AI is a rapidly growing technology that has the power to transform digital products and offerings. Gravvy aimed to enhance user experience and differentiate themselves by introducing generative AI into their product. Over a rapid 2-month design iteration, the focus was on creatively leveraging a Large Language Model (LLM) to assist users in crafting a personalized charitable giving portfolio. The team explored innovative ways to utilize LLM capabilities, honing in on the idea of aiding users in the creation of a charitable giving portfolio. This involved understanding user preferences, values, and interests to dynamically generate suggestions for charities aligning with individual preferences.
Through iterative design cycles, the integration of generative AI seamlessly became a pivotal feature of Gravvy. Users can now leverage the platform to discover and support charities tailored to their philanthropic inclinations. The LLM works as a virtual assistant, offering personalized recommendations and streamlining the process of curating a diverse and impactful charitable portfolio.
How Might We use Generative Artificial Intelligence to create a better charitable donation experience?
I was the sole UX & UI designer from concept to launch. I was responsible for creating a new flow that could easily be adopted into the current product. My process consisted of crafting user journeys, rapid iterations of wireframes, and assisting in user research.
October 2023- Dec 2023 (Launching Jan '24)
Figma, Illustrator, Chat GPT
For this project, I collaborated with a software developer. We met frequently with our Stakeholders for design critiques and workshopping sessions.
Through a few cycles of iterations, our team came up with a design that both pleased the Stakeholders and made for a better user experience. Initially, Stakeholders requested for the LLM to play a bigger role in the recommendation process. However, as I presented the first iteration our team quickly realized that an LLM can present a lot of problems that we will not be able to address in this short time-frame, and secondly that the end-result might be better suited by using machine for parts of our experience.
Success metrics I intend to measure post launch include the following Key Performance Indicators: user retention rate, user feedback, time on task and task completion rate.
This project's main constraint was time. We were tasked 2 months to ideate, flow, wireframe, and develop this feature. Other challenges included AI limitations and pitching to Stakeholders.
We were given 2 month deadline to conceptualize and develop this feature.
Working with an LLM proved to be a difficult task for our team and due to its limitations and challenges. Prompt engineering will remain an ongoing process in the product.
Giving a persona to the LLM for our chatbot enhances humanization, brand consistency, user connection, and overall user experience. It allows the chatbot to communicate in a way that aligns with the company's identity, fostering a more engaging and memorable interaction with users. It also allowed us to discuss safety mechanisms, tone of voice, and response length.
Research for this project was largely focused around Large Language Models. It required weeks of testing various LLMs evaluating for accuracy and errors, overall performance, cost-benefit analysis, comparison with alternatives, and overall feasibility. By conducting comprehensive research in these areas, our team was able to make an informed decision about whether to integrate the LLM into our product and how to do it in a user friendly way.
Our product is unique in that users are able to specify allocations to each charity in their portfolio. However, this request to the LLM was found to perform poorly and gave inaccurate results.
LLMs are a great resource when asked in a format they like. We found that giving an LLM full control over our product puts the onus on the user to ask good questions.
Through testing, we found that having to think of questions, wait for responses takes a much longer time to generate a portfolio than other methods we've considered. The abandonment rate would likely be high if our process took a long time to complete.
The initial ask for this feature was to create a simple chatbot that acts as a "Giving Coach" that assists users in creating a personalized giving portfolio. We quickly realized how complex LLMs can be and how we can't fully rely on them yet to give predictable and truthful answers, which is important in our use case. From a user experience perspective, this approach was not ideal to retain user attention. While Chat GPT and other generative AI bots are interesting and fun to use for those of in the field, most non-technical people would not spend time chatting with a bot to create something that could take minutes with another approach, and as a new product on the market we need it to be engaging.
Our team's solution was to present the user with a short sequence of questions and gather data for their preferences. Through machine learning, we would extract recommendations using our own dataset to provide the portfolio. This solution was simple, easy to understand, and quick to complete. Although this would make for a great user experience, general feedback from Stakeholders was that utilizing an LLM in our product remained high priority. So, our team went back to work!
Leverage agile frameworks to provide a robust synopsis for high level overviews. Iterative approaches to corporate strategy foster collaborative thinking to further the overall value proposition.
Social connections may also impact the relevancy of some search results, either within a social media network or at a ‘mainstream’ search engine. Social media itself is a catch-all term for sites that may provide radically different social actions. For instance, Twitter is a social site designed to let people share short messages or “updates” with others.
As we expect most first-time users to use Gravvy on a mobile device, we needed to ensure that our design would work exceptionally on smaller screens. For this final iteration, I took a mobile-first approach and designed for the smallest screen before the larger screens in order to accommodate user needs, align with current technology trends, and improve the overall user experience.
Designing for a generative AI product was both fun and challenging. Technology was our biggest constraint and informed most of our design decisions. This project taught me how valuable feedback is when rapidly iterating.
I also learned how to advocate for my designs and explain design rational to stakeholders.
The project's highlight was a collaborative design process with a small team. I enjoyed seeing how we integrated quickly and developed a new feature in just a few months. I look forward to seeing the success in this feature and gathering user feedback post launch!
