Monday, 13 February 2017

Marketing Mix Modelling: Challenges and Best Practices

The optimal allocation of funds across different channels of marketing is crucial for all organizations since investment decisions need to be made depending on the contribution each channel makes to the overall sales. Marketing Mix Modelling (MMM) helps quantify the contribution of various factors to sales and recommends fund allocation across multiple channels in order to achieve better ROI, efficiency and effectiveness. MMM is an analytical approach which is widely adopted across industries today to measure and optimize marketing budgets. While MMM has proved to be an effective technique to allocate funds more analytically, its implementation is key to achieve optimum results.
Key Challenges:
  1. The data needs to be understood thoroughly and delinked from mixed effects of any overlapping campaigns
  2. Validation of coefficients with borderline significance is important to maintain stability and consistency of new data before implementation
  3. Irregular market segments with thin and discrete history is a serious challenge for modelling and prediction. Such markets are dealt by ‘Proxy Modelling’ using higher levels of data and predictions, which are levelled by their proportional representation in the portfolio
  4. During implementation, the prediction and optimized allocation is made for all market segments by default without considering their real-time demands. If needed, depending on the marketing plans and priorities, budgeting and allocation has to be regulated as per the prevailing business or forecasting scenarios
  5. Thin market segments with irregular history may not appropriately fit for building ‘S curves’ to reflect the sensitive cost-revenue relationship; such market segments can be predicted by grouping them based on business considerations
Best Practices
  1. For superior insights, the objectives of MMM and what it plans to achieve should be clearly set by:
    1. Identifying drivers of revenue and quantifying impact
    2. Optimizing spend across different marketing channels for maximum return
    3. Time-series forecasting for future plan of action
  2. Every touchpoint in the customer journey should be defined, tracked and measured for proper accounting of cost and revenue components by marketing levels such as geography, channel etc.
  3. Revenue regressed on cost or raw variables (clicks, impressions) by channel should be accounted and available at the same granular level (either through derivation or already set up by the company.)  Data should be set at the same level – especially the cost variables since they are available at higher levels and have to be broken down to the lowest granular level on which the model is built
  4. It’s important to check key variables for both statistical and business significance
  5. Building an ‘S curve’ (sigmoid shape) to plot the growth rate of revenue as a function of cost in percentile scaling will help determine the ‘Spend Limits.’ Fitting of ‘S Curves’ to data should be done by tuning the shape, scale parameters of a chosen distributional form with respect to the empirical distribution of cost and revenue
  6. ‘Optimal Point’ should be discovered where revenue growth rate is maximized for a given cost
  7. Cost allocation by the channels that maximize the overall revenue should be optimized
  8. Test and control markets should be compared and then the feedback can be used to refine the model performance
Common Mistakes:
  1. To prevent incorrect results, disproportionate values and volatile distribution of data should be checked, trimmed and transformed
  2. Missing data should be dealt with before modelling, else it could lead to inefficient results
  3. Do not choose incorrect transformation for data in order to ensure the linearity and stability of the variables
  4. To avoid wrong attribution to marketing promotions, time-series data should be converted into a cross-sectional form before building the models by accounting and adjusting for seasonality and auto correlations in the data. If needed, models should be built on de-seasonalized and stationary data
  5. Data must be aggregated and summarized at requisite time intervals to correct the data imbalance like missing revenue to a cost point or vice-versa
  6. Spend limits are acceptable up to the saturation point in an ‘S curve.’ Promotional costs should be planned in a range between the discovered minimum and saturation points to avoid losses. Similarly, a minimum spend threshold should be maintained for stable markets
Since the stakes are high for brand building, following the best practices while implementing the model and taking care of the challenges that come along the way can provide high ROI and improve marketing decisions extensively. An MMM model can provide a consistent and more accurate set of metrics, which will help marketers influence the overall consumer journey.

No comments:

Post a Comment