Architecting an AI-Driven Sales Intelligence Ecosystem
Collaboration
Challenging the Brief
Rather than blindly executing a single dashboard, I sat down with the Product Manager to dissect user workflows and PRD specifications. Through mapping sessions, we realized the product served distinct personas: Sales Managers (who need macro analytics), Strategy/Coaches (who need transcript breakdowns and scoring rubrics), and Sales Reps (who need micro-tasks and pipeline prioritization).
Partnering with Engineering Early
To ensure complex data visualizations, multi-call stitching logic, and real-time AI recommendations wouldn't choke system performance, I aligned early with frontend engineers to map out API constraints, data table virtualization limits, and state-management rules for interactive tables and rapid filtering.
Design Exploration
Discarded Iterations
Early explorations attempted to use expandable table rows and nested accordions inside a single screen. User testing and design reviews proved this cluttered the UI, forcing users to scroll endlessly. This led to the structural breakthrough: splitting the product into dedicated, purpose-built dashboards.
Figma Showcase
I built a robust, scalable Figma architecture utilizing Component Properties, Advanced Auto-Layout, and Component Variables to handle complex data states (such as alternating lead temperatures - Hot, Warm, Cold - and dynamic metric cards). This kept developers happy with clean, tokenized handoffs and predictable component behaviors.
Design Solution
The final design architecture was structured into three distinct solutions tailored to the core personas, integrating advanced AI and PRD-backed logic:
Sales Insights & Rigor Dashboard (For Managers)
Centralized high-level analytics tracking team performance, talk time, and aggregate objection trends across business units to spot bottlenecks instantly.
AI-Driven Call Intelligence & Multi-Call Aggregation (For Coaches & Reps):
Smart Call Stitching: Designed the interface to ingest and synthesize up to three fragmented discovery calls around an anchor timestamp (60-minute window), ensuring reps get a single, unified summary note rather than disjointed logs.
Structured Rubrics & Red Flags: Built scalable UI modules displaying 5 core scoring dimensions (Background & Trajectory, Goals & Timeline, Past Interviews/Gaps, AI/GenAI Interest, IK Fit), an Overall Average Score, dynamic Lead Temperature (Hot/Warm/Cold), and a dedicated Red Flags section for unexpected call anomalies.
Graceful Error States & Edge Cases: Partnered with engineering to handle edge cases seamlessly—such as automatically posting a fallback notice (“Transcript unavailable; manual DCD review needed”) when a recording fails, ensuring zero false-positive data is pushed to HubSpot, alongside automated PII masking.
The Temperature Engine & Prioritize Leads (For Sales Reps)
A dynamic workspace allowing reps to filter pipelines by lead temperature, view primary objections, and instantly execute tasks via quick-audit IDs to eliminate lag between data and action.
Outcomes and Learnings
Impact & Quantified Success:
Streamlined data-to-action workflows, significantly reducing the time sales reps spent hunting for pipeline context.
Improved objection-resolution confidence by pairing reps with real-time AI-recommended playbooks directly inside their daily task list and HubSpot overview notes.
Slashed engineering handoff friction through a clean, tokenized component structure and clear rules for BigQuery versioning .
Retrospective & What I Would Do Next:
While the three dashboards successfully targeted specific personas, a future iteration should explore customizable widget layouts, allowing power users to pin cross-persona metrics to a personalized home feed based on changing quarterly goals.
