Your Complete Guide to the Key Analytical Stages and Methods

Vasudha Khandeparkar | Head of Insight

The big question – what every good fact finding mission starts with! The route to the most complex data-led solutions always start with a question. Unless you know the question you are trying to answer and how it fits within your marketing plans, the world of data analytics can seem like a vast, complicated and daunting realm. This guide walks you through the key questions you should be asking at each stage of an analytics project and the techniques you should use to answer them.

Let’s start with a broad question. Consider a scenario where a company wants to identify why their customers are lapsing so they can understand the decrease in their database growth.

Where should we start?

The first stage to get right in a data-led approach is setting the baseline. The first questions you ask paint a picture by asking ‘what’ happened. With our scenario in mind, you would start with very basic questions like:

  • What is the database size?         

  • What is the make-up of the database?

  • What are the different sources of acquisition or database growth?

  • What are the different sources of loss?

  • What are the different customer types?

These questions help build a solid foundation and will allow you to gain an understanding of your key numbers.

How do we maximise the insight we generate from our information?

Before you can make any changes, you need to know what to change - sounds obvious right! But it is this stage that will allow you to focus your analytics on the most appropriate areas, in turn helping derive the most benefit for your business. You are in essence trying to do a post mortem on your data! Again going back to our initial problem of changes in database growth, our questions would usually begin with ‘how’:

  • How has my database size changed over time?

  • How have volumes changed across acquisition channels?

  • How have volumes changed across loss sources?

  • How many individuals are engaged with the brand?

  • How many individuals are disengaged with the brand?

  • How do various factors influence a customer’s journey with the brand?

These questions allow you to determine which factors are potentially important in building brand loyalty. It is the process of asking these questions that will allow you to identify where the loss in growth is actually stemming from.

Now I have this information, what  future analytics do I focus on?

Now you have identified the variables that have the biggest impact on your customers, you can start to understand how variable changes impact behavioural changes. You can also put variables together to see how a combination of factors influences customers. Looking at our database growth problem again, if the decline was due to lapsing and unsubscribed customers, we would seek to understand:

  • What type of customer is most likely to lapse or unsubscribe?

  • What are the factors that indicate a customer is more likely to lapse or unsubscribe?

  • Are there any forms of communication that make an individual unsubscribe?

  • Is more contact likely to make an individual unsubscribe?

  • What is a customer’s behaviour prior to lapsing or unsubscribing?

However, if the decline in growth were due to a slowdown in customer acquisition, we would want to understand:

  • Across which channels is customer acquisition slowing down?

  • What are the historic customer types from each channel?

  • What is the value of customers from different channels?

  • How has engagement across channels changed over time?

  • How has activity across acquisition channels changed over time?

  • What factors improve acquisition across different channels?

So now you know the questions to ask. Let’s explore how you find the answers?

There is a close link between the question you are trying to answer and the analysis technique chosen. So let’s now focus on the techniques you could use to analyse and derive insights from your data.

Setting the baseline

Reporting attached to your database will allow you to answer all your preliminary questions around size and make-up of your database. Campaign performance reports will then allow you to understand how many customers were acquired through your different channels and how many customers unsubscribed or disengaged with your brand. A simple database count of number of customers opted in will also give you an indication of the movement in your marketable base.

To generate further insight, simple segmentation applied on top of reporting gives a better understanding of your customers. Having a look at a basic variable like location would help you visually identify how your database is changing based on customer location.

 A data post mortem

The techniques and tools used to understand historical data depend a lot on the questions you are trying to answer. Reporting which shows trends over time would allow you to understand aspects like changes in volumes and channels. This would also give you visibility of how growth has changed over time.

Key techniques you could implement:

  • Pulling in multiple factors will help answer questions around customer type and engagement

  • RFM and engagement segments would allow you to classify customers that are either engaged or disengaged and also understand the value of customers within your database

  • Cluster analysis would help you identify variables that help break down your database into actionable customer segments

  • Further analysis into content and offers will help you to understand what factors influence a customer the most when making a purchase

  • In the context of our problem, a time series analysis would help understand if all calculations were done at the same time and we were comparing like for like. If historically there has always been a period when a large number of customers lapsed, this would lead to a new question stream

Driving action from insights

Overlaying two or more analysis techniques would allow you to look at how different variables influence customer behaviour. For example, overlaying frequency of communication with engagement and RFM segments would help identify customers who are lapsing and potential issues with communications that are resulting in them disengaging. A predictive model could help identify whether a customer is likely to lapse. A CHAID model or decision tree could help identify points at which you need to intervene to prevent a customer from lapsing. In the database growth scenario, if the problem is with acquisition, an attribution model or cohorts analysis would help identify the best prospects.

All you need are the right questions     

Dr Jonas Salk said, ‘The answer to any problem pre-exists. We need to ask the right question to reveal the right answer.’ The same applies to data and analytics.

Share:

RedEyeUK @RedEye UK