Motley
Motley
Motley
Discover
Discover
Discover
Key Deliverables:
Key Deliverables:
→
→
→
Web application;
Web application;
→
→
→
Landing page;
Landing page;
→
→
→
Marketing & security pages;
Marketing & security pages;
→
→
→
Brand visual system.
Brand visual system.
Timeline:
2025 → Ongoing
Timeline:
2025 → Ongoing
Role:
Lead Product Designer
Role:
Lead Product Designer
Role:
Founding designer → Head of Design
(team of 3 designers)
Engagement:
Embedded, long-term
Engagement:
Embedded, long-term
Motley is an agent-native semantic layer platform that turns enterprise data warehouses into a governed, auditable source of truth for AI agents and recurring reporting workflows.
Motley is an agent-native semantic layer platform that turns enterprise data warehouses into a governed, auditable source of truth for AI agents and recurring reporting workflows.
Existing AI reporting tools either hallucinate metrics or demand a full data-platform team to deploy. Motley needed an interface that makes a deeply technical product — semantic models, MCP servers, reusable masters — approachable for data engineers and business users generating the same report every quarter, without losing the rigour the warehouse demands.
Existing AI reporting tools either hallucinate metrics or demand a full data-platform team to deploy. Motley needed an interface that makes a deeply technical product — semantic models, MCP servers, reusable masters — approachable for data engineers and business users generating the same report every quarter, without losing the rigour the warehouse demands.
0→40+
0→40+
Enterprise workspaces deployed
Enterprise workspaces deployed
94%
94%
Onboarding completion rate
5x
5x
Faster time to first report
Define
Define
Define
Motley sits between two crowded categories: legacy semantic layers like dbt MetricFlow and Cube — built for BI dashboards, not autonomous agents — and AI reporting tools like Gamma or text-to-SQL assistants, which are fast but unreliable, returning different numbers on different runs.
Motley sits between two crowded categories: legacy semantic layers like dbt MetricFlow and Cube — built for BI dashboards, not autonomous agents — and AI reporting tools like Gamma or text-to-SQL assistants, which are fast but unreliable, returning different numbers on different runs.
Customer research showed neither side actually solves the problem teams have today: AI tools that hallucinate metrics, or governed data platforms that no agent can use.
Customer research showed neither side actually solves the problem teams have today: AI tools that hallucinate metrics, or governed data platforms that no agent can use.
22 expert interviews
22 expert interviews
We spoke with data engineers, revenue operations leaders, and product teams already trying to bolt AI onto their analytics stack. The recurring pain wasn't speed — it was trust. Existing tools couldn't guarantee that the same query returned the same answer twice, which made them unusable for anything customer-facing.
We spoke with data engineers, revenue operations leaders, and product teams already trying to bolt AI onto their analytics stack. The recurring pain wasn't speed — it was trust. Existing tools couldn't guarantee that the same query returned the same answer twice, which made them unusable for anything customer-facing.
This led us to ask: "How might we let agents generate reports data teams trust enough to ship without manual review?"
This led us to ask: "How might we let agents generate reports data teams trust enough to ship without manual review?"
Personas
Personas
Maya, the Data Engineer (32, Senior Data Engineer):
Maya, the Data Engineer (32, Senior Data Engineer):
→
→
Owns the semantic layer for a fast-growing analytics team;
Owns the semantic layer for a fast-growing analytics team;
→
→
Tired of reconciling conflicting metric definitions across ad-hoc SQL;
Tired of reconciling conflicting metric definitions across ad-hoc SQL;
→
→
Needs auditability, versioning, and agent-ready metric exposure.
Needs auditability, versioning, and agent-ready metric exposure.
Daniel, the Customer Success Lead (38, Head of CS):
Daniel, the Customer Success Lead (38, Head of CS):
→
→
→
Prepares 30+ QBR presentations every quarter, mostly by hand;
Prepares 30+ QBR presentations every quarter, mostly by hand;
→
→
→
Wants automation that keeps the narrative intact, not generic AI slides;
Wants automation that keeps the narrative intact, not generic AI slides;
→
→
→
Needs consistent outputs and an audit trail when numbers get questioned.
Needs consistent outputs and an audit trail when numbers get questioned.
