tend to be more accessible in cost than global vendors.
The lag problem: MMM takes months to produce results
The most significant limitation of MMM for tactical campaign management is time. Building a credible MMM model requires 18–36 months of historical data, several weeks of modelling work, and iterative validation. The results come months after the campaigns have run. This makes MMM unsuitable for in-flight campaign optimisation or for evaluating a single campaign's performance — it is a strategic tool, not a tactical one.
For most India CTV advertisers, MMM is appropriate as an annual or semi-annual investment to calibrate the broader media mix. It should be complemented by faster-moving measurement approaches — incrementality tests, brand lift studies — for campaign-level decisions.
India context: practical guidance on CTV and MMM
India's MMM ecosystem for CTV is still developing. Several factors shape what is feasible:
- Data consistency: CTV in India is relatively new as a significant media channel. Brands that have been running CTV campaigns for less than two years may not have enough historical data points for a robust MMM that includes CTV as a distinct channel.
- Linear vs streaming split: Many Indian brands still run the majority of their TV budget on linear (GEC, news, sports). If CTV represents less than 15% of total TV spend, it may not be modellable as a separate channel — it may need to be included as part of a broader TV channel or separated only in later model iterations as the spend grows.
- IPL and cricket seasonality: India's media calendar is dominated by cricket, particularly the IPL. Any MMM for India must account for the extreme media inflation and audience spike during IPL — and must correctly attribute sales lifts to the IPL period versus general CTV advertising. Failure to model the IPL as a separate factor will contaminate the CTV coefficient estimate.
- Recommended approach: Start with an MMM that includes linear TV as a channel and treats CTV as a sub-component or separate line if spend is sufficient. Run annual model refreshes as CTV spend grows. Complement with quarterly incrementality tests for in-year CTV campaign decisions.