Using Data in Digital Marketing – Econometrics

Econometrics in Digital Marketing involves finding the relationships between different economic factors. It can explain, for example, how much a change in on-line advertising will affect a critical metric such as on-line sales. Also that the three factors with the biggest influence over sales are a particular set of Social Media Ads plus a TikTok campaign plus on-line searches for your products.

At its heart is a model (i.e. a mathematical equation) that represents how these different factors are connected. It depends upon data, and (typically) the further back the data goes, the better.

These assertions are backed up by statistics and probability. You will see metrics that explain how strong these relationships are and how well they fit real-world data.

How it works in Digital Marketing

To illustrate the value of Econometrics in Digital Marketing, we will show how it is currently used in a UK University. It’s a big organisation but the principles apply whatever the size of organisation.

1. What are the outcomes that matter?

The first step is to establish the metric that represents real value to the organisation. In the case of our University it is the number of Applicants for Undergraduate Programmes. We want to grow this metric because the Tuition Fees are a dominant source of funding.

Why is this step important?

Digital Marketers have a wide range of channels and activities at their disposal e.g. web pages, social media Ads, blogs, posts, videos, and so on. However, because you are working to a limited budget and tight time constraints, you will want to focus on the particular channels and activities that have the greatest influence over the metric that matters most to the business.

You therefore start by agreeing what this metric is e.g. the number of University Applicants. You then use Econometric principles to work out which channels and activities exert the greatest influence over this metric.

2. Which channels and activities influence the metric that matters?

This is where the historical data comes in (we’ll look at what happens if there isn’t any history further below).

Data is needed for all the channels and activities of interest. And there could be many. You might, for example, suspect that it’s your Facebook Ads campaign that is having the greatest impact, aided and abetted by a series of specialist blogs, plus regular organic TikTok and Instagram posts.

But a hunch is not enough, clarity and evidence are required.

You start your Econometric Model by gathering data into one place for all the channels and activities that might be relevant. You then assemble the data into a Matrix – think of this as a set of columns and rows in a spreadsheet.

The first column typically contains the dates covering the period for which you have data. A column is then added for each channel and activity in the scope – as in Figure 1.

Figure 1 – Matrix used for Econometric Modelling

E.g. the University Matrix included a column for Facebook Ads which contained the number of Ads posted on each date listed. It also contained columns spanning Google metrics, blogs published, TikTok campaigns, Instagram campaigns, SEO metrics, Domain metrics – and a whole host more.

The final column contains counts for the key metric of interest – in our case a count of Applications received on each date listed.

Software then calculates the strength of correlation between all the columns in the Matrix (i.e. the extent to which the increases and decreases in one column mirror those in another column).

This will give a first insight into whether there are Predictors for the key metric i.e. columns containing counts that correlate well with those in the final column i.e. for Applications.

For example, in our University, out of all the metrics available, the best Predictors for Applications were levels of engagement with Facebook and Google Ads, and website page views arising from Google Searches.

Figure 2 shows all the values plotted cumulatively (useful for seeing how well two sets of numbers move together and how quickly total values grow).

Applications (the key metric of value) are shown by the dashed line. Combined totals for Ad Engagement and website page views arising from Google Searches are shown by the columns.

Figure 2 – the cumulative growth in Applications and the total for Ad Engagement and Page Views from Google searches

There is a good level of correlation between the columns and the line (they increase at the same time and more or less to the same degree), but it’s not perfect. It’s important therefore to quantify the degree to which the metrics shown by the columns serve as Predictors for the Applications shown by the line. This involves some supporting statistics which are explained further below.

3. Digital Marketing works!

For the Digital Marketer this is something of a Holy Grail. You can now explain how the growth in your key metric (e.g. Applications) was achieved – and therefore what you need to focus on for it to grow further.

You can also quell any doubts that your Employer or Client might have about the value of Digital Marketing to their organisation.

They will, however, want to know about spend, in particular how much of the money spent on Digital Marketing influenced these outcomes, and what it will cost going forward.

4. Connecting spend with the metric that matters

The first piece of reassurance needed is that the spend so far has actually made a difference.

A new Matrix is required, but with just three columns.

Date
Spend
The metric that matters (e.g. Daily Applications)

Thankfully, in the case of our University, a significant level of correlation was shown to exist between spend and Applications – as shown at Figure 3.

The data is presented cumulatively once again – important for explaining rates of growth needed to achieve a target (Figure 4).

