Kognitos set out to reinvent business automation by enabling non-technical users to automate workflows using natural language (english). The goal was to evolve the platform from a technically powerful but expert-dependent system into an intuitive, AI-native interface accessible to everyday business users in operations, finance, etc.
The primary persona is a business user who is not particularly tech-savvy and needs to automate workflows using natural language (english).
Typical example scenarios for the software include things like automating claims processing, or automating invoice processing.
Key flows that needed to be designed.
An end-to-end design and prototype that allows users to describe an automation using natural language, deal with clarifications and errors, iterate on it, and then test the automation.
Users start by describing the automation they want to create.
Users can edit the automation by writing directly in englsih or prompting the Kognitos agent. The agent detects issues or clarifications.
Users now run the automation and see the results. The user has to provide inputs. Errors are sometimes detected and the user is prompted to fix them.
Users can deploy the automation to their system.
The system take the initial prompt and runs with it. Clarifications happens after it creates the intial SOP (Standard Operating Procedure).
The system asks for clarifications before it creates the intial SOP. This ensures it can create a more accurate SOP to start with.
After informal usability testing with ~15 participants and discussions with internal stakeholders, we gained the following insights:
The system should take the initial prompt and run with it. Clarifications can happen after it creates the intial SOP in a more contextual way.
Adding inputs should happen in a more contextual way so that the user can go through them one-by-one and ensure they are valid.
The run results need to be a much more interactive and exploratory experience versus static information.
I designed and built a high-fidelity prototype that addresses the insights gained from the first iteration as well as uses Kognitos design system.
Users start by describing the automation they want to create.
Users can edit the automation by writing directly in englsih or prompting the Kognitos agent. The agent detects issues or clarifications.
Users must clarify details of the automation before it can run.
Users must provide valid inputs for the automation to run (e.g. email addresses, files, dates, etc.).
Users sometimes have to handle exceptions and errors that occur during the automation running.
Users can explore the results of the automation by prompting the Kognitos agent.
It was also necessary to design a landing page and a dashboard for the automations. Here users can view all their automations, create new ones, and manage them.
The dashboard is the main page for users to manage their automations. It provides a high-level overview of the automations they have created, as well as the ability to create new automations.
Users can view the details of an automation, including recent runs, inputs, and results.
Dark mode and light mode felt disconnected and incoherent. I strove to make the design system more consistent by exploring a visual design direction and color palettes that would work well across both modes.
I designed and built a branded page loader using CSS animation.