Sarah, the Embedded Builder (34, Product Manager):
Sarah, the Embedded Builder (34, Product Manager):
→
→
→
Shipping in-product reporting without a dedicated data platform team;
Shipping in-product reporting without a dedicated data platform team;
→
→
→
Needs governed primitives for queries and documents that work out of the box;
Needs governed primitives for queries and documents that work out of the box;
→
→
→
Needs embedable primitives and predictable output to ship fast.
Needs embedable primitives and predictable output to ship fast.
Problem Statement
Problem Statement
Agent Reliability
Agent Reliability
Agent Reliability
LLMs writing SQL against raw warehouse schemas hallucinated joins, redefined metrics silently, and returned different numbers on different runs — making them unusable for anything customer-facing.
LLMs writing SQL against raw warehouse schemas hallucinated joins, redefined metrics silently, and returned different numbers on different runs — making them unusable for anything customer-facing.
Fragmented Metric Definitions
Fragmented Metric Definitions
Fragmented Metric Definitions
Every team defined core metrics like ARR, active users, or churn slightly differently across dashboards, exports, and ad-hoc queries — and no one caught the drift until it surfaced in a board deck.
Every team defined core metrics like ARR, active users, or churn slightly differently across dashboards, exports, and ad-hoc queries — and no one caught the drift until it surfaced in a board deck.
Manual Recurring Reporting
Manual Recurring Reporting
Manual Recurring Reporting
QBRs, monthly check-ins, and pipeline reviews were rebuilt by hand every cycle — impossible to audit, easy to break, expensive to scale across customers.
QBRs, monthly check-ins, and pipeline reviews were rebuilt by hand every cycle — impossible to audit, easy to break, expensive to scale across customers.
In-Product Reporting Cost
In-Product Reporting Cost
In-Product Reporting Cost
Shipping reporting features inside SaaS products required staffing a full data platform team — most product teams couldn't justify it and shipped nothing instead.
Shipping reporting features inside SaaS products required staffing a full data platform team — most product teams couldn't justify it and shipped nothing instead.
How might we?
How might we?
→
→
→
HMW expose warehouse data to AI agents without losing metric integrity?
HMW expose warehouse data to AI agents without losing metric integrity?
→
→
→
HMW make recurring business reports as repeatable as code, not as fragile as slide decks?
HMW make recurring business reports as repeatable as code, not as fragile as slide decks?
→
→
→
HMW design a semantic layer that both data engineers and business users can actually read?
HMW design a semantic layer that both data engineers and business users can actually read?
→
→
→
HMW give every generated report a visible, governed audit trail?
HMW give every generated report a visible, governed audit trail?
→
→
→
HMW let product teams embed AI reporting in their own products in weeks, not quarters?
HMW let product teams embed AI reporting in their own products in weeks, not quarters?
Key metrics
Key metrics
To track success, we monitored:
To track success, we monitored:
→
→
→
Time to first governed query — Measuring how fast new teams reached production-ready output;
Time to first governed query — Measuring how fast new teams reached production-ready output;
→
→
→
Master reuse rate — How often a template was rerun vs. rebuilt from scratch;
Master reuse rate — How often a template was rerun vs. rebuilt from scratch;
→
→
→
Document acceptance rate — Share of generated reports shipped without manual editing;
Document acceptance rate — Share of generated reports shipped without manual editing;
→
→
→
Agent query success rate — Queries returning governed, validated answers on first attempt;
Agent query success rate — Queries returning governed, validated answers on first attempt;
→
→
→
Onboarding completion — New workspaces reaching their first published document.
Onboarding completion — New workspaces reaching their first published document.
Iterate
Iterate
Iterate
Our design process for Motley was driven by the need to balance technically demanding concepts — semantic models, MCP integrations, governed query resolution — with an experience approachable enough for business users generating recurring reports. To achieve this, we collaborated closely with the data engineering, AI, and platform teams, ensuring every decision held up to both the rigor the warehouse demanded and the speed customer-facing workflows required.