The Applications are shown by the line, and the total spend on Digital Marketing shown by the columns.

The box on the right contains supporting statistics referred to earlier. An explanation is shown below the chart.

Figure 3 – the cumulative growth in Applications and Digital Marketing spend

The coefficient of Spend as a Predictor for Applications (0.07) – describes, on average, how many Applications you get for £1 of spend. A figure like 0.07 is not particular helpful, but we can instead say that 1 Application requires a spend of £ = £14 (approx.) – useful when it comes to working out the budget needed to reach a particular Applications target.
The P Value for the Null Hypothesis – is a piece of statistics jargon. It is, however, the most important piece of information.

This is the probability that the two sets of records used in the chart i.e. the daily records for both spend and Applications could have appeared if there was no relationship between them (known as the Null Hypothesis). The number 1.29E-20 uses Scientific Notation to express an incredibly small number (a decimal point followed by 19 zeros).

In other words, the idea that they are not related can be dismissed. This is, instead, strong evidence that spend and Applications are related.

Strength of Correlation – explains the degree to which the values for spend and Applications move together i.e. if one goes up or down, how much the other goes up or down.

Values that move perfectly together will have a correlation value of 1.0 (or -1.0 if they move in opposite directions). If they have a value of 0 then there is no link between them at all. A value of 0.6 indicates a good level of correlation between the two values.

Whatever software you use – e.g. Python, R, Matlab, Excel – these, and other statistics that help to paint a fuller picture, will be available.

5. Setting Digital Marketing budgets

With confirmation in place that the spend so far correlates with the metrics that matter, it then becomes possible to separate out the spend into component elements and establish which have yielded the greatest influence.

You can also use this relationship to plan for the future e.g. by setting an Applications target, confident that Direct Marketing should help to achieve it, and by calculating the budget needed to get there.

In Figure 3 below, the cumulative totals for spend and Applications have been extended to reflect an Applications target of 7,000 to be achieved by the end of September.

Figure 4 – the cumulative growth in Applications and Digital Marketing spend extended to show the spend needed to achieve a target of 7,000 Applications

But there are caveats and words of caution to heed

At this point there is a lot that the Digital Marketer can explain. E.g.

How Digital Marketing has added value so far.
What you have to do to drive further growth.
How and where spend exerts influence.
What spend is needed to reach a target by a given date.

However, this is all based upon findings from data spanning a particular period of time. Data taken from a different period, or from a much shorter period, might yield a different set of findings.

Also, amongst the supporting statistics is one that explains how much of the variation in the key metric (e.g. Applications) is explained by variation in the Predictors. You might see a low percentage figure (e.g. 25%) even though the Predictors are clearly very effective.

This indicates there are additional causes of variation, and therefore additional Predictors, not reflected in the data used so far.

University Applicants, for example, might also be influenced by advice from teachers, choices made by friends, parental preferences.

This doesn’t undermine the importance of Predictors that you have identified. But it’s a reminder to keep searching for additional ones as they will add to the accuracy of your findings.

What if there isn’t any relevant history?

This will be covered in a future blog, however, a short example can illustrate what’s required.

Let us say that you are launching a membership organisation such as an on-line club.

The metric of value would normally be the number of people applying. But you have yet to recruit your first member.

You have however had a sales funnel in action for a month e.g. TikTok posts > Landing Page > Apply Here button > Application Form submitted.

You can see from the data you have that some engagement with the TikTok posts has taken place, but the time spent on the Landing Page has been very brief.

This metric (i.e. time spent on the Landing Page) becomes your metric of value. And you use data to help improve the design and content of both your TikTok posts and Landing Page in order to lengthen dwell time.

When you are seeing evidence that the Apply Here button is being clicked, these clicks become your new metric of value. Your use of data then focuses upon the improvements needed to achieve more clicks.

This movement along the sales funnel requires some granular data collection and exploration which is not within the scope of this blog. But it will be covered in blogs that follow.

There’s much more to discover

This blog has only scratched the surface of Econometrics in Digital Marketing. There is a whole world to discover about the use of data to influence the effectiveness of Digital Marketing. Look out for further blogs on the subject.

—-

Paul Clarke is the Director of FiguringOutData.com, a Data Analytics Service Provider supporting teams make the best use of data to guide Digital Marketing and Social Media management.

For information about our services and for guidance in the use of data to support Digital Marketing please get in touch using the contact information above.