Team
Design: Me (Founding Designer)
Product: Harald Rosshirt (Head of Product)
Engineering: Nico Moschopolous (CTO)
Problem
In private M&A, proper nouns—what we call entities—are everything. If “Orion Energy” shows up in ten different spellings, asking Junior for Orion’s average NPS pulls in messy, diluted results. Multiply that across 40+ transcripts, and you’ve got noise.

(Dramatic reenactment)
The catch: users rarely review transcripts. They think they’re “close enough” and don’t realize entities are important. Before this solve, the only way to clean entities was inside the review flow—a flow almost no one used. We needed to break entity cleaning out entirely, make it dead simple, and remove its dependency on transcript review.
Our engineers built phonetic clustering to predict and auto-dedupe these variations.
In private M&A, proper nouns—what we call entities—are everything. If “Orion Energy” shows up in ten different spellings, asking Junior for Orion’s average NPS pulls in messy, diluted results. Multiply that across 40+ transcripts, and you’ve got noise.

(Dramatic reenactment)
The catch: users rarely review transcripts. They think they’re “close enough” and don’t realize entities are important. Before this solve, the only way to clean entities was inside the review flow—a flow almost no one used. We needed to break entity cleaning out entirely, make it dead simple, and remove its dependency on transcript review.
Our engineers built phonetic clustering to predict and auto-dedupe these variations.
Discovery
I worked closely with engineering to understand what a cluster might contain: number of entities, average mentions, and edge cases we’d need to handle.
I worked closely with engineering to understand what a cluster might contain: number of entities, average mentions, and edge cases we’d need to handle.
I worked closely with engineering to understand what a cluster might contain: number of entities, average mentions, and edge cases we’d need to handle.
Hypothesis
If users could merge entities outside the main review flow, we’d improve Junior GPT’s output by giving it cleaner, more accurate source material.
If users could merge entities outside the main review flow, we’d improve Junior GPT’s output by giving it cleaner, more accurate source material.
If users could merge entities outside the main review flow, we’d improve Junior GPT’s output by giving it cleaner, more accurate source material.
Solution
We designed a compact pattern that lets users:
Remove entities that don’t belong
Jump to transcript references
Unmark specific mentions
Select or add the correct spelling
The goal: if the cluster’s right, merging should be a single click. The “Suggested Merge” card handled the workflow, while a prominent notification badge on the Entities tab drove engagement.
We designed a compact pattern that lets users:
Remove entities that don’t belong
Jump to transcript references
Unmark specific mentions
Select or add the correct spelling
The goal: if the cluster’s right, merging should be a single click. The “Suggested Merge” card handled the workflow, while a prominent notification badge on the Entities tab drove engagement.
We designed a compact pattern that lets users:
Remove entities that don’t belong
Jump to transcript references
Unmark specific mentions
Select or add the correct spelling
The goal: if the cluster’s right, merging should be a single click. The “Suggested Merge” card handled the workflow, while a prominent notification badge on the Entities tab drove engagement.
Impact
Within a month of launch, usage jumped from zero to 180+ merges a day across ~1,000 DAU.
Within a month of launch, usage jumped from zero to 180+ merges a day across ~1,000 DAU.
Within a month of launch, usage jumped from zero to 180+ merges a day across ~1,000 DAU.
Conclusion
A niche but powerful pattern—proof that small, focused tools can have outsized impact.
A niche but powerful pattern—proof that small, focused tools can have outsized impact.
A niche but powerful pattern—proof that small, focused tools can have outsized impact.
More Case Studies
←
→
©2025 Alex Fortney. All Rights Reserved, etc. Please don't steal my things.