Our design process for Motley was driven by the need to balance technically demanding concepts — semantic models, MCP integrations, governed query resolution — with an experience approachable enough for business users generating recurring reports. To achieve this, we collaborated closely with the data engineering, AI, and platform teams, ensuring every decision held up to both the rigor the warehouse demanded and the speed customer-facing workflows required.
First Iteration:
First Iteration:
→
→
→
Started with a chat-first interface — a single prompt field where users described the report they wanted and the AI assembled the entire document end-to-end;
Started with a chat-first interface — a single prompt field where users described the report they wanted and the AI assembled the entire document end-to-end;
→
→
→
Treated the semantic layer as background infrastructure, surfaced only when the agent needed it;
Treated the semantic layer as background infrastructure, surfaced only when the agent needed it;
→
→
→
Output was a flat document — generated text and charts in a single scroll, with no structural separation between sections;
Output was a flat document — generated text and charts in a single scroll, with no structural separation between sections;
→
→
→
Validated the core thesis that agents could produce coherent business reports — but exposed deeper friction below the surface.
Validated the core thesis that agents could produce coherent business reports — but exposed deeper friction below the surface.
Second Iteration:
Second Iteration:
Testing with data engineers and customer success teams revealed a consistent pattern: trust came from control, not magic. Users needed to see which data sources were selected, which metrics were used, and which parts of the document they could safely edit without breaking governance.
Testing with data engineers and customer success teams revealed a consistent pattern: trust came from control, not magic. Users needed to see which data sources were selected, which metrics were used, and which parts of the document they could safely edit without breaking governance.
→
→
→
Replaced the single prompt with a structured Master editor — typed slide blocks (Title, Summary, Chart, Custom) that map cleanly to the semantic layer;
Replaced the single prompt with a structured Master editor — typed slide blocks (Title, Summary, Chart, Custom) that map cleanly to the semantic layer;
→
→
→
Surfaced data source selection upfront, with explicit toggles for each connected dataset before generation begins;
Surfaced data source selection upfront, with explicit toggles for each connected dataset before generation begins;
→
→
→
Introduced custom values ({customer_name}, {time_period}) as first-class primitives, so a Master could be reused deterministically across dozens of accounts;
Introduced custom values ({customer_name}, {time_period}) as first-class primitives, so a Master could be reused deterministically across dozens of accounts;
→
→
→
Added a persistent audit panel showing which queries resolved, which metrics fired, and where the agent had to fall back to raw data;
Added a persistent audit panel showing which queries resolved, which metrics fired, and where the agent had to fall back to raw data;
→
→
→
Repositioned MCP integration setup earlier in onboarding, so agents like Claude or Cursor could trigger generation before users finished configuring the workspace.
Repositioned MCP integration setup earlier in onboarding, so agents like Claude or Cursor could trigger generation before users finished configuring the workspace.
These iterations reframed the product from AI that writes reports to a governed system that lets agents write reports your data team can defend. The shift unlocked enterprise adoption — moving Motley out of the prototype category and into production reporting workflows.
These iterations reframed the product from AI that writes reports to a governed system that lets agents write reports your data team can defend. The shift unlocked enterprise adoption — moving Motley out of the prototype category and into production reporting workflows.
Design
Design
Design
The Design phase translated research into a coherent product surface — one that could be operated confidently by data engineers writing semantic models and by customer success leads generating QBRs the same week. Every visual and interaction decision was tested against a single question: does this help the user trust what just got generated?
The Design phase translated research into a coherent product surface — one that could be operated confidently by data engineers writing semantic models and by customer success leads generating QBRs the same week. Every visual and interaction decision was tested against a single question: does this help the user trust what just got generated?
