THOUSANDS OF
CONVERSATIONS
AT YOUR
FINGERTIPS
An Agent tailor built for sales leaders to gain deep insights in the sales journey from every touchpoint in the sales cycle.
Project Details
Company
Scoot — Enterprise virtual events platform
Role
Design Lead
Scope
Defined and shipped an AI-driven sales intelligence interface to consolidate fragmented revenue workflows
Business Context
Enterprise B2B SaaS serving revenue and sales leadership teams
Primary Focus
Reducing reporting time and increasing confidence in sales insights
tl;dr
Sales leaders were manually synthesizing thousands of conversations across fragmented tools, limiting visibility and slowing decision-making. I defined and shipped an AI-driven interface that transformed unstructured transcripts into a continuous, queryable system with embedded source traceability. This shift reduced reporting time from hours to minutes per week and increased confidence in revenue insights.
The Problem
Revenue leaders lacked a cohesive way to extract signal from thousands of recorded sales conversations. Insights were scattered across 6–8 tools, requiring manual review and synthesis. This process was time-intensive, error-prone, and limited the ability to act quickly on emerging trends.
Despite the availability of raw data, decision-making was slowed by fragmentation and lack of trustworthy summarization.
The Solution
- •Built an AI interface capable of querying 5,000+ conversations
- •Implemented persistent conversational context across sessions
- •Embedded source-linked transcripts in every summary
- •Prototyped and validated model behavior in n8n before engineering integration
- •Designed role-based access and permission-aware data retrieval
The Results
- •Reduced executive reporting workflows from 2–6 hours per week to under 15 minutes
- •Consolidated 6–8 fragmented tools into a single interface
- •Increased executive confidence in AI-assisted decision-making
- •Enabled faster identification of sales trends and performance gaps
The Strategic Decision
Rather than building another dashboard layered on top of static metrics, I proposed a conversational AI interface that allowed executives to directly query unstructured sales transcripts in natural language.
The key design challenge was trust. Generative summaries without traceability risked hallucination and reduced confidence in high-stakes decisions. To address this, I structured the system to preserve conversational continuity across sessions and embed linked source references within every response. This required prioritizing traceability and decision confidence over raw response speed — a deliberate tradeoff to support enterprise use.
2–6 hrs → 15 min
Weekly Reporting Time Saved Per CRO
8
different touch points reduced to one source of truth
8-24x
times faster than traditional campaign analysis
5,000+
conversations simultaneously queryable
An AI Agent that is context aware, extensible to the org, and easily navigable from narrow to broad contexts and results.
Strategic insight: The challenge wasn't generating answers — it was preserving contextual continuity across fragmented sales interactions.
Audiences
The MVP was tailor built to meet the needs of the most immediately actionable user type: The Chief Revenue Officer.
CRO
The CRO manages large scale, complex sales initiatives and needs quality insights into the effectiveness and health of their campaigns, funnels, deals, and team members. Easily accessible and deep insights into sales conversations is of obvious value to these users.
Mapping CRO needs with agentic parameters
My process began with extensive research on CRO's workflows, tech-stacks, pain points, and needs. I then became an expert on agentic workflows and how to best match technical and experience parameters to meet those needs.
Using a combination of SME interview recordings, and LLM based research, I assembled a folder of relevant research that I used Notebook LM to synthesize and aggregate conclusions from.
I took all of my research insights and mapped them into a comprehensive CRO persona.
I mapped goals, needs, pain points, major topics and lines of questioning and began to conceptualize the type of experience that would be most beneficial to the CRO.
I then went on to map out the user journey, and collaborated with engineering to define the necessary backend that would enable the desired feature set.
I created an ideal, happy path user journey that I then brought to engineering leadership to understand what was technically feasible, most cost effective, and the clearest path to MVP. This helped me to refine my user journey and design into something that met both user needs and technical constraints. I built a functional n8n prototype to test out the quality and accuracy of various LLMs and query parameters.
Building the prototype before designing the interface
Before designing a single screen, I built a functional prototype in n8n to evaluate which LLMs and query parameters produced accurate, trustworthy results at scale.
This step is one most designers skip — but for an AI product where the quality of outputs directly determines user trust, validating the model behavior before committing to an interface was non-negotiable.
Testing variables included: model selection, context window size, prompt structure, and how the system handled ambiguous or low-confidence queries.
Finally, I created low-fidelity mockups that matched our desired user journey for final refinement and handoff.
This allowed me to design and validate more specific aspects of the experience by matching the specs with the backend functionality that we had previously established. I was then able to write up accurate and actionable acceptance criteria, and hand the epic off to engineering.
Rather than designing isolated features, the work focused on redefining how users access and explore organizational knowledge — shifting from navigation-based interfaces toward continuity-driven interaction.
Conversational intelligence layer replacing navigation-based analytics
Transformed fragmented sales data into a conversational intelligence layer, allowing CROs to query organizational knowledge through natural language instead of navigating disconnected dashboards. Designed context filtering and shared journeys to support both exploration and repeatable workflows.
Contextual continuity – from thousands of conversations to a single quotation
Designed traceable insight architecture that allows users to move seamlessly from high-level summaries to source-level quotations, ensuring trust and verifiability — a critical requirement for AI-assisted decision making.
Strategic Trade-off — We prioritized traceability and context continuity over raw generation speed because accurate, auditable insights were essential to executive confidence and operational decision-making.
Maximizing customer value from technical lift and infrastructure.
Leveraged underlying infrastructure to expand product value beyond initial scope, enabling rapid deployment of additional agentic experiences (e.g., help/knowledge bots) without significant engineering overhead — demonstrating scalable system architecture.
From fragmented tools and inputs to real-time sales intelligence
The platform transformed thousands of unstructured sales conversations into a single, queryable system executives could rely on for decision-making.
By consolidating 6–8 fragmented tools into one interface, leaders could directly explore trends, performance gaps, and campaign signals across 5,000+ conversations without manual synthesis.
This shift dramatically accelerated executive reporting workflows, reducing weekly analysis from 2–6 hours to under 15 minutes, while increasing confidence in AI-generated insights through source-linked traceability.
Future Vision
Being able to query all Scoot meetings is a very valuable MVP – but future versions could include the ability to import and integrate additional data and content from external meetings, CRMs, and other databases in order to make Deep Journey the true source of truth for insights on the entire sales cycle.
What I'd Do Differently
The MVP was scoped tightly to the CRO persona, which was the right call for speed — but it meant we underdesigned for the sales rep persona who would ultimately be the heaviest user of the data being queried. I'd have pushed for at least one round of rep-facing validation before shipping to ensure the access control model and organizational visibility layer matched how reps actually think about their data ownership.
I'd also have designed a more explicit onboarding moment for first-time AI interaction — many CROs hadn't used a conversational analytics tool before and the blank prompt state created hesitation we hadn't fully anticipated.
Designed & built with love in Figma + Cursor with Claude Code
© 2026 Jon Ezell. All rights reserved.