You can follow fewer and fewer customers from click to purchase. Cookie banners, browser restrictions and privacy rules leave gaps in your data. Your reports show fewer conversions than there really are. That makes it hard to decide where your budget should go.
Still, you want to know what your marketing delivers. Fortunately, there are ways to measure that without tracking every click. This article covers three of them: modelled conversions, marketing mix modelling and incrementality testing. You will also read about how they complement each other.
Why you can no longer track everything
In the past, a website placed a cookie. That is a small file in your visitor’s browser. It let you see which ad led to a purchase. This method is called click-based attribution. Attribution means assigning a conversion to a channel.
That method now breaks down in several places:
- Visitors decline cookies in the cookie banner.
- Browsers such as Safari and Firefox have been limiting visitor tracking for some time.
- People switch devices. They see an ad on their phone and buy later on their laptop.
- Many ads work without a click. Someone sees your video, remembers your brand and later searches for your name.
The result is that your data is incomplete, but not useless. You see part of the picture. The question is how you estimate the rest.
Modelled conversions and your own measurement
Google and Meta fill in missing data with statistical models. These are called modelled conversions. The platform looks at visitors who did give consent. Based on their behaviour, it estimates how many conversions occurred among the other visitors.
This works, but not automatically. The model is only as good as the data you supply. If your measurement is messy, the model’s estimates will be messy too. So start by getting these basics in place:
- A well-configured cookie banner. Make sure the visitor’s choice is passed on correctly to your advertising platforms. At Google, this is called Consent Mode.
- Clean conversion tracking. Only measure the actions that matter, and do not count them twice.
- First-party data. This is data you collect yourself, such as customers’ email addresses. Provided you have consent, you can safely pass it on to improve measurement.
- Server-side tagging. With this, measurement data goes to the platforms via your own server instead of directly from the browser. That makes your measurement more stable.
Please note: modelled conversions remain estimates. They show what the platform thinks happened. They do not prove that an ad caused the purchase. For that, you need the other two methods.
Marketing mix modelling: the big picture
Marketing mix modelling, or MMM for short, does not look at individual visitors. It looks at your total revenue and your spend per channel over time. A statistical model then searches for patterns. Does revenue rise when you spend more on Google Ads? Does anything change when you reduce social advertising?
Because MMM does not use cookies, privacy rules do not affect it. The model also captures the effect of channels that do not generate clicks, such as video, radio or seasonal influences. This gives you a fairer view of each channel’s contribution.
You do not need to buy expensive software for it. There are now free open source options, such as Robyn from Meta and Meridian from Google. Open source means the code is available to everyone.
MMM also has limitations:
- You need data covering a longer period. Think of at least a year, preferably longer, often at weekly level.
- Your budgets must have changed from time to time during that period. If you always spent the same amount, the model can learn very little.
- It gives insight at channel level. It does not tell you which ad or search term works best.
- It is suited to strategic decisions, not to day-to-day adjustments.
So think of MMM as a map of the whole area. For the details of each street, you need other tools.
Testing incrementality
Incrementality is about one question: what would have happened without these ads? A conversion that would have happened anyway adds nothing. A conversion that came only because of your ad is incremental.
You test this with an experiment. You switch off a channel for a few weeks in one region and let it keep running in a comparable region. The difference in revenue shows what the channel really adds. This is called a geo test or holdout test.
Here is how to approach it:
- Choose a clear question. For example: does our brand campaign in Google Ads add anything?
- Choose two regions that are similar in size and buying behaviour.
- Switch off the campaign in one region. Leave everything in the other region unchanged.
- Let the test run for at least several weeks. If your sales process is long, you will need longer.
- Compare the revenue or enquiries of both regions, before and during the test.
Make sure you do not run other major activities in that period, such as a sale in only one region. Otherwise you will not know what is causing the difference. It is also wise not to over-interpret the result. Small differences can be down to chance.
A test sometimes costs a little revenue, because you are switching something off. On the other hand, you find out where your budget really works. That often pays for itself quickly.
How to combine the methods
Each method has its own role. Modelled conversions give you daily figures in the platforms. MMM shows how your budget should be divided across channels. Incrementality tests check whether the outcome of those figures and models is correct.
Suppose you have a webshop selling garden furniture. You spend money on Google Ads, Meta and email. In Google Ads, you see a good result for your brand campaign. Meta appears to deliver little in the reports. Based on that, you are considering cutting the Meta budget.
First, you run an MMM on several years of weekly data. The model suggests that Meta contributes more than the platform itself reports. Brand search traffic in Google seems to partly follow your social campaigns. That is an indication, not proof.
So you switch Meta off for a number of weeks in one region. You compare that region with a similar region where Meta keeps running. If revenue in the test region falls noticeably, you know that Meta really adds something. You therefore keep the budget in place and adjust the split if necessary.
This is an example, not a promise about outcomes. It mainly shows how you arrive at a better decision step by step.
Frequently asked questions
Do I need MMM if I have a small budget?
Often not. With a small budget and few channels, there is too little data for a reliable model. Start with good in-house measurement and simple tests. MMM becomes more interesting when you use several channels and have been collecting data for some time.
How long does an incrementality test take?
As a rule of thumb, a few weeks. With a short purchase journey, that may be enough. For expensive products with a long consideration phase, a longer test is needed. Also avoid scheduling the test in a period with unusual peaks, such as the holiday season.
Are modelled conversions reliable?
They are a useful estimate, but not a fact. The quality depends on your own measurement and the amount of consent data. Use them to steer campaigns day to day, and check the big decisions with MMM or an experiment.
Book a free 30-minute introductory call
Do you want to know how your marketing is performing now that cookies are less useful? We would be glad to help you think it through. In a free 30-minute introductory call, we look at how your measurement currently stands. We discuss which method suits your situation and what you can improve first.
You will not get a sales pitch, but an honest view of the possibilities. Get in touch with Adventio Marketing in Amersfoort and book a time that suits you.