Color Scheme & Visual Identity:
Color Scheme & Visual Identity:
→
→
→
A light, near-neutral base was chosen as the foundation — Motley sits next to terminals, IDEs, and BI tools, and needed to read as an instrument, not a marketing surface. Dark themes carry premium consumer associations; light themes carry precision and density, which is what a data tool earns trust with;
A light, near-neutral base was chosen as the foundation — Motley sits next to terminals, IDEs, and BI tools, and needed to read as an instrument, not a marketing surface. Dark themes carry premium consumer associations; light themes carry precision and density, which is what a data tool earns trust with;
→
→
→
Violet was established as the primary brand signal — applied to AI-generated artifacts, governed actions, and anywhere the semantic layer surfaces. Crimson (#FF0059) was reserved as a secondary accent for state, emphasis, and editorial moments, used with restraint;
→
→
→
A mint-teal CTA was introduced for primary confirmations (Edit, Generate, Publish) — a deliberate counterweight to the violet system, giving the interface a clear signal for "this is the moment of commitment";
A mint-teal CTA was introduced for primary confirmations (Edit, Generate, Publish) — a deliberate counterweight to the violet system, giving the interface a clear signal for "this is the moment of commitment";
→
→
→
Typography is built on Figtree — a geometric sans with the warmth a B2B data tool needs to avoid reading as cold infrastructure, while holding up at the densities the product demands. Used across body, product UI, and marketing surfaces with weight changes carrying the hierarchy.
Typography is built on Figtree — a geometric sans with the warmth a B2B data tool needs to avoid reading as cold infrastructure, while holding up at the densities the product demands. Used across body, product UI, and marketing surfaces with weight changes carrying the hierarchy.
User Experience & Engagement:
User Experience & Engagement:
→
→
→
A progressive disclosure model was used throughout — semantic models, MCP integration, and query resolution are exposed gradually, with the underlying machinery accessible but never forced on users who don't need it;
A progressive disclosure model was used throughout — semantic models, MCP integration, and query resolution are exposed gradually, with the underlying machinery accessible but never forced on users who don't need it;
→
→
→
Master and Document editing was rebuilt around typed primitives — Slide Title, Summary, Chart, Custom Values — each mapped directly to a governed query in the semantic layer, so users edit business logic without writing SQL;
Master and Document editing was rebuilt around typed primitives — Slide Title, Summary, Chart, Custom Values — each mapped directly to a governed query in the semantic layer, so users edit business logic without writing SQL;
→
→
→
A live preview anchors the entire editing experience — every change to a Master resolves in real time against connected data sources, so users see exactly what their customer or stakeholder will see;
A live preview anchors the entire editing experience — every change to a Master resolves in real time against connected data sources, so users see exactly what their customer or stakeholder will see;
→
→
→
A persistent audit trail surfaces which sources, queries, and metrics fired during generation, turning "the AI made this" into a defensible chain of governed decisions — the single most important affordance for enterprise adoption.
A persistent audit trail surfaces which sources, queries, and metrics fired during generation, turning "the AI made this" into a defensible chain of governed decisions — the single most important affordance for enterprise adoption.
Adoption & Distribution Strategy:
Adoption & Distribution Strategy:
→
→
→
SLayer, the open-source semantic layer, was positioned as the top of the acquisition funnel — engineering teams adopt it for free under MIT, integrate it into existing warehouses, then graduate to the hosted Motley Cloud when they need orchestration, audit, and team controls;
SLayer, the open-source semantic layer, was positioned as the top of the acquisition funnel — engineering teams adopt it for free under MIT, integrate it into existing warehouses, then graduate to the hosted Motley Cloud when they need orchestration, audit, and team controls;
→
→
→
MCP-native architecture turned every connected agent — Claude, Cursor, Codex, or custom — into a distribution channel: any team using these agents can query Motley without changing their primary toolchain;
MCP-native architecture turned every connected agent — Claude, Cursor, Codex, or custom — into a distribution channel: any team using these agents can query Motley without changing their primary toolchain;
→
→
→
Onboarding was designed around a single first-value milestone: a connected data source resolving a governed query end-to-end — measurable, sub-30-minute, and shareable as a public artifact, modeled on the developer-led growth motion proven by Vercel and Linear;
Onboarding was designed around a single first-value milestone: a connected data source resolving a governed query end-to-end — measurable, sub-30-minute, and shareable as a public artifact, modeled on the developer-led growth motion proven by Vercel and Linear;
→
→
→
Templates (Masters) function as a soft network effect — every reusable QBR or pipeline review template authored inside a team can be cloned, versioned, and shared across workspaces, compounding value the longer a customer stays.
Templates (Masters) function as a soft network effect — every reusable QBR or pipeline review template authored inside a team can be cloned, versioned, and shared across workspaces, compounding value the longer a customer stays.
