Smart Discovery 2.0
The feature uses machine learning models to provide business insights and build automated dashboards.
I joined the Predictive Analytics team in late 2017 as the primary UX and Visual designer. Smart Discovery 2.0 was announced at Sapphire 2018 as one of SAP’s exciting new innovations in Analytics Cloud.
|Product Expert||2017 - 2018|
What is Smart Discovery 2.0?
Smart Discovery is a feature in the SAP Analytics Cloud product. It lives in the ‘Story’ space, where designers and business analysts build visualizations and reports through data models and traditional data discovery.
Smart Discovery uses regression and classification predictive models to analyze customer datasets and find potential insights.
Usually it takes designers and business analysts days of data entry and manual manipulation to discover and then report on the insights that inform business decisions.
How might we help business users quickly find and understand key influencers, outliers, trends, and ask “what if” questions when building a smart report?
To make the data discovery process intuitive, scalable and visually expressive.
- Enable People To Share & Manipulate Findings: The previous Smart Discovery dashboard had read-only limitations. My task was to design a workflow that allowed people to edit or re-purpose the generated insights.
- Create a story-driven user journey: Smart Discovery needed to be more intuitive for users with any level of expertise.
- Provide Insight Through Improved Visualizations: Smart Discovery identifies key influencers by looking at the record level data, but the feature needed to show data in a visualization that business users could easily understand.
As the primary UX Designer of Smart Discovery & Smart Insights, I was responsible for the User Research, UX Design, and Visual Design for both features.
- User testing & interviews to inform customer journey maps to find innovation opportunities during a typical data discovery workflow.
- Wireframing & interaction patterns to guide users through the discovery preparation: selecting a target variable, navigating data size limitations, and overall improved error handling.
- Interaction prototypes in low fidelity (paper/Keynote) and high fidelity (Principle/Framer), which explored different ways people can discover and better understand key influencers.
- UX & Visual mockup specifications in Zeplin and provided further design support to development when technical limitations came up.
- Collaborate closely with Engineering leads and Product Experts when planning the feature roadmap for improvements and advocating for user needs.
Ask all the questions
After asking Product Experts and Engineers a ton of questions, I started evaluating Smart Discovery 1.0 for potential ‘UX quick wins’ that I could rapidly prototype and validate.
Here are some of the questions I started with:
How might we visualize large amounts of information in a way that’s easier to digest?
How might predictive findings be communicated in a way that is easier to understand using natural language processing?
How might color be used to communicate ranking, correlation, or clustering?
One of the biggest usability issues I wanted to address was how people interpret the key influencers.
I wanted to explore how to clearly instruct & communicate the impact of each influencer through:
Side by side comparisons
Progressively disclosing insights to reduce information overload
Improving visualization types.
After working through initial sketches, I jumped into rough prototyping and testing using Keynote.
The most viable proposal showed the top key influencers and dynamically added them to the dashboard. From here a user could learn more, explore different influencers, or view the relationship between two influencers.
I worked with the Product Expert to host design thinking workshops to rapidly ideate different visualization formats which could communicate and express “impact” the best.
I used color to define a system that made it easier to digest each visualization and visually link related insights.
In terms of visualization types, I landed on a stacked histogram that showed the target variable’s distribution by each predicted influencer group.
I ran a series of usability evaluations & user interviews with various SAP customers to validate my design proposal and gather more user needs for feature enhancements.
The results of these sessions helped me create a more useful design. I used affinity mapping to analyze the research findings and map to existing features to identify overlaps and help prioritize user needs. I also created a customer journey framework to identify the touchpoints shared by the 4 Core BI Personas (admins, designers, analysts and decision makers) and how future versions of Smart Discovery can help each user persona.
Once the final design was signed off, I polished the final visuals, created pixel perfect mockups and exported these to Zeplin for developer handoff. I continued to support the developers by providing alternative solutions when we hit technical roadblocks.
What I Learned
My key takeaways included navigating big data models, facing ambiguous user demands, and understanding regression and classification models.
Each lesson helped further my skills when facing complex technical problems, addressing ambiguous business needs, and designing a functional and delightful user experience.
Linkedin Blog: Smart Assist - SAP Analytics Cloud starts analyzing for you.
Continuing the Vision...
Smart Discovery 2.0 was soft released in Q2 2018. Since then I have been focusing on a UX Vision which is bringing dynamic natural language query (NLQ) driven user experience to data discovery. The goal is to help empower decision makers and non-analysts to discover key business insights in a way that’s easier, faster, and accessible from any application within the SAP Intelligent Enterprise.