Last August, a tweet from a data scientist reminding colleagues to get out of tables of data and talk to real people — you know, do qualitative research — went viral among these two constituencies. The resulting debate reflected the state of the relationship between data science and design: They’re not on the same page. One participant even deleted one of his tweets, overwhelmed by the volume and combative tone of the feedback.
The worlds of data science and design are indeed colliding. More and more companies want to transform experiences with machine learning. The best organizations rely on customer analytics to make decisions. At the same time, the size of design teams has exploded, as has the influence of design thinking — design’s cousin discipline for the masses.
These groups aren’t working together effectively. Data scientists overlook the value of design, and designers don’t get what’s possible with data and machine learning. At the intersection: a counterproductive divide between data scientists and designers. They don’t see the world the same way, and they don’t use the same methods. But instead of finding common ground, many retreat to what they’re comfortable with.
We went looking for a better way forward by researching the right design methods for what Forrester calls a “data-fueled product” — a digital product that recognizes patterns and anomalies relevant to a user’s goals in large quantities of data and adjusts parts of its user interface in response.
Unlike traditional digital products that just provide information and allow users to complete tasks but don’t set out specifically to help, data-fueled products bring the best of design and data science together. Seeing suggestions about the next song you’re likely to enjoy? That’s a data-fueled product. Seeing best-fit doctors based on symptoms and location? That’s a data-fueled product. Showing a support rep the relevant knowledge-base articles based on the caller’s concerns? A data-fueled product.
Some of the best practices our research revealed:
1) Construct a team that comes from different backgrounds and life experiences — then build a culture that can openly discuss potential bias and misuse.
2) Understand concepts like the algorithmic imaginary (how users think of what the algorithm does) and what’s assistive versus agentive (how much the technology does on the users’ behalf).
3) Sketching — yes, sketching. Whether teams sketched the model, the interface, or the experience, as soon as colleagues began to visualize projects and products, everything went better.
For more examples and guidance about these challenges, see my new report: “Data-Fueled Products: How To Thrive On The Design And Data Science Collision.”
As always, if you’re working on these challenges or have questions, get in touch — I’d love to hear from you.