Reimagining enterprise search as an action-first, conversational AI Homepage
Reimagining enterprise search as an action-first, conversational AI Homepage
Reimagining enterprise search as an action-first, conversational AI Homepage
In development
As Product Designer, worked on WebUI App & Desktop App.
As Product Designer, worked on WebUI App & Desktop App.
As Product Designer, worked on WebUI App & Desktop App.
Tools used:
Tools used:
1,554 domains had AI switched on.
Only ~500 were using it.
1,554 domains had AI switched on.
Only ~500 were using it.
The homepage wasn't doing its job. Users landed, grabbed a file, and left, unaware that agents were available to perform their tasks. The redesign had one goal: close the gap between capability and comprehension.
The homepage wasn't doing its job. Users landed, grabbed a file, and left, unaware that agents were available to perform their tasks. The redesign had one goal: close the gap between capability and comprehension.
The homepage wasn't doing its job. Users landed, grabbed a file, and left, unaware that agents were available to perform their tasks. The redesign had one goal: close the gap between capability and comprehension.
Key Product Decisions
Key Product Decisions
Key Product Decisions
1 | The homepage was a navigation tool. We needed it to be a workdesk.
1 | The homepage was a navigation tool. We needed it to be a workdesk.
1 | The homepage was a navigation tool. We needed it to be a workdesk.
Monthly clicks on
Monthly clicks on
17,400
17,400
17,400
Bookmarks
Bookmarks
1,500
1,500
1,500
Tasks
Tasks
Tension: Data favored Bookmarks. Removing or deprioritising it risked disrupting familiar behavior.
Tension: Data favored Bookmarks. Removing or deprioritising it risked disrupting familiar behavior.
Decision: We reduced emphasis on Bookmarks to surface Tasks (signature requests, document reviews, and security alerts).
Decision: We reduced emphasis on Bookmarks to surface Tasks (signature requests, document reviews, and security alerts).
Why: High usage reflected habit, not value. Tasks represented work that required attention but remained buried.
Why: High usage reflected habit, not value. Tasks represented work that required attention but remained buried.
Before
Before
After
After
2 | Generic agent names don't get clicked. Context does.
2 | Generic agent names don't get clicked. Context does.
2 | Generic agent names don't get clicked. Context does.
AI engagement dropped from
AI engagement dropped from
38%
38%
38%
Week 1
Week 1
12%
12%
12%
Tasks
Tasks
Tension: Improving onboarding would increase awareness but not sustained usage.
Tension: Improving onboarding would increase awareness but not sustained usage.
Decision: We shifted from a generic 'AI Assistant' to contextual prompts powered by an LLM scoring layer.
Decision: We shifted from a generic 'AI Assistant' to contextual prompts powered by an LLM scoring layer.
Why: Users don’t think in product terms. The system reads recent files, tasks, and activity to surface actions in the user’s context, not the product’s language.
Why: Users don’t think in product terms. The system reads recent files, tasks, and activity to surface actions in the user’s context, not the product’s language.
Before
Before
After
After
3 | The homepage wasn't where desktop users worked.
3 | The homepage wasn't where desktop users worked.
3 | The homepage wasn't where desktop users worked.
42%
42%
42%
Desktop users accessed Egnyte through Finder or Windows Explorer.
Desktop users accessed Egnyte through Finder or Windows Explorer.
Tension: Driving users back to the homepage required behavior change, which rarely sustains.
Tension: Driving users back to the homepage required behavior change, which rarely sustains.
Decision: We introduced a lightweight desktop panel with AI input and pending tasks.
Decision: We introduced a lightweight desktop panel with AI input and pending tasks.
Why: Instead of pulling users into the product, we embedded the product into their existing workflow.
Why: Instead of pulling users into the product, we embedded the product into their existing workflow.
Before
Before
After
After
The challenge wasn’t to add more capability.
But to let AI become part of how work actually gets done.
The challenge wasn’t to add more capability.
But to let AI become part of how work actually gets done.
The challenge wasn’t to add more capability.
But to let AI become part of how work actually gets done.