Context
By 2023, AI was table stakes in enterprise SaaS — but only when it was accurate, explainable, and trustworthy.
At Bigtincan, AI directly impacted:
Content discovery
Sales productivity
Platform trust
When it worked, adoption accelerated.
When it failed, trust collapsed.
Starting point
SearchAI launched as a semantic layer alongside keyword search.
It proved the tech.
It exposed the problem:
AI was outpacing the platform’s ability to support it coherently.
Developer's Proof of Concept
Pressure
With GenieAI on the roadmap and acquisitions accelerating, AI could no longer ship as standalone features.
It had to become a shared platform capability.
Core tension
AI raised expectations — but the platform wasn’t built for intelligence as a system.
Conflicting search models
Fragmented retrieval logic
Product-specific AI behavior
Minimal analytics & experimentation
UX involved too late
The challenge wasn’t adding AI.
It was stopping intelligence from fragmenting the experience.
The problem
AI-powered search and traditional search could return different answers on the same page.
Across the platform:
SearchAI and keyword search disagreed
GenieAI relied on SearchAI, but UI patterns weren’t consistent
Some areas had no real search — only filters or manual navigation
Users couldn’t tell:
Which answer to trust
Where AI applied
Why results differed
Whether content was complete or scoped
Where trust broke
A sales rep searches for a pitch deck before a customer call.
On the same screen:
SearchAI highlights one document as the “best result”
Traditional search ranks a different file first
Both look valid.
Neither explains why.
The rep hesitates — not because AI failed, but because the system presented two conflicting truths.
The root issue
What we learned
These weren’t isolated UX issues. They were symptoms of:
Fragmented retrieval models
Inconsistent discovery entry points
No unified search surface
AI logic disconnected from user mental models
The pattern was clear:
AI needed to help users find, understand, and act on content — not live as scattered tools across the platform.
The constraints
AI logic defined before design involvement
Limited analytics and experimentation tooling
Acquisition and merger priorities shifting the roadmap
Multiple products with inconsistent search foundations
Impact was qualitative and directional, validated through Customer Success feedback, leadership alignment, and recurring usability patterns — not clean quantitative attribution.
What I owned
I focused on experience definition, interaction patterns, and trust models, working within engineering-defined AI constraints.
SearchAI
UX and interaction design within dev-defined logic
Visual and UI refinement to reduce confusion and duplication
Desktop and mobile experiences
GenieAI
Research and early concepting
User mental models and trust patterns
UI foundations for Knowledge Scopes
Handoff prior to full execution
Global AI Search (POC)
Design-led research and problem framing
Platform-level recommendations
Exploratory wireframes to align cross-functional teams
Phase 1 — SearchAI Stabilization
2023
Intervention
Clarified AI vs keyword results within legacy constraints
Outcome
Usability improved, core issues remained
Phase 2 — GenieAI & Trust Patterns
2023-2024
Intervention
Expanded AI into an assistant with Knowledge Scopes
Outcome
Enabled safe enterprise adoption
Scope as a Trust Boundary
Enterprise adoption required control.
Knowledge Scopes functioned as a trust mechanism:
Users understood what AI could and couldn’t see
Admins reduced risk without disabling value
Select a knowledge scope
Admin create a knowledge scope
Phase 3 — Global Search POC
2025
Intervention
Reframed AI as shared discovery layer across products
Status
Direction-setting. Not shipped due to acquisition priorities.
AI as an Overlay, Not a Destination
AI lived alongside existing UI, not in a separate experience:
Inline summaries
Contextual suggestions
Side-panel reasoning
Action prompts aligned to workflows
Why it mattered:
Lower cognitive load, higher trust, no forced behavior change.
POC concept showing AI layered into search results instead of pulling users into a standalone experience.
Prompt-based refinement explored how AI could narrow search intent without obscuring results.
Separate Retrieval from Interpretation
Search (retrieval): find the right things
AI (interpretation): summarize, explain, suggest
Combined moments: accelerate decisions
Why it mattered:
AI felt understandable, not magical or random.
Concept model separating search (retrieval) from AI interpretation to make system behavior easier to understand.
Optimize for “Next Steps,” Not Clever Answers
High-value AI moments enabled action:
Summaries without opening files
Cross-item suggested actions
Creation flows triggered in context
Why it mattered:
AI supported momentum, not distraction.
POC exploration of AI-driven summaries and actions triggered directly from search results.
Reduced confusion between AI and non-AI results
Clearer mental models for AI outputs
Safer adoption through scoped knowledge
Positioned AI as a platform capability, not a feature
Informed future search and consolidation discussions
Reduced long-term UX debt risk
Customer Success confirmed improved trust and comprehension
Leadership aligned on the need for unified search
Global Search POC referenced in platform-level decisions
Where it stalled
Platform and merger priorities paused execution
Limited analytics constrained validation
AI systems evolved faster than UX maturity
The real lesson
What I'd do differently
Push for earlier design involvement in AI architecture, especially around retrieval and ranking.
Several UX issues were locked in by technical decisions made before design input.
Even so, defining platform-level principles such as separating retrieval from interpretation, treating AI as an overlay, and optimizing for next steps created a durable foundation beyond any single feature.
