BOBBY KING
Case study: designing for big data at the Home Office
I worked with user researchers and stakeholders to design a case management system which would balance human decision making with AI recommendations.
The brief
The government needs a way of tracking purchases of dangerous chemicals, or substances which can be used to manufacture illicit drugs. Staff at the home office needed a case management system to manage referrals from two sources: a public web form and data submitted by retailers.
The solution had to be easy to learn while enabling the specialists to decide how to handle each referral. The purchase of a poison or an explosive could be suspicious, or it could be a routine purchase.
What I did
Learned about the teams daily tasks and habits
I planned out a programme of research with the user researchers and developed prototypes for them to test on Home Office staff.
We learned about how the team functioned. Every morning the team leader reviewed the new cases which arrived and allocated these to team members. They needed a way of detecting which cases were more suspicious.
Developed flags, search and filters
The case workers had a system which wasn't based on the Home Office design system. When we arrived it was difficult to filter the cases properly. I simplified the filters and worked with the developers to understand the search algorithm.
We also had to consider flags which were created by a machine-learning algorithm. As the algorithm was going to evolve over time we needed to offer some simple flags for identifying where the system was recommending the most urgent cases.
I developed a design system for the development team. They rebuilt the case management system to make it more useful and usable.
Iterated on user flows
After their team leader assigned them a case, the case managers would do their investigation and close the case or escalate it. They would assign a priority for the cases they passed on.
I also considered complex edge cases - for example, when a report had come in from two sources at once (an online retailer and an in-store purchase). It was possible that a member of the public buying a poison would attempt to buy it in more than one place. We needed to be able to flag useful links in the data.
The outcome
The work took around a year to complete. We ran multiple rounds of usability testing to iterate on our solution. At the end, the Home Office had a new product to offer their internal staff. As I was part of the Home Office community of designers I played back the new design to colleagues. I collaborated with designers who had worked on case management systems in order to develop the Home Office design system.