How to succeed with marketing measurement in the AI and data privacy era

Seasoned marketing teams know the ROI figures in their ad and analytic platforms are wrong. Not only does increased user privacy limit the data, but most budget decisions are hard to make, even with perfect data. For example;  

• Is retargeting and branded search really as good as it seems?  How many of these conversions would we get anyway? 

• What is the ROI of my social media campaign? I use discount codes and UTM tagging of links, but surely there must be some effect that is hard to capture with data?  

• What is the effect of my latest event and tradeshow appearance?  The sales department say they get more inbound calls in the weeks after, but it’s hard to measure. 

• What is the effect of internal factors like new product releases and other external factors affecting my business for example weather, interest rates and energy prices? 

Larger strategical questions like “Is brand X really worth investment or is it a race to the bottom against competition?” or “Our brand strength is different in Sweden and France so we need to know how much we should invest top-funnel for maximum long-term growth,” also requires input from many factors from outside the usual digital domains. 

2. Start testing for true incremental effect 

To find out the true value of a marketing channel, you need to set up an incremental test. This will show you how much effect you would have without the channel and from there you can calculate its value. The golden standard here is a GeoLift test that will balance the effect from your different regions. While any channel can use GeoLift testing, digital ad channels that can be split by region are ideal. Any marketing team can set up a GeoLift test and after 2-4 tests you will find that you have solid data to back budget decisions for 1-2 channels. Most of the times, the findings using GeoLift are 20-65% different from what analytics and ad platforms tells you. 

3. Build a statistical model (MMM) to catch all effects and verify 

The next step is to incorporate the findings from the tests in a model to calculate the effect and find out the optimal budget for all channels. 

In the creative golden age of advert­ising during the 60’s advert­isers started to measure the effect of advert­ising using statistical models. If you had time series of your advert­ising investments and your sales, you could measure the effect of advert­ising. The fancy name for this type of model became Marketing Mix Model, MMM. The downside of this method was that it was very time-and resource consuming and therefore reserved for only the largest advert­isers. 

Today, in the age of cloud computing and AI, time and resources are less of an issue. Any marketing department can use a statistical model to measure the effect of marketing investments cross-channel, online and offline. When the model is verified by additional GeoLift test, most of the time the model will be proven as strikingly accurate. 

4. Keep working with basic digital measurement 

The new way of working does not completely disqualify the old. Basic digital measurement in for example Google Analytics, Meta or Google Ads are still great for many purposes. They can answer more granular questions and are still crucial to the complete understanding of how your marketing works. Therefore, you should still put efforts towards applying the latest tech upgrades in this domain like server-side measurement and Google/Meta signal engineering. 

What needs to be in place 

To succeed with modern marketing measurement, you need certain components in place. These can be created inhouse or be built together with external agencies and vendors. 

• Statistics skillset - To succeed with GeoLift testing and statistical modelling you need statistics expertise to set inputs and make business sense of output. GeoLift is fairly easy to grasp for anyone with basic statistical knowledge, but modelling requires advanced experience. The statistics skillset differs from basic digital measurement that is usually skewed towards platform use and tag implementation. 

• Data governance – You need have a certain level of control of your marketing data. If you already have a marketing data warehouse in place, you have a good starting point.  

• Marketing Mix Model - There are open-source econometric modelling packages like Metas Robyn or Googles Meridian that you can use free of charge. There are also many MMM vendors with proprietary models and data connection technology that help you through the modelling process.