• Martti Sutinen

Design for Data Scientists: Searching for Purpose

Or 3 Lessons from (Not) Applying Design

What is design and why should I love design as a data scientist? These were the questions I wanted to answer when I first got service designers as colleagues. Let me start by answering them right away, then I’ll share experiences from projects. And if you are a designer, read ahead, I’ll also describe data science shortly

As a data scientist, I should care about design because it helps me make customers happy by putting focus on people. Design gives a solid foundation for the problem context and goal setting, resulting in engaging data-driven solutions. It gives tools for communicating stories with customers. If data science is a team sport, then the winning team most likely includes both expert designers and data scientists, as well as engineers.

What makes me say that data scientists should love design and users?

I have been working as a data scientist for several years in an enterprise with more than 10k employees and I currently work in a company with fewer than 50 experts. I have worked with customers in manufacturing, forestry, energy, media, and healthcare. During all this time, I have never had to regret working with industrial or service designers. But I have regretted the several times I have not had that opportunity.

Data Science in a Nutshell

Data science is an approach to solving complex problems for people by using data (anything that is or can be recorded) and by borrowing methods from mathematics, statistics, engineering, and computer science. I dare say that it borrows from consulting, design, and janitoring as well. Imagine inspirational storyteller meets curious math nerd meets hardworking developer.

We are not always actual scientists. I think the science part comes from approaching problems with an open and critical mind. Science inspires us to work in an iterative manner, creating hypotheses and testing them with real world observations. We must understand limitations of the used data and our processing of the data, the models we apply, and the conclusions and insights we draw.

Lesson 1: Data Analysis is Pointless Without Known Context and Needs

Successes with design are easier to come by and spot than successes with data science lacking design effort. Luckily it is never too late for design to save the day, as I’ll show below.

I once worked on an Internet of Things project where we wanted to analyze movement in office spaces. The start of the project was a minimum viable product with a very strong technology focus. Sensors were installed, a dedicated network put into place, and the operation was monitored. However, exploratory analysis showed that the recorded data did not provide very informative content for further analysis. The sensors had been changed and moved around without notice or record, causing the data to be inconsistent. I then wanted to know what the original pain points to solve were. It turned out that there were none; just a hypothesis that office analytics would be interesting and the sensor data would provide some insights.

My literal reaction. By mintchipdesigns from Pixabay.

Luckily our analysis visualizations looked awful, so an industrial designer joined the project. She conducted interviews around the office and worked together with us data scientists to understand what could be done with the data to benefit users. She provided mockups, and soon we found we had a clear concept to work on. The data collection was improved, and we could test a movement data visualization on end users.

Lesson 2: Do Not Specify the Solutions in Advance, Just the Needs, Goals, and Constraints

In another case, a client ordered a customer segmentation analysis, which seemed straightforward to me. Some external geo-information data was combined with the customer data and several clustering algorithms were tested. The clustering provided interesting results but I did not expect that the new segments would be all that valuable for the intended marketing purposes.

I decided to try out a dimension reduction algorithm in order to map a large number of customer metrics into just a couple of main, aggregate factors. I put together a dashboard for identifying customers with the factors relating to volume and activeness. This reduced the number of metrics to consider in ranking from a hundred or so to only two.

The client preferred the new approach and started testing it out internally. In data science, as in design, we should approach problems with an open mind, prototyping alternative solutions and testing their fit on the defined problem context. The uncertainty regarding solutions should be recognized in project plans and contracts.

Driven by Purpose

Alright, if design is so powerful, how can we work together? What is it that a designer does that I can use in my work as a data scientist? What do designers even deliver besides ideas?

Good news. It is easy for us to work together because we share surprisingly many attributes. If you read the descriptions of design and data science above, you might have noticed that research and an iterative way of working are key parts in both of them. Both design and data science are about sensemaking, with empathy and data. But the most important similarity is purpose. We find a problem that people have and we are driven by the purpose to understand and solve that problem, without knowing the answers or even the challenges in advance.

Design Deliverables and Process

Purpose, as mentioned above, is communicated with design deliverables. The business model canvas and value proposition are summaries of how the offering and its business model creates value competitively for both the user and the business. The customer journey map, consisting of narratives, and the service blueprint, with inputs, processes, and outputs, visualize the service flow and experience from the customer’s perspective. Sketches, wireframes, and prototypes visualize the intended interfaces. The deliverables are often presented on canvases.

Workshops are a great way to collect data on needs. By Skitterphoto from Pixabay.

The features roadmap describes how problems are prioritized in the implementation, based on cost benefit or impact vs. difficulty analysis. Personas, use cases, and user story maps show who the users are and what pain points they are trying to solve, and they are useful in managing agile projects and tasks during the implementation. Taxonomies depict how content and information is organized.

Design is a research driven creative process. The double diamond shown below shows how design work alternates between divergent and convergent steps, or creating and testing hypotheses with data. The data is collected from user and stakeholder interviews as well as metrics collected from the processes and interfaces. Research briefs describe the goals and methods of research, as well as the design brief describing the goals and context of the whole design work.

The Double Diamond. By Olga Carreras Montoto from Wikimedia Commons.

Three Ways to Combine Design and Data Science

Data science fits into design frameworks very well, in three ways. First, data science informs design decisions through data collection and analysis on user flow and experience. Secondly, data science expands design possibilities by introducing new ways of interacting with users as well as automating previously manual processes such as claims handling.

Finally, data science benefits from the purpose and the problem definitions formed in the design process. The starting point is not the currently usable data but the validated business and user goals, needs, and constraints. Data science and design work should not be isolated, as they influence each other. Working on the same team with daily communication makes sense and I recommend it.

Data science and design work should not be isolated, as they influence each other.

Lesson 3: Data Science is About Solving Challenges of Trust, Excitement, and Growth

In a case I’m currently working on, we are solving customer churn and retention with a data-driven predictive and diagnostic approach. The problem concerns a variety of stakeholders; customers and employees from sales to management. The minimum viable product was about showing business decision makers that the technology has potential, with little focus or investment placed on design.

Now we are proceeding to productization, and the customer has realized the importance of bringing a user experience team onboard. Experience shows that for business users, it is not about providing just a model, but about bringing visibility and openness to data sources and processing. When we produce new insights with analysis, we are also transforming existing processes. This makes managing the change and experiences vital. We’ll continue to use design where it excels; that is, nurturing the positive relationship with users and achieving real business impact.

If you liked this article, please recommend or share it, and I will continue writing about similar topics! How do you feel about combining design and data science? Discuss in comments below.

At Valuemotive, we offer teams of designers, data scientists, and developers to solve challenges together with our customers. We also offer workshops and training to get started with a data-driven culture and artificial intelligence.