AI Search
Senior Product Designer
2025
Problem
Keeping pace with changing expectations
As a platform built to help investors uncover insights, Affinity recognized the need to evolve beyond manual workflows and align more with what customers expected of digital experiences as AI became more prevalent.
Expectations were shifting quickly and many customers felt Affinity was no longer meeting their baseline needs. Within just a year, our survey showed a significant increase in teams planning to incorporate AI into their workflows, which highlighted the opportunity to rethink how the product could better support them.
Growing customer demand and the evolving AI landscape created a strong sense of urgency among leadership and we knew we needed to move quickly to avoid falling behind the market.
Research
How can AI deliver meaningful value?
Although we knew that AI needed to be included in our broader strategy, we wanted to avoid adding an AI chat experience just because everyone else was doing it. We were looking to understand where the real need was and identify how it could deliver genuine value.
I led a round of interviews with some of our largest customers to better understand their pain points in the current experience, how they were using AI in their workflows at that time, and where they felt Affinity was falling behind.
Since Affinity is a relationship intelligence platform, users spent a significant amount of time evaluating relationship strength, gathering context, and determining the right person to engage for a potential opportunity. A lion's share of this process was manual and prone to blind spots since information was fragmented across emails, notes, and CRM records.
We learned that some customers were even resorting to third-party AI tools, like ChatGPT and Perplexity, to manually piece together insights outside of Affinity, which was a strong signal that the product wasn't supporting their needs. We saw an opportunity for AI to synthesize this fragmented information and help users make faster and more confident decisions.
Hypothesis
Making it faster and easier to discover insights
We hypothesized that introducing AI into the CRM could synthesize relationship data across emails, meetings, and notes and help users to identify the right types of people to engage more quickly and confidently.
This could help address gaps in the existing experience and increase engagement by making Affinity more valuable within customers' everyday workflows.
Approach
How do users want to interact with their data?
There were several potential ways users could access insights and we wanted to understand which interaction model felt most useful and natural. We explored three different approaches:
Recommendations: Proactive insights surfaced in context
Search: Finding answers to specific questions
Dialogue: Refining ideas through conversation
We presented each concept to customers to gather feedback on where each model worked best and which they preferred.
While customers found value in all three model types, we chose to start with search since it offered the best balance of customer value and feasibility.
From a technical perspective, search also gave us a way to collect real user queries and understand how people naturally asked questions. Those interactions would be valuable training data for improving response accuracies and informing future models.
By starting with search, we could be able to establish confidence in the underlying capabilities before expanding into more proactive and conversational experiences.
Explorations
Starting with a high-traffic entry point to meet users where they are
Rather than creating a separate destination for AI search, we chose to integrate it into Affinity's existing global search. Since global search was already one of the product's highest-traffic entry points, this gave us a stronger opportunity to drive adoption by introducing the new capability within a familiar workflow.
When I began my explorations, I assumed users wouldn't care whether results came from a keyword search or AI — as long as they found what they were looking for, the method wouldn't matter to them.
However, we learned that users approached each result type with different expectations and wanted ones that were AI-generated to be clearly distinguished so they could evaluate them through the right lens.
Separating the two made even more sense since they both supported different types of interactions. Keyword search returned results instantly as users typed, while AI search required users to submit a query and wait for the system to generate a response.
Iterations
The need for explainable results and the ability to ask follow-up questions
We used Claude Code with the Figma MCP to build an interactive prototype populated with customers' actual data. This made the feedback sessions feel more realistic since they were interacting with familiar entities. As we reviewed our initial concepts with customers, a few additional insights emerged:
Trust was a major barrier: Users were skeptical of AI-generated results and wanted greater transparency into where the information came from. This was especially important since the ML team was still building a POC and the responses weren't always accurate. To preserve trust, we needed to prioritize clear citations and direct links to sources so users could quickly verify answers for themselves.
Users didn't always know exactly what they were looking for: They often started with a vague idea and needed help refining it, which validated the value of a conversational experience that could narrow in on the right answer. This also aligned with the AI tools they were already using and familiar with.
Although conversational AI wasn't part of the initial scope, it was important for us to still think about the broader ecosystem to ensure that AI Search could seamlessly evolve into a richer experience over time.
Final Designs
The vision and path to get there
We presented our AI strategy to the broader organization and shared a vision for evolving Affinity into a more intelligent partner that could proactively surface insights, answer questions through an AI-powered search, and help users refine questions through a conversational experience.
Longer term, the vision also included agentic workflows where users could automate their tasks to help streamline their workflows more.
We paired that vision with a phased roadmap showing how we planned to bring it to life, with AI search positioned as the first milestone. Sharing the broader strategy gave teams visibility into the product's long-term direction and helped them understand how AI could influence their own roadmaps and priorities.
In the final designs, we refined the original segmented control into a clearer, more discoverable affordance that made AI search easier to find. For power users, we also introduced keyboard shortcuts that aligned with existing system patterns.
During early testing with ML engineers, we knew the AI wouldn't always generate accurate results in this first phase. Instead of trying to hide these limitations, I focused on communicating when results were of low confidence or when no matches were found.
I felt that occasional inaccuracies were acceptable as long as users understood the system's level of confidence and didn't feel like they were being misled.
To align leadership and the broader organization around our long-term direction, I created hi-fi mockups and an interactive prototype that brought our vision to life and illustrated how the experience would evolve across future phases.
Our vision was to evolve AI Search into an intelligent partner that supported conversational exploration and proactively surfaced contextual recommendations throughout users' workflows, making it easier to uncover opportunities and take action.
Lastly, we established a set of AI design principles to ensure every AI interaction felt predictable and consistent. I partnered with our marketing design team to create a shared framework that served as guidance for designers when incorporating AI moments into their features.
Impact
AI with a clear purpose
Our strategy laid the foundation for Affinity's broader AI roadmap and was introduced to customers as part of our future vision. Early feedback confirmed that AI-powered search solved an immediate pain point while also demonstrating the long-term value of our direction.
Takeaways
Design's role in creating alignment and clarity for stakeholders
This initiative drew significant leadership attention and cross-functional visibility, as multiple teams relied on it to inform their own roadmaps. Keeping everyone aligned became just as important as designing the product itself. My PM, engineering lead, and I regularly shared work in progress through weekly product reviews and updates over Slack to keep the broader organization aligned on our decisions and rationale behind them.
Clearly communicating the long-term vision was essential for aligning the organization around a shared direction. By presenting hi-fi concepts early, we helped stakeholders understand where we were headed, why we were making certain decisions, and how each phase contributed to our broader strategy.
This project helped reinforce to me that design can extend beyond just the product itself. It can play a critical role in aligning teams around a common vision and help navigate ambiguity with confidence.