AI Summaries
Helping users quickly understand what matters on a CRM record by transforming long activity histories into concise, actionable summaries.
USERS
Sales, Customer Service, and Marketing reps
DESIGN DURATION
2 weeks
PORTFOLIO READ TIME
5 minutes
PROBLEM
As customers scaled, record pages accumulated long, chronological activity feeds.
Important context became buried, making it difficult to quickly understand what happened and what mattered.
Users needed fast, actionable insights to make decisions without scanning dozens of events.
An already very dense record page leads to heavy scrolling
OPPORTUNITY
Transform record pages from passive activity logs into active decision support tools by leveraging existing AI capabilities to scale insight across growing accounts without adding complexity or training overhead.
CONSTRAINTS
assumptions & resources
There were no existing AI summary patterns on CRM record pages, so the team had to design from first principles.
Because the underlying model didn’t exist yet, we had limited ability to test or validate prompts in advance. Tight timelines pushed us to ship an initial version quickly and rely on real usage to learn and iterate.
risks & dependencies
The biggest risk was user trust AI outputs were unpredictable, and edge cases weren’t fully known upfront.
Progress was also dependent on backend engineers and the ChatSpot AI team, which constrained scope and iteration speed and required close cross-team collaboration.
what does success look like?
adoption
We used message count (# of prompts generated) to measure active usage and demand i.e., whether users were choosing this feature as part of their workflow (and how it compared to other AI features), not just seeing it on the page.
retention
We used 14-day return rate to confirm the feature delivered repeat value beyond novelty. If users came back within two weeks, it was a strong signal the summaries were helpful in ongoing, real-world work.
usefulness
We used a usefulness feedback score to evaluate quality and trust directly whether the generated summary actually helped users, and where it failed (too long, missing key info, wrong emphasis), which numbers alone can’t explain.
RESEARCH THEMES
Due to a compressed timeline, we partnered with internal support and sales reps as proxy users to quickly pressure-test early concepts. Product and Design aligned on shipping a focused first iteration to validate behavior in-market, prioritizing real usage data over extended upfront research.
intent
Users wanted to quickly understand what a contact or company cared about without reading every interaction.
action
Reps asked for clear guidance on what to do next so they could act immediately after reviewing a record.
recency & momentum
Users needed to know how recently and how actively an account had engaged to prioritize outreach.
PROCESS
Unlike a traditional design to handoff to build workflow, this project required continuous collaboration with engineering as the AI model evolved. We designed and validated in parallel, adapting to new outputs and edge cases in real time.
In addition to working in tandem with backend engineers, I partnered closely with Content Design and participated in regular design critiques with both the AI team and my core design team to refine structure, hierarchy, and trust patterns as the experience took shape.
Content structure
Because AI patterns were still evolving internally, there wasn’t a clear standard for structuring summaries. With limited content design support, my PM and I defined the framework ourselves.
We explored paragraph-style and chat-based formats, but ultimately chose a structured, scannable approach that follows how reps prepare: engagement, intent, recency, then next steps. We favored precise timestamps to support CRM accuracy.
This became our working schema with backend, refined around what ChatSpot could reliably generate while minimizing hallucination risk.
UI EXPLORATION
I explored multiple UI patterns to seamlessly integrate AI into the record experience, balancing discoverability, workflow continuity, and alignment with the evolving AI design system.
Entry point
Selected a contained card over a floating trigger to create clear structure and visibility, while controlling vertical height on an already dense record page.
Contextual in-place rendering
Displayed the summary directly within the activity feed to keep insights anchored to surrounding activity and avoid fragmenting the workflow.
Density within a moving system
Integrated the feature using the evolving ChatSpot AI component library, adapting to shifting visual patterns while ensuring the summary felt cohesive within the record experience.
