Taxonomy
Breaking down product meaning into connected insights
(Domain)
Analytics
(Year)
2024
(Services)
Product Design
ABOUT
THE PRODUCT
The Taxonomy project redefined how customer feedback is structured and explored.
The platform unifies feedback from app stores, communities, support tools, social channels, and sales calls, then structures it into a clear hierarchy. Predictions (keywords, themes, reasons) are organized into a taxonomy that lets teams filter, segment, and quantify insights at scale.
The legacy flat model led to redundancies, misclassifications, and heavy manual cleanup, limiting its ability to map feedback to business metrics or product areas.
The new system introduced two major shifts:
Adaptive Taxonomy: A multi-level, hierarchy (Keywords L1 > L2 > L3 + Themes/Sub-themes + Categories) that removes noise, prevents overlap, and aligns insights with real business structures.
Knowledge Graph: A relational model linking feedback to Users, Accounts, Products, and Opportunities—unlocking cross-object insights like revenue impact of specific themes.
Together, these upgrades turned a noisy system into a scalable, business-aligned intelligence layer.
Context
& Problem
As the platform matured, the limits of the flat keyword + reason model became clear, creating friction for users who relied on taxonomy to navigate insights. The structure failed to capture relationships or business alignment, leaving users with noisy, inconsistent, and incomplete data, these problems being:
Poor scalability and business alignment: The rigid structure didn’t map to real product or business hierarchies, required heavy manual upkeep, and couldn’t adapt to evolving data or organizational needs.
Unstructured and redundant taxonomy: The flat hierarchy lacked relationships between keywords and reasons, leading to overlapping labels, duplicate predictions, and fragmented insight discovery.
Low coverage and accuracy: Nearly half of feedback remained unclassified or misclassified, reducing trust in insights and weakening the system’s analytical reliability.
The result: users couldn’t reliably find, trust, or act on insights, undermining workflows and leaving the product underutilized.
Goals
The design effort focused on making the taxonomy and underlying architecture intuitive, scalable, and user-friendly:
Design a scalable taxonomy: Build a hierarchical structure that removes duplication and gaps while aligning taxonomy with real business structures—products, features, and teams—for clarity and consistency.
Enable smarter, connected insights: Lay the groundwork for cross-object relationships between Feedback, Accounts, Users, Products, and Opportunities to unlock richer, revenue-linked analysis in future iterations.
Streamline management and improve accuracy: Simplify taxonomy maintenance with intuitive UI and AI-driven suggestions, increasing prediction precision, recall, and coverage while reducing manual effort.
Initial Research
Our discovery process drew from a mix of customer conversations, support channels, and usage analytics, which consistently highlighted a critical issue: the taxonomy was difficult to manage, unreliable, and too labor-intensive. Users frequently spent hours cleaning duplicates, merging overlapping categories, and second-guessing the accuracy of insights. This friction directly impacted adoption, with analytics confirming that 40–50% of feedback records remained untagged, effectively blocking deeper analysis.
To validate these patterns, we looked beyond our own ecosystem and conducted a competitive and adjacent domain review. Platforms in the analytics and feedback space, as well as enterprise systems like CRM and knowledge graph models, revealed three consistent patterns:
Hierarchical structures with progressive granularity enabled users to move smoothly from broad to specific insights.
Business-aligned taxonomies mapped directly to product areas or organizational structures, improving relevance.
AI-powered suggestions reduced the manual overhead of classification and taxonomy maintenance.
Finally, we quantified the impact of these gaps to set measurable design criteria:
Coverage: Only 40–60% of records were classified due to the flat structure; we set a goal of 80%+ coverage with the update.
Manual effort: Taxonomy upkeep was unsustainable; we targeted a 70% reduction in ongoing maintenance with a more consistent and simple structure.
Together, these insights made it clear that the redesign needed to deliver a seamless, business-aligned, and adaptive taxonomy experience, one that improved accuracy while reducing user effort and unlocking deeper, more actionable insights.
Lo-Fi Design
Low-fidelity explorations focused on restructuring how taxonomy was displayed and managed:
Early concepts of multi-level keyword hierarchies (L1 > L2 > L3), The idea was to build a sort of a graphical representation that emphasises on the knowledge graph and lets the user understand and navigate through the connections seamlessly.
Wireframes for Theme/Sub-theme evolution from Reasons was scoped with.
Early testing with internal and pilot users validated the graphical approach for its clarity, scalability, and intuitive deep-dive experience.
