ENTERPRISE WEB APP
Led design of B2B machine learning app used by analysts to reduce loan delinquencies 30%.
Industries: Enterprise Technology, Business Services
Platform: Web
Competencies: Project management, research, contextual interviewing, UX and UI design, visual design, Agile software development
Deliverables: Requirements, sketches, wireframes, prototypes, design guidelines for data analytics apps
Team: Project Lead, two Data Analysts, one Business Analyst, two Researchers, two Developers
PROBLEM
Xerox Research had been providing predictive data analytics to internal BAs as a service. This required too many resources and the company needed to transition to a self-service model.
The pilot scenario was for financial services BAs to apply data analytics to help their big bank clients predict which borrowers might default. The banks could then reduce that risk by targeting those individuals with personalized communications.
COMPANY
Xerox is a former Fortune 500 company (now Conduent) providing digital print technology and related solutions. Offerings include managed document services, workplace solutions and graphic communications.
CHALLENGES
-
I did not have a good grasp of data analytics and I had to come up to speed quickly
-
It was difficult to simplify such a complex process while providing opportunities for a deeper dive
-
This was my first time collaborating with offshore developers
The target users were internal Xerox BAs who were experts in the business aspects of financial services.
They were responsible for understanding the problems their clients faced and finding solutions; they had a high level of business and vertical knowledge, with limited technical knowledge around data analytics.
To learn more about expert use of data analytics, I partnered with two user researchers to interview/observe 12 internal algorithm specialists and 14 external knowledge analysts, and to observe three internal teams developing analytics tools.
Themes emerged from these contextual inquiries that would help me design a tool for non-experts, for example:
Problem-definition is critical to success
“There’s so much data available that a lot of business types seem to have the impression that now they can just throw computers at it, and boom, they’re going have all kinds of information. You need to know what you’re looking for before you start looking…”
Keep things simple, but consider providing an advanced mode
“I would like to very easily say: I want to see this. No, I don’t like that. I want to go to this other thing. Let me actually play with the data rather than spend all my time trying to get it and match it.”
I did competitive analysis of seven popular self-service data analytics tools on the market along with Excel, to identify the most common features.
I discovered 16 key features to integrate into my designs.
1. Ability to integrate data from multiple sources 2. Support of structured and unstructured data 3. Automated analysis and visualization generation 4. An “advanced” mode for those with more statistical knowledge 5. Data exploration features such as filter, sort and drill down 6. Ability to create relationships between different columns of data in different tables 7. Include many types of visualizations 8. Include predictive analytics (if applicable) 9. Ability to add custom analytics and visualizations 10. Natural language summaries 11. Search capabilities 12. Ability to add informal notes about insights or observations 13. Ability to schedule processing 14. Collaboration tools 15. Ability to share and export data 16. Mobile access
Using the research information, I distilled the data analytics process into five high level steps, then integrated specific information from the loan risk reduction scenario.
-
Loan risk prediction
-
Web portal usage
-
Phone calls
-
Demographics
-
Payments
-
Macro-economics
-
Regression models for risk
-
Borrower scoring
-
Segmentation analysis
-
Geographic risk
-
Comparisons over time
-
Automated call
-
Automated email messages
I worked with the project team leader and our partner BA to define and document 11 high level platform requirements and 21 tasks that users should be able to complete. Examples include:
-
Enable BAs to make actionable business decisions that will help improve the quality of communication (e.g., calls, web) to the customer and reduce costs
-
Allow them to easily connect to existing data sources
-
Provide action output that can be used in existing communication channels
-
Support existing approval and reporting workflows
-
Provide a way to track outputs from predictive modules to ensure their validity
I led brainstorming and sketching sessions with the team, then wireframed in PowerPoint. Although not a typical wireframing tool, it allows key team members view and give feedback more easily.
I used as many existing design patterns as I could, but most of the platform required new patterns to support more complex workflows.
I integrated the five-step data analytics process into a new, wizard-like design pattern that made it easy for BAs to set up and run a solution.
30% reduction
in loan delinquencies and defaults for multiple big bank clients
The app was expanded for use with a variety of industry data sets.
I wrote an internal resource, User Experience for Analytics: Design Guidelines, which was used to inform many additional analytics projects and identify innovation opportunities.
The app was successfully transferred from the research group to a Xerox business unit for product implementation and deployment.