Spend Efficiently: Four Lessons on Optimizing your Media Investments

Optimizing media budgets can be quite a challenge in today’s omnichannel world. Increasing amounts of information become available to marketers, who are expected to take all these factors into account in their decision-making process. They face questions such as:

  • Which media channels are the most effective?
  • How do these (online or offline) media channels interact with each other?
  • What is the perfect timing for campaigns?
  • How do special events, competitor pricing or external factors affect decisions?

Most marketers will agree that advanced analytical techniques are required to model this complex landscape. However, widely used reporting systems like Google Analytics don’t automatically take the relevant factors into account. As we can see in the following example, this leads to sub-optimal media investment decisions.

Imagine your goal for this month is to maximize the amount of conversions without increasing the total media investment. This requires the reallocation of media budgets among channels. The following data from Google Analytics is available over the previous month:

Campaign Clicks Cost CPC Conversions Conversion Rate Revenue ROI
Display 5.342 € 10.000 € 1,9 1.023 19.1% 100.254 € 10
Facebook 3.253 € 5.000 € 1,5 324 10% 28.512 € 6
AdWords 12.374 € 30.000 € 2,4 1.894 15.3% 164.778 € 5

What would be the optimal re-allocation of media budgets over these channels? The Display channel looks promising due to its high return on investment (ROI) and high conversion rate. It would be perfectly reasonable to assume a higher investment in this channel would lead to the highest revenue. Upon further analysis, however, the opposite could also be true.

Optimize your media investments with the following four lessons.

1. Use Multi-Touch Attribution (MTA)

When studying the table above, it is important to consider how each of these metrics is calculated. Many systems still use the last-click model. This model assigns the last touchpoint the customer has interacted with all the credit for the conversion, and nothing is attributed to other channels in the customer journey. The last click model therefore overvalues the channels at the end of the customer journey, and undervalues the channels at the beginning of the customer journey. Google and other search channels are often heavily overvalued using the last-click model.

When implementing multi-touch attribution (MTA), the first step is to capture all relevant contact points with your customers to create a realistic representation of the actual customer journey. The MTA model can then be used to calculate the added value of each channel in the customer journey, taking the order of touchpoints and any interaction between them into account. Using this information, the exact contribution of each individual touchpoint can be calculated.

In the example above, the type of model is used to calculate the performance metrics is not defined. So, what exactly are we looking at when we see 1.023 conversions for Display?

  • If conversions are calculated using a last-click model, the 1.023 conversions in the Display channel indicate Display was the last-clicked channel in a converting customer journey 1.023 times. We can’t conclude that Display was actually responsible for 1.023 conversions.
  • If conversions are calculated using an MTA model, the 1.023 conversions in the Display channel indicate Display is responsible for 1.023 conversions.

Lesson 1: it is important to know the meaning of the metrics you are looking at, as well as how they are calculated. For media investment decisions, you should always use the more realistic MTA model to attribute conversions.

2. Consider Marginal ROI Instead of Total ROI

Looking at the data above, it seems reasonable to shift budget from AdWords to the Display channel, as the Display channel has the highest ROI. However, the law of diminishing returns states the more you invest in a certain channel, the lower the marginal returns of a certain channel.

investment elasticities media budget allocation

The Display channel could already be relatively saturated, such as at the investment level of €10,000 in the graph above. Investing more in display will not give you an ROI of € 10, but a marginal ROI of € 1. Look at the current level of investment in Facebook. This channel is not yet saturated, and has marginal ROI for an investment in Facebook is € 5,7.

Calculating marginal ROI and investment elasticities is more complex in comparison to calculating ROI, and will require non-linear models. Logarithmic or s-curved models are often used. It is also necessary to take additional factors such as salary weeks and events into account (see lesson 3).

Lesson 2: making decisions based on marginal ROI instead of on ROI can yield very different results. Steering on marginal ROI is more accurate. 

3. Correct for all Factors of Influence

Media and sales form a complex landscape with many factors of influence. The example above could yield completely different results under a strong price proposition.

Factors of influence are important to include in predictive models. Pricing strategy, the weather or holidays can all influence sales and media effectiveness. Any relevant information can be included in modeling, even economic trends or events like natural disasters or terrorist attacks.

It is very important to note that some of these factors, like big launches of new products, interact strongly with your media effort. Factors that do not interact with media should be included as control variables. Being aware of this interaction and know which factors have a strong impact on your channels.


Lesson 3: in order to allocate the media budget optimally, find the factors of influence for your business and incorporate these factors into both MTA- and investment elasticity models.

4. Do not Skip Levels

In order to make models work in practice, it is strongly advised to work towards the final goal using a stepwise approach. Start by building an understanding of the models, then implement them correctly, and only when you are certain the output makes sense, start interpreting it and base decisions on it. To prevent infamous black box models, create a plan detailing how to implement the models within the organization. Sharing results with the main stakeholders can be an important aspect of success.

Lesson 4: do not skip levels, and make sure you fully understand the models. Create a plan on how to implement the models within the organization.

Arno Witte

Arno Witte

Lead Data Science