Data-driven marketing musings

Data-Driven Marketing Musings

In the last 17 years of my marketing career, one of the things that have grown in importance is data and having a results-driven mindset. Data-driven marketing is key to effective marketing campaigns. Not only is it important to have this mindset because it helps you make strategic decisions with confidence, but it is also a tool you can use to get buy-in from stakeholders.

I have outlined below the principles and models I have used repeatedly throughout my career. This is not an exhaustive list, please add your favourite principles and models in the comments section. I'd love to hear about the ones that work best for you.

Cheatsheet to Attribution-models

Below is a high-level overview of the key models to consider when deciding on attribution, that is, which marketing/advertising channels generated results. There are 3 categories of attribution models:

1. Rule-based attribution models rely on statistical business rules to determine the method of attribution.?

• First-click / first interaction
• Last-click / last interaction
• Even / linear
• Position-based
• Last non-direct click
• Last Google ads click
• Time decay
• Uneven / non-linear?

2. Data-driven /Algorithmic attribution models | Uses statistical modeling and machine learning techniques to derive the probability of conversion across all marketing touchpoints?

3. Probability attribution models incorporate addressable and non-addressable media. Attribution is based on the probability of the user also viewing the brand’s adverts on those channels.?

Here's an infographic I created when working for Omnicom Media Group. https://annalect.com.au/insights/simple-guide-to-attribution/

If you don't have the data or tools available to you, I like to utilise the Correlation Coefficient Formula in Excel.

Simple correlation Correlation Coefficient formula is one I often rely on when I don’t have the platforms available to me to use the above attribution models. Correlation identifies the degree to which variables in our data set are dependent upon each other. It’s important to note that Correlation is not causation. But I personally still find correlation useful as it gives you a high-level glance at the relationships between key variables.?

For example, let’s start by identifying the X variable, which could be ‘Leads’ or ‘Opportunities’, and its dependency on Y, this could be paid media spend, paid social media advertising, organic social media posts, website visits, etc.

Some of the ones I like to compare are:

• Impact of (y) website visits, (y) paid media spend, (y) paid social media spend, (y) social media posts, (y) emails sent
• On (x) leads, (x) opportunities, (x) marketing qualified leads, (x) sales from?

The sky’s the limit when it comes to what you can compare, so long as you have an adequate data set (adequate = statistical significance).

Using Excel’s Correlation formula, setup your data as outlined in the GeeksforGeeks website, refer to the source link below:

https://www.geeksforgeeks.org/how-to-find-correlation-coefficient-in-excel/#:~:text=In%20Excel%20to%20find%20the,the%20cell%20range%20for%20both .?

1.0 - This indicates a positive correlation

0.5 - This indicates a moderate correlation

-1.0 - This indicates a negative correlation. This means, a strong opposing relationship. In other words, if (x) is leads, and (y) is emails sent, then the more emails sent, the less likely you will see an increase in leads.?

The closer to 1, the stronger the correlation.?The closer to 0, the lack of correlation as it’s almost zero or is zero.

Some useful basic rule of thumb principles for other market research studies and gathering data:

• When conducting A/B testing, traffic/visits must be at least 5,000 visits minimum
• But measuring conversions is key to deciding on the winning set. At least 1,000 conversions. In saying this, I will always fall back on the absolute minimum to hit statistical significance in any market research sense, which is 100 units. This sample size is a representation of your population set.?
• Time: at least 7 days of testing. Spend the time you need to get the data and hit statistical significance
• Market research: 100 respondents minimum. Ideally, 1,000 for a B2C research piece, but 100 will suffice if you don’t have the respondents or budget to reach 1,000.

Caveat - there are many variables at play when it comes to hitting statistical significance. It also depends on the ultimate size of the set you’re deriving the sample size from. Some professionals believe 30 is an adequate number when it comes to some research methods.

AI-powered platforms

If you're looking to digitise the way you report to make sound marketing decisions, then I urge you to consider the AI-smarts behind the platform.?

When it comes to comparing data sets based on numerics, there are endless tools available. But what’s important to factor in is non-numerical data. How do you measure important non-numerical data?

Human reasoning is required to make AI work effectively when determining non-numerical data.

The key is to attribute numerics against the data. For example, when monitoring ‘Innovation or technological trends’, rank a Risk rating out of 10.?This can get quite overwhelming, which is why it's worth considering AI-powered platforms to help you with a shorter path to actionable insights.

If you'd like an example of an AI-powered platform that provides a 360-degree view of financial and non-financial performance, visit Skyjed www.skyjed.com Or visit the 360 product domain framework webpage directly https://www.skyjed.com/features/product-domain-framework.

If you have any other useful insights or rule of thumb principles, please comment below.