Dividing audiences like Donald Trump

Related Articles


Google Consent Mode V2: No longer just an optional extra


Austrian data protection authority finds use of Google Analytics an unlawful transfer of personal data


Dividing audiences like Donald Trump

My talk today is going to be about dividing audiences like Donald Trump. It’s a nice catchy headline. Really what we’re talking about is segmentation. Today I want to give you two really tangible, simple tricks that you can use to segment audiences and to segment some other dimensions that we don’t often talk about, maybe products, transactions and so on.

I want to provide you a framework for doing that nice and easily. We already do lots of great things for segmentation, so within paid media we do lots of analysis to understand how we should be adjusting our bids. Within the content teams we’re doing lots of analysis into personas. All of this is just different types of segmentation. I wanted to try and generalise a framework for segmenting dimensions and metrics from Google Analytics and demonstrate it to you in a couple of really simple ways.

My first little technique is for uncovering opportunities for any metric. The first thing you need to do is pick your metric and for an e-commerce client we might typically look at revenue, average order value or the average basket size, straightforward stuff. We’re using Google Analytics data for this example, but it can apply to anything. Hopefully with Google Analytics it’s very familiar to all of you, you should be able to follow along without a problem.

The next step is to build your questions. The kinds of things that we want to know about our audiences and the people purchasing are who, where, when, what and how (there is a why, but you need some qualitative data for that). In Google Analytics, we have access to all these different dimensions, age and gender, city and country, that can help us answer these who, where, when, what and how type questions. I suggest once you got your metric, you look at these questions and select your dimensions as well.

Then we can get the data for these using Query Explorer or Custom Reports, we can very easily put them in. You can see here we’ve got hour, day of week, and revenue, you just drop them in and then export the reports as you normally would and bring it into Excel, or you can bring it into something like Power BI.

Once we are at that stage, you can very easily create a grid. All it is, is one dimension down this side and one dimension across the top and then all the values in the middle here. So this is the hour of the day from midnight through to 11pm. Then Monday to Sunday across the top. Then this is revenue. You can see really clearly when we start to heatmap that, we can use conditional formatting in Excel to visualise it and we can create a heatmap of when customers are spending money, when they’re actually converting. Already you can begin to see some really clear segments jump out.

I pulled out five, early morning Wednesday to Friday AM, 8AM-5PM on weekdays, Sunday mornings, early week evenings, and weekend afternoons. Once you’ve got those segments, it’s really easy to start getting ideas, getting creative, thinking about how you can market to these people, what might these people be doing at this time. It opens up opportunities for further exploration.

My second trick is for discovering powerful audience segments. So far we’ve looked at segmenting a metric and understanding how a metric can be segmented in different ways. What I would like to talk about now, is how we can build audiences, so this might be individual users, but we can also apply the same technique to things like products if we’re trying to understand a product catalogue, for example.

In this example, you pick two metrics and one dimension, as opposed to one metric and two dimensions. You do the same thing as before, you put the dimensions into a custom report in Google Analytics, or we can use the Query Explorer to get the data, exactly the same thing. You can plot it on a scatterplot and it looks horrible and nasty and it’s really not very easy to analyse at all. But if you find the average (or median) of each, here we’ve got revenue and we’ve got average basket size – the quantity of products purchased and transaction. We’ve got and average revenue per transaction of £78.28 and an average basket size of 29. We can plot these averages on the chart and it creates quadrants for you.

Then we can very easily begin to analyse these and have a think about what these quadrants are showing us. We can come up with different ideas as to how these different audiences might be best reached. What I was thinking when I saw this, is in the top-right you’ve got people spending a lot of money, they’re buying a lot of products at once. It could potentially be B2B customers. In the top-left you’ve got people buying a lot of items but not spending a lot of money, so it might be people looking for a good deal or a big bulk discount. So that could be really good for, say the client wants to make some room in the warehouse and clear some old stock, send these guys an e-mail saying we’ve got a big discount on these, why don’t you come and buy them? In the bottom-right you have the high value product buyers, people buying really high ticket items but not very many of them.

In summary, I want to enable you guys to be working with segments in your day-to-day work and to use this framework to build some really powerful segments that will help you out. Very, very easy to do. Pick a metric, pick a dimension, one of the two depending on what you’re looking to explore. Pick two accompanying metrics or dimensions, the opposite of whatever you’ve already chosen. Get the data with custom reports and with the Google Analytics Query Explorer. Visualise and analyse them appropriately as heatmaps or scatterplots and then do analysis to create segments and come up with the ideas for how these people can best be reached. Finally, act on the data. Once you’ve got these segments and have come up with a hypothesis, test it, if it works, implement it.