Best practices for combining data science and design
By Ricky Hennessy, Sheetal D Raina & Amanda Ward
At Fjord, we’ve been combining data science and design on integrated teams for the past three years. By bringing together these two disparate disciplines, we’ve been able to deliver tremendous value to our clients through data-driven products and services that address user needs and solve critical business problems. Despite the enormous potential, most organizations struggle to enable effective collaboration between data science and design.
In this article, we’ll share best practices and lessons learned on how data scientists and designers can work together to deliver transformative products and services.
Create a Shared Understanding
With different backgrounds and ways of seeing the world, it’s important to create a shared understanding between data scientists and designers so we can get the most out of this powerful collaboration. Instead of parallel tracks working in isolation, take advantage of opportunities for knowledge sharing and question asking. By remaining flexible and open to different approaches, data scientists can benefit from a deep understanding of user needs and designers can unlock possibilities presented by large sets of data.
When working with one of our aerospace clients, our data science team was performing exploratory data analysis while our design team was interviewing users. By co-locating the team and providing opportunities for spontaneous check-ins, the design team was able to validate research findings using real data, and the data science team was able to gain insight into the peculiarities contained in the data. This led to a much deeper understanding of both the end users and the data.
Don’t Lose Sight of Business Goals
Data science is a process for solving business problems using data. Before we can begin the process, it’s critical that we have a clear understanding of the business. Too often, the design process leads to an overly user-centric view that ignores real-world realities and constraints. On the other hand, data scientists can become so focused on solving technical problems that they lose sight of core issues that need to be addressed. For any data science and design collaboration to be successful, it’s critical to align user needs with business goals. Ensuring that the team is equipped with business understanding will lead to solutions that deliver value while also addressing user needs.
While working with a Fortune 100 manufacturing company, it became apparent that the initial problem at hand was too broad. If we had focused exclusively on what users were looking for or what project stakeholders wanted, we would have developed solutions that delivered little to no value for the organization. Instead, we viewed user needs and stakeholder input through a “business goals” lens, allowing us to narrow our focus to high value problems.
Execute Through Seamless Collaboration
Defining the right problem starts with the utilization of a “business goals” mindset. However, defining the problem is just the beginning. Individual understanding leads nowhere unless there is effective collaboration between team members. At every stage, integrating various viewpoints with different skills and expertise helps the team solve the right business problem in unison.
At Fjord, we have developed a framework that translates business problems into solvable data science problems using a human-centered design approach. The key lies in integrating strategy, design, and data science in a way that they are more potent than when they work separately.
Incorporate Data Science Into the Design Process
At Fjord, we find opportunities to supplement the design process with data science techniques. During user research, we analyze data that contains a wealth of quantified user behavior. Through analytical techniques like clustering, we use this data to form data-driven mindsets. We then combine the results with outputs from qualitative research to form detailed and holistic views of users. During interface design and optimization, we utilize clickstream data to inform design decisions and create personalized experiences. Including data science in the prototyping process provides an entirely new medium for designers: machine learning. Experiences can now predict user needs and adapt to changing environments. During product delivery, data scientists can create a data collection strategy for future predictive modeling and automated product improvements.
Account for Data Access Issues
Oftentimes, the data we need is inaccessible. It could be that the data doesn’t exist, that there are security issues preventing access, or that the team controlling the data is unwilling to give access. Before kicking off a project, it’s critical to identify data requirements, determine accessibility constraints, and develop a “no-data” contingency plan. Here the most likely scenarios to consider:
- Relevant data has been identified and access has been granted by day one of the project. This is the ideal situation and is possible with careful planning; however, it’s highly unlikely and never a good idea to assume this will be the case. For every project, depending on data complexity, it’s best practice to plan for 1–4 weeks to gain data access.
- Relevant data has been identified, but technical problems are preventing access. The best practice in these situations is to plan for 1–4 weeks of time where there will be no data access. Additionally, there should be clear understanding about any data-dependencies and assumptions that will not be met if data access is not granted in a timely manner. This is also when a “no-data” backup plan is critical. One common way to deal with a lack of data is to create scripts that generate synthetic data. This will allow design and development teams to continue moving forward while data access issues are sorted out.
