When seeing MarketPulse in action, many people have asked me about how response curves work, and what they really mean. In this article, I'll show you a few examples of marketing response curves, and explain how they relate to the real world, and how they reflect the Return on Marketing Investment (ROMI) for different campaigns.
Assumptions behind the MarketPulse approach
In order for our approach to work, it needs to make three key assumptions:
- Marketing results are not random, but are heavily influenced by some of the things you and your competitors do, such as marketing, promotions, price changes, etc.
- Usually, the effects of these actions are not immediate, but take time to build up, and they also last for a period of time, decaying until the effect has died out
- It is not necessarily productive to try to understand these effects from a behavioural point of view; rather, it is possible to get insight by taking a purely data-driven, quantitative approach
The second of these points is manifested by using response curves to model the effect that each marketing action has on sales, from the time the action takes place, until it no longer has any effect.
Response Curves: Some Examples
The graphs below (from screenshots of an actual model using our software) show response profiles for some representative marketing activities. The one on the left (blue curve) shows the effect of TV advertising, while the one on the right shows the same blue curve, but also the effect of press spend (green curve) superimposed.
These response curves show the impact on sales from one unit of spend (say, one dollar or pound) over a 10 week period. Note the following:
- TV reaches a higher peak (just under 0.4) than press (about 0.3)
- TV reaches this peak faster, within about a week versus about 10 days for press
- However, TV decays much more quickly (by around week 5), while the effect of press lasts longer (still having an effect at week 10, where the graph ends)
- The net effect is that, although the press curve is not has high, it lasts longer, and the total area under the curve is greater for press (about 0.12 for press versus 0.09 for TV, shown in the graph legend)
So these response curves suggest that press is having a greater effect than TV for the same amount of money spent, even though it has a lower peak and takes longer to get there.
For another example, the graph to the right shows the effects of a promotion campaign, and is actually composed of two different effects:
The total area under the combined curve is what matters (.039), showing that the positive effect (.122) more than offsets the negative effect (-.084), suggesting that this campaign had a positive ROI.
Response Curves: their Meaning for ROMI
The examples above should make it clear that response curves show the impact on sales of £1 of spend, over time. Some marketing ROI metrics attempt to give you the total impact, but we believe the time profile is more useful and interesting because:
- It tells a rich "story" about the effect of each marketing lever, which makes more sense and allows for deeper comparison and understanding
- It allows you to compare different campaigns of the same type, along measurable dimensions (e.g., peak, time to reach peak, length of decay, and "shape")
- It makes possible the optimization techniques we deploy in MarketPulse, which allow for a better fit (for the right reasons) than standard multivariate regression
The meaning for Return on Marketing Investment is that the area under the curve is your total return, not just the peak impact on sales. This is an important and practical insight.
How we get response curves and use them
We derive response curves for our clients using our proprietary software, which manages a database of spend on all known drivers, and uses several optimization algorithms to estimate the response curves against known output metrics such as sales or market share. During the process, we examine the variance between predicted and actual sales, to help identify where critical drivers have been missed, and have identified several unobserved market dynamics this way.
Although we inspect the curves to make sure they are reasonable, it is important to note that the approach is data driven, and does not use inputs such as expected impact or duration in the process. We have found that this purely data-driven approach is robust and reliable.
The resulting curves are very useful. In additional to "explaining" how various marketing actions impact sales, they help us to create and interpret scenarios around different marketing spend allocations. This helps to improve company results, by reducing spend on less productive activities, and increasing those activities that work.