At the initial stages of Market Mix Modelling (MMM) projects one of the main points for discussion is the scope of modelling. Based on the project objectives, the project team needs to select the suitable model structure that allows to extract insights and recommendations at the required granularity. In this note I will address the most important aspects of the process.
Time Frame
Let’s start with the most straightforward one – the time frame of the analysis. Generally most of market mix models cover a 2-3 year period. Anything below two years might be too short to provide robust results. In the longer time period, broader market trends can have stronger impact than marketing activities. In addition, the longer is the time period, the more series might not be available in the required lenght. The best approach is to use 2-3 years of weekly data and extend the range of findings through bi-annual model updates.
Mix modelling is an ideal approach to measure longer-term carry over effects of brand-building advertising. We should allow for this through choosing the right end date point, i.e. the modelled period should include 6-8 weeks of data after the last campaign we went to measure properly.
Geographical Scope
Measuring effectiveness of marketing across different geographical markets is one of the most common scopes of modelling. It allows generating individual sets of insights for each modelled market. Therefore it is interesting for clients who want to optimise budget split across markets and/or they plan media and campaigns for each geographical market separately.
Geographical scope of modelling is essential when individual markets are very different, e.g. different demographic composition, climate, seasonality etc. and therefore their media plan and channel split need to be adjusted to market-specific characteristics.
For example, some products such as sun-block creams or sun glasses are in some markets only seasonal goods with very low base sales out of season. On the other hand, in markets with warmer climate, the same products can have relatively flat sales through the year. Marketing strategies should reflect these differences
The same holds if marketing and media activities vary largely between the markets. One aggregated model would not allow capturing correctly every activity taking place only in some markets. The sum of marketing effects from disaggregated models may be larger than the same effect estimated using aggregated series. The gap is likely to grow with the number of geographical units, the sales are disaggregated into.
Separate models allow for testing new media strategies at a smaller scale. Local Test&Learn strategies are suitable for the advertisers who rely on the local media.
On the other hand, in highly disaggregated models we might not be able to pick up the impact of activities planned at the broader level. As an example, smaller digital and pay TV campaigns which are in AU planned mainly at the national level cannot be well measured through granular models.
Product Categories
The product categories model split depends on how heterogeneous individual product categories are. The more they cannibalize each other, the stronger is the case for aggregating them into one model. If it is essential to build separate models to estimate halo advertising effects, one needs to control for potential cannibalization between product categories.
If product categories do not compete with each other, and they have distinct dynamics and consumption patterns - individual models will be useful. This holds, especially, if they compete with different brands. Even if the same brand advertising covers them all, the response can differ. Separate models can provide wealth of insights.
Product categories model structure is definitely more interesting than geographical scope as the communication activities of individual categories overlap, interact and affect each other. Good understanding of those dynamics enables building effective communications strategies.
Let’s take an example of cosmetic brand consisting of five product categories. The product categories serve different purpose and target different demographic groups. As such they do not cannibalize each other. However, they all benefit from brand equity on one side and contribute to building it on the other.
The project with separate models for each product category within the brand provided valuable insights with implications for timing of product campaigns and media channels. Optimised campaign execution and channel use leveraged brand equity and improved long-term business results.
Distribution Channels
The number and heterogeneity of used distribution channels provides the next angle for model scoping. The first idea that probably comes to one’s mind are retailers selling their products online and in “brick-and-mortar” stores. These two channels may attract different types of customers as well as interact and support each other. Modelling sales in both channels as well as intermediate metrics such as store visits and website visits build better understanding of how the two channels interact and how they differently respond to paid advertising. The media budget split can then be optimised to maximise overall sales and other engagement KPIs.
In addition, separate modelling of several offline distribution channels provide grounds for improving communications strategies. For example, soft drinks are often distributed through grocery stores, petrol and convenience stores and small independent kiosks. Modelling at the level of distribution channels can provide learnings for more effective timing of media and messaging throughout the year. While during the summer season, the cyclical uplift of sales in petrol and convenience stores dominate the sales, flat base sales in groceries are more important in colder months. This has implications for campaigns where the message, used channels and activations should be scheduled based on the time of the year and which distribution channel is the current priority.
Consumer Segments
If a marketer applies segment-specific marketing strategies – it imakes sense to build models separately for individual segments. Segment-specific models can be instructive for all aspects of campaign planning. Alternatively, communication strategies can be designed to maximise the overall effect and equity growth.
Given that mix modelling is often associated with FMCG and scan sales data, this part may deserve a bit of explanation. For the companies that do not have access to individual-level data, MMM has been the primary analytical method to understand sales dynamics. However, for those product or service-providers that collect customer-level data and can distinguish customer segments, mix modelling is a very effective method to quantify what drives acquisition of new customers and how well marketing activities perform.
Customer segments are based on age groups; channel through which they were acquired; shopping habits, demographic attributes or what type of product(s)/service(s) they prefer. The model split based on (sufficiently big) customer segments will enable better targeting of potentially more valuable customers.
To Sum Up
This note provides examples of what benefits more granular modelling structure offers. However, brand and marketing managers are increasingly cautious of costs that a higher number of models implies. These can be significant. Even though more granular modelling benefits from centralised data preparation; result reporting, the range of insights, simulation tools and post-project consulting are getting more complex.
Marketers weigh the benefit of comprehensive insights and recommendations against higher costs of modelling. In some situations, a simpler structure is sufficient. In other cases, the mix modeller should insist on more disaggregated model structure to meet the project objectives.
The table below summarizes the discussed points. Spending some time on better understanding what model structure could best guide brand marketing can save dollars in research budget, and, most importantly, spare disappointments with the lack of actionable insights.
When Useful |
Benefits |
Watch-outs |
|
Geographical Markets |
Media and marketing activities differ across geographical units. For media tests done at the regional level. Heavy use of local media. |
Provides base for geographical budget split. It can evaluate media planned at very granular level. Gives detailed recommendations. |
Level of disaggregation aligned with units of media planning. If planned at a higher level, try to get disaggregated metrics. |
Product/ Categories |
Brand contains a range of categories and strives to maximise overall brand equity. |
Provides base for budget split across categories. Optimise media channel usage, timing of campaigns and their spend, taking into account halo effect. |
Control for possible cannibalization between categories. What level of disaggregation makes sense. |
Distribution Channels |
Different consumers /consumption patterns across channels, various seasonal dynamics. | Enables better planning of campaigns and more effective campaign timing. |
Control for potential cannibalization. How to take into account synergies of having several channels. |
Consumer Segment |
Customer segment data are available, and there is a reason to believe that segments have different dynamics. | Optimisation across segments allow segment-specific message and focus across channel. | Customer segments must be sufficiently big and aligned with marketing strategy. |