Hi-Fi Designs
High-fidelity prototypes introduced a redesigned Taxonomy Manager UI:
Clear hierarchical navigation of keywords, themes, and categories.
Inline editing of taxonomy nodes with real-time AI suggestions.
Visualized Keyword Paths (L1 → L3) for context.
A chat-like agent to recommend taxonomy edits (e.g., adding new nodes after a feature launch) although scoped as a P2 release due to implementation scoping.
Transparency features like impact previews of edits and audit logs for governance.
Testing with design partners showed strong positive reception: the new flows felt intuitive, business-aligned, and scalable, addressing long-standing frustrations.
Outcome
Improved classification quality: 25% improvement in recall, boosting coverage to 80%+.
Reduced noise & overlap: 40% fewer duplicate/overlapping reasons.
Taxonomy management efficiency: ~70% reduction in manual maintenance.
Business alignment: Teams could now structure taxonomy in ways that mirrored product/feature groups, improving self-serve insights.
Customer retention: Customers saw faster time-to-insight and fewer manual workflows, strengthening trust in the platform.
Takeaway
What worked: Moving from a flat to a hierarchical structure unlocked clarity and adoption. Pairing taxonomy with business alignment and AI-powered editing created immediate value.
What was missing: Migration of saved dashboards and feeds required heavy manual effort, flagged as an area for automation in the future.
Learnings: Complex re-architectures demand progressive disclosure in UX. By guiding users step-by-step (L1 <> L3, Theme <> Sub-theme), the system remained approachable despite its depth.
Taxonomy
Breaking down product meaning into connected insights
(Domain)
Analytics
(Year)
2024
(Services)
Product Design
ABOUT
THE PRODUCT
The Taxonomy project redefined how customer feedback is structured and explored.
The platform unifies feedback from app stores, communities, support tools, social channels, and sales calls, then structures it into a clear hierarchy. Predictions (keywords, themes, reasons) are organized into a taxonomy that lets teams filter, segment, and quantify insights at scale.
The legacy flat model led to redundancies, misclassifications, and heavy manual cleanup, limiting its ability to map feedback to business metrics or product areas.
The new system introduced two major shifts:
Adaptive Taxonomy: A multi-level, hierarchy (Keywords L1 > L2 > L3 + Themes/Sub-themes + Categories) that removes noise, prevents overlap, and aligns insights with real business structures.
Knowledge Graph: A relational model linking feedback to Users, Accounts, Products, and Opportunities—unlocking cross-object insights like revenue impact of specific themes.
Together, these upgrades turned a noisy system into a scalable, business-aligned intelligence layer.
Context
& Problem
As the platform matured, the limits of the flat keyword + reason model became clear, creating friction for users who relied on taxonomy to navigate insights. The structure failed to capture relationships or business alignment, leaving users with noisy, inconsistent, and incomplete data, these problems being:
Poor scalability and business alignment: The rigid structure didn’t map to real product or business hierarchies, required heavy manual upkeep, and couldn’t adapt to evolving data or organizational needs.
Unstructured and redundant taxonomy: The flat hierarchy lacked relationships between keywords and reasons, leading to overlapping labels, duplicate predictions, and fragmented insight discovery.
Low coverage and accuracy: Nearly half of feedback remained unclassified or misclassified, reducing trust in insights and weakening the system’s analytical reliability.
The result: users couldn’t reliably find, trust, or act on insights, undermining workflows and leaving the product underutilized.
Goals
The design effort focused on making the taxonomy and underlying architecture intuitive, scalable, and user-friendly:
Design a scalable taxonomy: Build a hierarchical structure that removes duplication and gaps while aligning taxonomy with real business structures—products, features, and teams—for clarity and consistency.
Enable smarter, connected insights: Lay the groundwork for cross-object relationships between Feedback, Accounts, Users, Products, and Opportunities to unlock richer, revenue-linked analysis in future iterations.
Streamline management and improve accuracy: Simplify taxonomy maintenance with intuitive UI and AI-driven suggestions, increasing prediction precision, recall, and coverage while reducing manual effort.
Initial Research
Our discovery process drew from a mix of customer conversations, support channels, and usage analytics, which consistently highlighted a critical issue: the taxonomy was difficult to manage, unreliable, and too labor-intensive. Users frequently spent hours cleaning duplicates, merging overlapping categories, and second-guessing the accuracy of insights. This friction directly impacted adoption, with analytics confirming that 40–50% of feedback records remained untagged, effectively blocking deeper analysis.
