By Davide Russo, Courtney Bradford - March 9, 2021
We recently launched a project that’s been months in the making—a completely redesigned Sisu experience.
Our vision for Sisu is to be a decision-making hub for our users—whether they’re technical analysts, growth marketers, or product managers. To accomplish this, we needed to reimagine the end-to-end product experience to be simple and intuitive. Read on for an in-depth view of how we built this new experience and some key takeaways.
Sisu started out of Peter’s research at Stanford’s DAWN Lab where he focused on bridging the gap between having access to data and using it to make decisions. In the first year, the team focused on productizing and scaling the underlying analytics engine.
As with many products that have early traction, the team added features organically in response to user feedback. This approach resulted in many useful features but increased product debt. We heard from users that while the “engine” was incredible, the “car” was becoming hard to drive. With this feedback, we knew we had to take a new approach to the design and experience of Sisu.
For a Product Manager or a Designer, this is one of the most exciting situations to be in. We love working on products that deliver value but where there’s so much potential for impact.
At Sisu, we use some of the core Design Thinking principles, including an adaptation of the “Double Diamond” below. (You can read more about the original framework here).
Likewise, the product team grounds every design decision in user pain points. For this project, that meant first understanding the mental models that users have around making decisions with data. Then we dug into how they think about metrics, how they get answers from their data, and how they make decisions with data. And at every step, we tried to understand their underlying assumptions, pain points, and desires.
The output of this research helped us define a shared definition of (1) who we are building for and (2) what problems we want to solve, and enabled cross-functional teams to align their strategies and priorities towards a shared vision.
With this shared understanding and focus, it was time to get to work. We assembled a team of PMs, designers, and engineers to explore critical user journeys and develop low-fi UX. Ignoring everything that existed in the product, we thought about this problem from a first principles approach. Then, we focused on what steps it takes to make decisions with data and what the ideal experience should be. After multiple iterations, we converged on an information architecture and UX that felt intuitive and easy to use. For example, we moved from grouping analyses by type (e.g., time comparison vs. group comparison) to grouping them by metrics (e.g., LTV, Average Order Value) and built a new prototype to use in another round of user research.
You’re probably picking up on a theme here. We’re obsessed with making sure we’re building the right features, and we put things in front of users early and often. This round of UXR focused on usability and accessibility. You know that you’re on the right path when users look at something, and they get it right away. (Pro tip: If something you’re working on needs explanation, it’s a sign you need to continue iterating.) Users loved the prototype and shared great suggestions for improvements. This gave us the confidence we needed to start dev work.
A major product overhaul like this is a heavy engineering undertaking. The necessary information re-architecture made it impractical to ship small incremental features and instead required one big rewrite and release. For example, analyses were previously stand-alone pieces of information, and users often had to type the same SQL query multiple times. In this new release, we introduced the ability to create and maintain a common definition for metrics across the company. In addition, we changed analyses to be specific to a metric rather than stand-alone, which broke every paradigm we had.
To help users validate data quality, we decided to surface stats like min, max, and median with each analysis. This change meant users no longer had to switch between SQL and Python. We also added the ability to preview the underlying data table to make spot checks easier, requiring new endpoints and tight collaboration across platform and product engineering.
One of the trade-offs we made was prioritizing information architecture and core functionality over the look and feel. For example, we debated at length where users would customize their analyses and what those levers should be. We wanted these levers to feel contextual to the data insights to allow for easier iteration without the need for SQL. However, we spent very little time debating what those controls looked like.
Given our timeline, this was the right trade-off to unblock the development of core features but required revisiting the look and feel mid-project. While that isn’t ideal, it gave us more time to think through the design principles we’d use to ground our new look and feel.
With this time, we set down three guiding rules:
1. Optimize for what’s important to focus a user’s attention
Hick’s Law explains that the time it takes to make a decision increases with the number and complexity of choices. In analytics, being impactful relies on how quickly and efficiently you can find the insights that matter in your data.
While we couldn’t decrease the number of choices within Sisu without making a task more difficult or impossible for our users, we could reduce the complexity by gracefully sequencing information.
To do so, we had to start by creating an information hierarchy that arranges the elements in a way that intuitively reveals their importance. First, we grouped obviously related components and information together. Then we reduced visual noise by obscuring elements that are needed less often in an accessible, memorable way.
Users told us that browsing through hundreds of facts could feel overwhelming, so we introduced three features to help with that: fact grouping, fact drill-downs, and new filters on par with industry standards. These features would provide users multiple ways to zero in on the data and iterate efficiently to get to actionable takeaways.
2. Acknowledge user actions with quick feedback and make it easy to recover from mistakes
When it comes to data analysis, there’s a certain level of complexity that can’t be simplified. The interface needs to acknowledge that complexity and proactively guide users to achieve their goals as quickly as possible.
To aid users in setting up their analyses, we added:
To put users at ease while they work, we also added:
3. Good design is intuitive and aesthetically pleasing
Finally, we believe good design creates an intuitive experience, especially when the job to be done is complicated. According to the Aesthetic-Usability Effect, users perceive aesthetically pleasing design as more usable. With this redesign, we looked for ways to make the UI “easy on the eyes” for analysts that spend considerable time in front of a screen. We intentionally minimized the use of color by leaning on a cool gray palette to emulate depth. We used white space to improve readability, softened the components, and leveraged subtle visual indicators for actions. For us, “make it pretty” wasn’t an outcome. It was a key part of the design process that directly impacted the user experience.
We outlined success metrics for this launch and worked closely with our Marketing, Sales, and Sales Engineering partners for over a month to bring this to market. Since the launch, we’ve started collecting user feedback to inform our next iterations and fast follows.
This launch is just the beginning. We can’t wait to share all the exciting changes coming over the next months and have fun along the way.