And saves €500.000 a year - every year.
MACHINE LEARNING MEETS PRODUCT DESIGN
MACHINE LEARNING MEETS PRODUCT DESIGN
This case study features the overhaul of an
outdated software that was used by
employees on a daily basis.
Through combining design and machine learning
we automated 95% of manual workflows and
resolved years of logged employee complaints.
As the lead designer for this project I was
responsible for the research, workflow
optimization, component design and
prototyping.
This project opened the possibilities to
include machine learning for other parts of
the business and enable further workflow
optimization as well as cost savings.
This case study features the overhaul of an
outdated software that was used by
employees on a daily basis.
Through combining design and machine learning
we automated 95% of manual workflows and
resolved years of logged employee complaints.
As the lead designer for this project I was
responsible for the research, workflow
optimization, component design and
prototyping.
This project opened the possibilities to
include machine learning for other parts of
the business and enable further workflow
optimization as well as cost savings.
Overview
Where it all started
Where it all started
"Checking invoices sometimes feels like looking at an Excel sheet for several hours. It’s a challenge to keep concentration up, especially when the tool randomly decides to crash because it didn’t like to work that day."
"Checking invoices sometimes feels like looking at an Excel sheet for several hours. It’s a challenge to keep concentration up, especially when the tool randomly decides to crash because it didn’t like to work that day."
Ciara S. - Legal & Business Analyst Team Manager
Ciara S. - Legal & Business Analyst Team Manager
The previous software was created piece by piece over the years and was maintained by several different developers, most of which had left the company a long time ago. It was slow, crashed often, had dead interactions and no design guidelines to speak of.
Needless to say, the analysts who worked with the tool had left mountains of feedback over the years and it was a major decision factor for the re-design.
The previous software was created piece by piece over the years and was maintained by several different developers, most of which had left the company a long time ago. It was slow, crashed often, had dead interactions and no design guidelines to speak of.
Needless to say, the analysts who worked with the tool had left mountains of feedback over the years and it was a major decision factor for the re-design.
Research - Understanding the problem
Research - Understanding the problem
A deep understanding of the current tool, its shortcomings and what the employees that use it think was the first step.
A deep understanding of the current tool, its shortcomings and what the employees that use it think was the first step.
Workflow Documentation
The current workflow was documented to gain a better understanding of the tasks that analysts are doing and to identify the most impactful areas for the automation.
Feedback Ticket Analysis
A total of 76 feedback tickets were logged over a 2-year period. These were sorted in main drivers and to have a comprehensive feature list.
User Surveys & Interviews
8 Zoom Meetings
2 Shadowing Sessions
60 Survey Results
Analysts, Team leaders and new employees were interviewed. In total that meant:
Feedback Ticket Analysis
A total of 76 feedback tickets were logged over a 2-year period. These were sorted in main drivers and to have a comprehensive feature list.
Scoping - Figuring out task priority
Scoping - Figuring out task priority
The research findings were used to create an updated workflow, which identified the key areas that needed to be tackled first for both the engineering and design teams.
The research findings were used to create an updated workflow, which identified the key areas that needed to be tackled first for both the engineering and design teams.
A desired state was created in FigJam and it was used to get executive approval and allowed product management to create a project timeline.
This was invaluable and improved executive communication during progress reports as we could easily document feedback and show our planning.
"We need to move quick. Our Engineers need to start as soon as possible."
Ashley H. - CTO
Sketching - Bringing ideas to life
Sketching - Bringing ideas to life
Through the updated workflows we identified several design components that weren't part of the current design language system. Especially R.A.G indicators that were used to quickly communicate the machine learning states were a new idea that wasn't used anywhere else in the software.
To figure out the logic of the interactions and to create the first wireframe ideas, pen and paper sketches were used to brainstorm several ideas in a time-efficient manner.
Through the updated workflows we identified several design components that weren't part of the current design language system. Especially R.A.G indicators that were used to quickly communicate the machine learning states were a new idea that wasn't used anywhere else in the software.
To figure out the logic of the interactions and to create the first wireframe ideas, pen and paper sketches were used to brainstorm several ideas in a time-efficient manner.
The project was considered a success because it automated more than 95% of the manual steps that had been done by analysts previously. As a result, it was calculated that this new dashboard will save around €500,000 a year and it was used as an example to explore other areas of the business where machine learning can be implemented.
The project was considered a success because it automated more than 95% of the manual steps that had been done by analysts previously. As a result, it was calculated that this new dashboard will save around €500,000 a year and it was used as an example to explore other areas of the business where machine learning can be implemented.
Prototyping - An agile MVP approach with rapid iteration
Prototyping - An agile MVP approach with rapid iteration
After finalizing a direction with the help of the sketches, mid to high fidelity prototypes were created in Figma as a foundation to gather more user feedback and as a guideline for frontend developers.
These prototypes were basic and only covered the essentials as we wanted to reuse as many components that we already had and to be able to have several iteration meetings every day.
After finalizing a direction with the help of the sketches, mid to high fidelity prototypes were created in Figma as a foundation to gather more user feedback and as a guideline for frontend developers.
These prototypes were basic and only covered the essentials as we wanted to reuse as many components that we already had and to be able to have several iteration meetings every day.
The main challenge of this approach (Other than the sheer volume) was that developers already started development work even without the final design being approved.
As a result, many features were created and scraped, ultimately resulting in a decrease in productivity and it caused frustration with the development team.
The main challenge of this approach (Other than the sheer volume) was that developers already started development work even without the final design being approved.
As a result, many features were created and scraped, ultimately resulting in a decrease in productivity and it caused frustration with the development team.
Results