User Experience in Automating Document Based Workflows
Eliiza employs a labelling user interface to locate and label fields within digital documents; e.g. fields like Business name, provider number etc. on an insurance claim. These labels help train machine-learning algorithms for automated document processing.
The ‘necessary evil' of manual labelling is repetitive and sometimes requires rework. The early interface for the labeller was cluttered and unintuitive. Pretzel’s mission was making the process simple enough that anyone could learn to use it as well as making it more efficient and accurate.
We conducted a four-week discovery that included competitor research to discover best practice, co-design with stakeholders as well as user interviews that uncovered these pain points with the current process:
- The boring and repetitive nature of labelling
- Poor information hierarchy, lack of task status or progress
- Inability to collaborate easily with others.
These insights informed our design principles: ‘Let me view and select easily’, ‘Show me what to focus on’ and ‘What’s next?’. Prioritised features decluttered the interface improving focus and reducing errors. Co-designing with developers ensured technical feasibility and stakeholder alignment. We created user flows, wireframes, an annotated prototype and a product roadmap from MVP to target state to guide future iterations.
Outcomes and Results
By understanding and solving for key pain points, we made labelling simpler; improving the interface and experience, within technical requirements and constraints. We made the repetitive and tedious task of training machine learning models more intuitive and efficient and built a strategic product feature roadmap for future iterations. The improved experience benefited users and ensured greater machine learning accuracy.