To validate these patterns, we looked beyond our own ecosystem and conducted a competitive and adjacent domain review. Platforms in the analytics and feedback space, as well as enterprise systems like CRM and knowledge graph models, revealed three consistent patterns:
Hierarchical structures with progressive granularity enabled users to move smoothly from broad to specific insights.
Business-aligned taxonomies mapped directly to product areas or organizational structures, improving relevance.
AI-powered suggestions reduced the manual overhead of classification and taxonomy maintenance.
Finally, we quantified the impact of these gaps to set measurable design criteria:
Coverage: Only 40–60% of records were classified due to the flat structure; we set a goal of 80%+ coverage with the update.
Manual effort: Taxonomy upkeep was unsustainable; we targeted a 70% reduction in ongoing maintenance with a more consistent and simple structure.
Together, these insights made it clear that the redesign needed to deliver a seamless, business-aligned, and adaptive taxonomy experience, one that improved accuracy while reducing user effort and unlocking deeper, more actionable insights.
Lo-Fi Design
Low-fidelity explorations focused on restructuring how taxonomy was displayed and managed:
Early concepts of multi-level keyword hierarchies (L1 > L2 > L3), The idea was to build a sort of a graphical representation that emphasises on the knowledge graph and lets the user understand and navigate through the connections seamlessly.
Wireframes for Theme/Sub-theme evolution from Reasons was scoped with.
Early testing with internal and pilot users validated the graphical approach for its clarity, scalability, and intuitive deep-dive experience.
Hi-Fi Designs
High-fidelity prototypes introduced a redesigned Taxonomy Manager UI:
Clear hierarchical navigation of keywords, themes, and categories.
Inline editing of taxonomy nodes with real-time AI suggestions.
Visualized Keyword Paths (L1 → L3) for context.
A chat-like agent to recommend taxonomy edits (e.g., adding new nodes after a feature launch) although scoped as a P2 release due to implementation scoping.
Transparency features like impact previews of edits and audit logs for governance.
Testing with design partners showed strong positive reception: the new flows felt intuitive, business-aligned, and scalable, addressing long-standing frustrations.
Outcome
Improved classification quality: 25% improvement in recall, boosting coverage to 80%+.
Reduced noise & overlap: 40% fewer duplicate/overlapping reasons.
Taxonomy management efficiency: ~70% reduction in manual maintenance.
Business alignment: Teams could now structure taxonomy in ways that mirrored product/feature groups, improving self-serve insights.
Customer retention: Customers saw faster time-to-insight and fewer manual workflows, strengthening trust in the platform.
Takeaway
What worked: Moving from a flat to a hierarchical structure unlocked clarity and adoption. Pairing taxonomy with business alignment and AI-powered editing created immediate value.
What was missing: Migration of saved dashboards and feeds required heavy manual effort, flagged as an area for automation in the future.
Learnings: Complex re-architectures demand progressive disclosure in UX. By guiding users step-by-step (L1 <> L3, Theme <> Sub-theme), the system remained approachable despite its depth.
Taxonomy
Breaking down product meaning into connected insights
(Domain)
Analytics
(Year)
2024
(Services)
Product Design
ABOUT
THE PRODUCT
The Taxonomy project redefined how customer feedback is structured and explored.
The platform unifies feedback from app stores, communities, support tools, social channels, and sales calls, then structures it into a clear hierarchy. Predictions (keywords, themes, reasons) are organized into a taxonomy that lets teams filter, segment, and quantify insights at scale.
The legacy flat model led to redundancies, misclassifications, and heavy manual cleanup, limiting its ability to map feedback to business metrics or product areas.
The new system introduced two major shifts:
Adaptive Taxonomy: A multi-level, hierarchy (Keywords L1 > L2 > L3 + Themes/Sub-themes + Categories) that removes noise, prevents overlap, and aligns insights with real business structures.
Knowledge Graph: A relational model linking feedback to Users, Accounts, Products, and Opportunities—unlocking cross-object insights like revenue impact of specific themes.
Together, these upgrades turned a noisy system into a scalable, business-aligned intelligence layer.
Context
& Problem
As the platform matured, the limits of the flat keyword + reason model became clear, creating friction for users who relied on taxonomy to navigate insights. The structure failed to capture relationships or business alignment, leaving users with noisy, inconsistent, and incomplete data, these problems being:
Poor scalability and business alignment: The rigid structure didn’t map to real product or business hierarchies, required heavy manual upkeep, and couldn’t adapt to evolving data or organizational needs.
