AI Writing Assistant for Efficient Workflows

AI Writing Assistant for Efficient Workflows

AI Writing Assistant for Efficient Workflows

Web Feature for OpenGov, 2023

Overview

Open Gov provides state and county governments in the United states with cloud solutions. Their Procurement web platform allows government officials to create RFP documents, make bids & interact with sellers to procure resources. Writing these documents is a time-consuming and frustrating process for users.

To tackle this issue we introduced an generative AI feature to assist user in the document creation process. The feature we designed saves the user's time & gives the organisation a competitive edge. We focused on introducing the right level of automation, building user trust & ensuring ethical practices as the organisation introduced their first AI feature.


  • Role Product Designer ( UX Research, Ideation, Prototyping, Usability Testing, UI)

  • Team Design Lead, Project Manager, Product Owner, Developer

  • Outcome This project went live in Sept ‘23, it was debuted at a National US conference (NIGP '23) to over 2K+ attendees.

The Problem

Government officials spend a lot of time creating Requests for Proposals (RFP’s) for every new procurement on a very frequent basis.


  • 100+ Over a 100 RFP’s are filed per fiscal year to make bids to procure resources.

  • 23.8 Hrs It takes about 23.8 hours to write a single RFP, this varies based on project complexity and resources.

  • $665 Billion The government spends roughly billions on contracts to procure/buy resources per fiscal year.

Research

In the research phase, my methods were aimed at understanding the product, the stakeholders vision and the user goals & challenges. I discussed the constraints on the project & worked closely with the dev team to understand what we could and couldn’t achieve. I did a task analysis. Here is a snapshot of the discovery.

Who am I designing for

The primary user's for this feature are procurement officers/contracting officers/project managers. The key insights from user research relevant to this project told us that our users were -


  • Hesitant & Mistrust New Tech - Hesitance of adopting new technology was rooted in concerns about security, reliability, cost and broader implications.


  • Values Ethics, Security & Transparency - Our user’s top priority is transparency on how their data is being used, stored. Being transparent about this is really important to our user.


  • Variance in Technical Proficiency - In the government sector we see a wide range of technical proficiency due to diverse backgrounds.

What do users struggle with

Since user's spent a large amount of time on the scope of work, for the scope of this project we focused on improving this task-flow. The scope of work is a detailed and crucial part, requiring careful preparation.


Below is a Hierarchical Task Analysis diagram outlining the three ways in which user's went about creating a scope of work.


  • We saw that while writing manually user's spent a lot of time thinking and articulating before they even started writing.


  • When users chose to use the template library, most time was spent finding relevant resources, deciding which parts of it made sense for their project & then modifying it.


  • Using suggested content required a lot of effort on the user's part to read, understand and then map that information to their own project requirements.


Design Principles

Human-Centred Automation : Our goal was to empower users by amplifying human abilities rather than pure automation. Positioning a tool as a co-pilot or assistance, and allowing a lot user control & flexibility.


Seamlessly integration in the Ecosystem: I wanted the user’s to be able to adopt and use this feature easily. My aim was to reduce the friction to try out a new feature by leveraging user’s existing mental models and familiar patterns.


Transparency Build’s User Trust: Rooted in our user’s value in ethics, transparency and security. I wanted to ensure open communication on how the AI works, how the data is being used and saved.

Proposed Solution

Generative AI co-pilot to write the “scope of work section” of the RFP documents.


Why AI over other automation tools?

  • Contextual Understanding and Adaptability

  • Fosters Innovation & Competitive Edge

  • Content Customisation

  • Speed & Efficiency

Iteration and Explorations

I explored a couple of task flows, I narrowed down to two ways to integrate the AI co-pilot feature. One was a modal option, we explored this option since other automation tools throughout the platform used modals. The other was a non-modal option. I created prototypes and tested these out. In the next sections, I go over all the insights we uncovered and the changes we made.

User’s are not prompt-engineering experts

We saw that user’s struggled with getting the right output. Prompt-engineering is crucial to ensure our user’s get good results from Gen AI. This was switched to the back-end. We changed the open-ended input box, to asking user’s for a very specific/customised input based on the context of their RFP title.

Confused when encountered long loading states

At-times it took up-to a couple of minutes to generate the output, in this case user’s would be confused if the system was stuck, what’s happening and abandon the task. To avoid this we added messages that indicated that the AI is still working on the output.

Modals weren’t the best UX

Initially we explored this feature as a modal because modals were used throughout the application and it was a familiar user pattern. But we changed this due to modals interrupting workflows, too easy exit, long task time, lengthy output which isn’t suitable for a modal.

User Education on AI and Data Usage

Clearly stating that Chat Gpt 3.5 is being used to generate responses. A learn more button that navigates user’s to a blog post about how AI work, how their data is being used.

Final Solution

Hand-Off

The hand-off files consisted of annotated files, documented animation components, prototyped for user flow, states & an archive with research & design iterations.

Outcome & Impact

  • This project went live in Sept ‘23, it was debuted at a National US conference (NIGP '23) to over 2K+ attendees.

  • We also won a company hackathon for this project!

  • I made some contributions to the companies design system while I was there. I create a few components through this project. I also audited 6 applications and the different components that different applications used and created a new consolidated design system in Figma.

Gauri Rajmane

gauri.rajmane@outlook.com