How To Design
With Warm Data
Much is currently being written about data as a new tool in the designer’s toolkit. Just as the virtues of data are being shared and sold in business, creatives are starting to wonder what role data might play in their process.
In the last 12 months, we’ve been bringing data into our design process and learning a lot about how to benefit from it. Like any tool, it isn’t an answer in and of itself, but there’s a lot to gain from data as a designer.
Warm data is a nascent term, the steps below are how we’re starting to understand it. The five areas in this article are things to think about when building data into your design process. They are by no means exhaustive, but as a starting point, they might help focus your energy and get you to valuable results sooner.
Data Driven Design
As a creative, the thought of data playing a role in my process initially seemed counter intuitive: how would I resolve the (apparently) competing methodologies? Did I want my designs to be rationalised down to a series of A/B tests?
Whatever my perceptions about data were they were shaped by the way other people talked about it. One thing that becomes abundantly clear when you enter this world: no one really knows what they’re talking about. To paraphrase Ray Bradbury:
Don’t let them flim flam you into following their advice.
As designers we need to define our own terms for the way we want to use data.
1. Hypothesise!
While I don’t think we should make our process look more scientific than it is, there are elements of the scientific approach that do help to focus your energy on the right part of the problem. When you enter the world of data, statics and analytics, it’s easy to get lost in details and dive down endless rabbit holes. Defining a hypothesis forces you to make a statement of fact that your data should either prove or disprove. Equally, a hypothesis will tell you what to measure, the metrics and amount of change.
However, the simple suggestion of ‘start with a hypothesis’ masks a difficult task. Coming up with a good hypothesis can take a lot of work and discussion; discussion that, at times, may feel like an endless round of opinion and counter-opinion. Debate here is important, so make space for it. It will become less necessary over time.
But, if you get stuck in a loop trying to figure out a good hypothesis, the act of doing will answer most of your questions: much like prototyping (something I’ve written about here) just get started.
2. Visualise!
The image above is of four different data sets with the same average, mean and mode. As raw data, these might seem broadly similar and it’s only through visualising that patterns emerge. The graphs are known as Anscome’s Quartet and emphasise the importance of moving beyond the numeric.
I’m essentially a visual designer and so displaying data in a graph or chart is often the only way I can truly understand it. However, as the graphs above show, visualising can also reveal hidden trends and forces. We’re also finding that charts are the best way to disseminate information between a group, as they act as Boundary Objects — helping different people see different things whilst uniting the group.
It’s important here to draw a distinction between visualising data and the trend of infographics. Infographics aim to summarise a mass of data from different sources into something digestible, illustrative and engaging. But that’s not what we’re aiming to do when visualising data. The goal here is to step beyond raw data.
Choosing the right format of graph or chart is important — and not always obvious.
Understanding the right chart type to use is a post in itself, but the best thing you can do is to experiment with multiple visualisation types. You may find that the most appropriate format only becomes clear when you try it.
As much as a complex or unusual format might seduce you, I’m afraid it’s often the basic formats of bar chart, scatter graph and pie chart that will get you the best results. Tools like Raw by density design and the d3.js library are fantastic for building complex and interactive visuals, but I spend more time using Excel and Google Spreadsheets chart tools.
3. Synthesise!
As a Human Centred Designer, interviews, discussion and co-creation sessions are core parts of my process. Making sense of the things we find is the most crucial part of any research and so regular synthesis sessions help us digest and summarise our research.
The same goes for data. Taking the time to document your process and findings is more important than a ‘good result’. Your documentation and post reflection will tell you more about the success of the test, than the goals you set. Review the method, the results and the unexpected findings.
Working with data will pull you into a cycle of rapid iteration as experiments raise more questions than they answer (a good thing). Synthesising will help you avoid drowning in data. At times it might feel like you’re not extracting enough detail. Remember that five iterations are better than five days of sifting through the results of one test.
While summarising the findings of an experiment, keep an eye out for ancillary information. Most analytic tools will give you access to a wide range of results, while you may be focused on one or two key metrics. Don’t forget to glance around the other data points captured. One recent experiment, using Adwords, to compare the preference between two value propositions, accidentally showed us the top ten US cities that were searching for our future product.
4. Collect it yourself
It will become more common to kick off a project by analysing the existing data streams in a clients’ systems. Some have data because it’s part of the way they work today, others might have a progressive view of data and be collecting it already. But more often than not you may need to collect it yourself.
Even if there is data out there, it might be difficult to get to, or listening for other things — not aligned with your needs.
The chances of useful data being available to design with is probably low. Fortunately, the tools to help you here are widely available and easy to use. Google Analytics and Crazy Egg will help you track website traffic. Other services, like Intercom, are geared up for tracking customer engagement at a deep level, and just this month Twitter has released a public dashboard to track your tweets.
We’ve also had success building our own tracking tools. Cheap prototyping hardware like Arduino and Raspberry Pi can make great data loggers, and cheap enough to leave in the field for weeks at a time.
Equally, off-the-shelf consumer tracking products can be a quick route to data capture. Look out for products with developer toolkits and APIs that will allow you to extract raw data.
You might also find value in public datasets: in the UK data.gov.uk and data.gov in the US both give you access to masses of big datasets. Be warned that just because the data is here doesn’t mean it’ll be easy to work with. You might need specialist software to play around with the data you find. (There are new skills to be added to the designer’s toolkit here).
5. Quantitative and Qualitative. In Parallel
At IDEO the emerging skill-set around data driven design isn’t a replacement for the existing process. Human centred qualitative design will always be our core, talking to real people and building empathy for those we design for is the key to appropriate and impactful design. What we’re finding is that the qualitative and the quantitative work harmoniously: where data can give us patterns, true behaviour and precise results, interviews with participants, reveal true attitudes and perceptions. The power comes when data directs us toward specific areas for investigation, and speaking to people reveals the nuance.
Data analysis is pretty bad at narrative and emergent thinking, and it cannot match the explanatory suppleness of even a mediocre novel.
It can be easy for data to give apparently clear results, but it’s only in asking people directly that you find the truth. Make sure you test important hypotheses in multiple ways, triangulating answers from different experiments. The power of multiple data points all pointing to the same conclusion should give you confidence that you’re on the right track.
I’ll leave the final note to Richard Fenyman, a scientist who came to the same conclusion.
I must first understand more or less how the answer probably looks. It’s hard to explain this very well, but I had to get a qualitative idea of how the phenomenon works before I could get a good quantitative idea.
Tell me what you think.