- No relevant data exists. This is a great opportunity for data scientists to be involved up front in forming a data strategy. Usually, we are constrained by data models and systems that don’t account for the end user experience. At Fjord, we’re able to use data science to design with data in mind and ensure we’re collecting the right data that’s needed for experiences that learn and improve over time.
Embrace Technical Constraints to Deliver Value and Gain Trust
When it comes to data, most organizations have made large investments in technology that place constraints on what is feasible. Whether it’s limited access to relevant data or underpowered systems, understanding and embracing these constraints ensures our work leads to implementable solutions and true organizational impact. Of course, this shouldn’t stop us from making recommendations that would lead to improved experiences, but by delivering value within existing constraints, we gain the authority and trust to become partners in long-term transformations.
While working with a state run healthcare exchange, we developed a machine learning model that predicts the likelihood of a citizen going uninsured. By working within the existing technical constraints, we were able to quickly turn this model into a tool for the outreach team to prioritize effort, which ultimately led to less uninsured citizens.
Facilitate Cross Discipline Communication
Communicating analytical results or making machine learning models explainable to stakeholders is critical. However, it’s equally important to have seamless internal communication on integrated data science and design teams. If you don’t understand each other’s craft or speak the same language, you will always be at a disadvantage.
Avoiding technical jargon and creating a glossary where technical terms gets translated into simple, understandable language helps bridge the gap between the two disciplines. Understanding each other’s line of work through a universal medium enables collaboration, and therefore collectively, allows the team of data scientists and designers to articulate business problems and translate algorithms and solutions into compelling stories.
Plan for Experimentation
This may come as no surprise, but the data science process borrows heavily from the scientific method, which is an iterative process involving the development of hypotheses and experiments to test those hypotheses. This differs from projects where outcomes are known, and the process is more deterministic — we need to plan accordingly. This means giving data scientists time to perform exploratory data analysis, which allows them to understand how the data is structured, what it contains, and ultimately how it can be used. This leads to a set of hypotheses on how data can be used to enable specific opportunities or solve important business problems. Analytical prototypes and POCs can then be developed to test hypotheses and determine feasibility.
Standardize Data Deliverables
As data scientists working inside a design agency, we’re often working alongside colleagues that have never worked with data scientists. This is why it’s import to develop consistent and standardized deliverables so that designers can gain familiarity with data content and have a better understanding of what we do.
Consider two perspectives:
- Data deliverables. We are developing standardized Notebook, PowerPoint and Word templates for exploratory data analysis, modeling, and communication. We are creating reusable libraries for Python, R and D3 to integrate our work with prototypes. We are building templates for conceptual data models, data strategy, and data architecture for common AI techniques such as recommendation engines, chatbots, machine learning models, predictive analytics, and deep learning.
- Design deliverables. We are modifying the company wide guidelines to include data annotations, information models ,and interactive data visualizations in all the standard design deliverables and tools. This will help ensure that data science outputs can easily be integrated with the broader team deliverables. Data scientists working with design should understand the design tools, should know how to incorporate data and data science content into the trendiest design tools your organization uses. Integrating deliverables means a seamless experience for the entire team.
Data is a core component of nearly every new product and service, and if we want to effectively design for this new world, we need a new form of collaboration — one that combines data science and design into a tightly integrated team. Despite the challenges of bringing these two very different disciplines together, at Fjord, we believe it’s the only way to deliver data-driven products and services that create value and delight users.
Ricky Hennessy leads Data + Design for Fjord US Central. He has a background in data science, design, and engineering. As a practical academic, he likes to apply state-of-the-art data science techniques to real-world business problems. He has a Ph.D. in Biomedical Engineering from The University of Texas at Austin, over a decade of experience in data science, and has spent the last four years working in design.
Sheetal D Raina has expertise in Data Science, Machine Learning, Computer Vision, Business Intelligence and Data Strategy. Sheetal brings over 20 years of experience in the technology industry. In addition to her expertise and love for solving business problems, she brings a broad spectrum of domain expertise augmented with a strategic sense of business and creative thinking.
Amanda Ward has expertise in data science, business analysis, solution design and consulting. She has her Master’s in Data Science from Monash University in Australia and 10 years’ experience as a Business Analyst / Product Manager on data-based projects. Amanda is new to the USA, and her background is in Australian telecommunications, government and education industries.