Unstructured and redundant taxonomy: The flat hierarchy lacked relationships between keywords and reasons, leading to overlapping labels, duplicate predictions, and fragmented insight discovery.
Low coverage and accuracy: Nearly half of feedback remained unclassified or misclassified, reducing trust in insights and weakening the system’s analytical reliability.
The result: users couldn’t reliably find, trust, or act on insights, undermining workflows and leaving the product underutilized.
Goals
The design effort focused on making the taxonomy and underlying architecture intuitive, scalable, and user-friendly:
Design a scalable taxonomy: Build a hierarchical structure that removes duplication and gaps while aligning taxonomy with real business structures—products, features, and teams—for clarity and consistency.
Enable smarter, connected insights: Lay the groundwork for cross-object relationships between Feedback, Accounts, Users, Products, and Opportunities to unlock richer, revenue-linked analysis in future iterations.
Streamline management and improve accuracy: Simplify taxonomy maintenance with intuitive UI and AI-driven suggestions, increasing prediction precision, recall, and coverage while reducing manual effort.
Initial Research
Our discovery process drew from a mix of customer conversations, support channels, and usage analytics, which consistently highlighted a critical issue: the taxonomy was difficult to manage, unreliable, and too labor-intensive. Users frequently spent hours cleaning duplicates, merging overlapping categories, and second-guessing the accuracy of insights. This friction directly impacted adoption, with analytics confirming that 40–50% of feedback records remained untagged, effectively blocking deeper analysis.
To validate these patterns, we looked beyond our own ecosystem and conducted a competitive and adjacent domain review. Platforms in the analytics and feedback space, as well as enterprise systems like CRM and knowledge graph models, revealed three consistent patterns:
Hierarchical structures with progressive granularity enabled users to move smoothly from broad to specific insights.
Business-aligned taxonomies mapped directly to product areas or organizational structures, improving relevance.
AI-powered suggestions reduced the manual overhead of classification and taxonomy maintenance.
Finally, we quantified the impact of these gaps to set measurable design criteria:
Coverage: Only 40–60% of records were classified due to the flat structure; we set a goal of 80%+ coverage with the update.
Manual effort: Taxonomy upkeep was unsustainable; we targeted a 70% reduction in ongoing maintenance with a more consistent and simple structure.
Together, these insights made it clear that the redesign needed to deliver a seamless, business-aligned, and adaptive taxonomy experience, one that improved accuracy while reducing user effort and unlocking deeper, more actionable insights.
Lo-Fi Design
Low-fidelity explorations focused on restructuring how taxonomy was displayed and managed:
Early concepts of multi-level keyword hierarchies (L1 > L2 > L3), The idea was to build a sort of a graphical representation that emphasises on the knowledge graph and lets the user understand and navigate through the connections seamlessly.
Wireframes for Theme/Sub-theme evolution from Reasons was scoped with.
Early testing with internal and pilot users validated the graphical approach for its clarity, scalability, and intuitive deep-dive experience.
Hi-Fi Designs
High-fidelity prototypes introduced a redesigned Taxonomy Manager UI:
Clear hierarchical navigation of keywords, themes, and categories.
Inline editing of taxonomy nodes with real-time AI suggestions.
Visualized Keyword Paths (L1 → L3) for context.
A chat-like agent to recommend taxonomy edits (e.g., adding new nodes after a feature launch) although scoped as a P2 release due to implementation scoping.
Transparency features like impact previews of edits and audit logs for governance.
Testing with design partners showed strong positive reception: the new flows felt intuitive, business-aligned, and scalable, addressing long-standing frustrations.
Outcome
Improved classification quality: 25% improvement in recall, boosting coverage to 80%+.
Reduced noise & overlap: 40% fewer duplicate/overlapping reasons.
Taxonomy management efficiency: ~70% reduction in manual maintenance.
Business alignment: Teams could now structure taxonomy in ways that mirrored product/feature groups, improving self-serve insights.
Customer retention: Customers saw faster time-to-insight and fewer manual workflows, strengthening trust in the platform.
Takeaway
What worked: Moving from a flat to a hierarchical structure unlocked clarity and adoption. Pairing taxonomy with business alignment and AI-powered editing created immediate value.
What was missing: Migration of saved dashboards and feeds required heavy manual effort, flagged as an area for automation in the future.
Learnings: Complex re-architectures demand progressive disclosure in UX. By guiding users step-by-step (L1 <> L3, Theme <> Sub-theme), the system remained approachable despite its depth.