Measure & Test
Measure & Test
Measure & Test
Data Analysis & Testing Approach:
Data Analysis & Testing Approach:
→
→
→
Instrumented the product with PostHog and internal logging from day one — every Master generation, query resolution, and document edit was tracked end-to-end, giving the team a continuous signal on where governance broke and where users hesitated;
Instrumented the product with PostHog and internal logging from day one — every Master generation, query resolution, and document edit was tracked end-to-end, giving the team a continuous signal on where governance broke and where users hesitated;
→
→
→
Found that documents generated from a pre-built Master were accepted without manual editing 3.2x more often than documents generated from a one-off prompt — confirming the core thesis that reusable, parameterized templates outperform ad-hoc generation;
→
→
→
Session replays and agent query logs revealed that first-run accuracy mattered more than raw speed — users tolerated a slower governed query, but abandoned the workflow entirely after a single hallucinated metric;
Session replays and agent query logs revealed that first-run accuracy mattered more than raw speed — users tolerated a slower governed query, but abandoned the workflow entirely after a single hallucinated metric;
→
→
→
Implemented a measurement loop where every customer-facing release was paired with a corresponding agent-evaluation suite — UX changes that broke MCP determinism were caught before reaching customers.
Implemented a measurement loop where every customer-facing release was paired with a corresponding agent-evaluation suite — UX changes that broke MCP determinism were caught before reaching customers.
Enhancements Following SOC 2 Readiness:
Enhancements Following SOC 2 Readiness:
→
→
→
As Motley approached SOC 2 Type 2 readiness, the product surface was upgraded to make compliance a visible signal of trust rather than a buried legal page:
As Motley approached SOC 2 Type 2 readiness, the product surface was upgraded to make compliance a visible signal of trust rather than a buried legal page:
As Motley approached SOC 2 Type 2 readiness, the product surface was upgraded to make compliance a visible signal of trust rather than a buried legal page:
•
•
•
Per-workspace access controls with role-based permissions;
Per-workspace access controls with role-based permissions;
Per-workspace access controls with role-based permissions;
•
•
•
Data residency and self-host signaling in the UI;
Data residency and self-host signaling in the UI;
Data residency and self-host signaling in the UI;
•
•
•
Audit log surfaced directly inside every Document and Master.
Audit log surfaced directly inside every Document and Master.
Audit log surfaced directly inside every Document and Master.
→
→
→
These signals shortened the enterprise sales cycle — security review questions that previously triggered a back-and-forth with the team were answered by the interface itself.
These signals shortened the enterprise sales cycle — security review questions that previously triggered a back-and-forth with the team were answered by the interface itself.
These signals shortened the enterprise sales cycle — security review questions that previously triggered a back-and-forth with the team were answered by the interface itself.
User Research & Iterative Refinements:
User Research & Iterative Refinements:
→
→
→
Early prototypes leaned playful — emoji-heavy empty states, casual microcopy, animated state transitions modeled on consumer AI products;
Early prototypes leaned playful — emoji-heavy empty states, casual microcopy, animated state transitions modeled on consumer AI products;
→
→
→
Interviews with data engineering leads and CS heads revealed the opposite preference: the audience operating governed data expected an instrument, not a personality. Playful affordances eroded trust rather than building it;
Interviews with data engineering leads and CS heads revealed the opposite preference: the audience operating governed data expected an instrument, not a personality. Playful affordances eroded trust rather than building it;
→
→
→
Microcopy was rewritten in the voice of a senior data engineer — direct, specific, free of marketing softening. Error states explain which query failed and why, not we're sorry, something went wrong;
→
→
→
A/B tests confirmed that structured, technical language increased completion of the first Master setup by 22% among data-engineer users, with no measurable drop among business users.
Expanded Integrations for Increased Credibility:
Expanded Integrations for Increased Credibility:
→
→
→
Added named warehouse connectors (Snowflake, BigQuery, ClickHouse, Postgres, Databricks, DuckDB, dbt) as a first-class surface in onboarding — visibly demonstrating warehouse-agnostic positioning;
Added named warehouse connectors (Snowflake, BigQuery, ClickHouse, Postgres, Databricks, DuckDB, dbt) as a first-class surface in onboarding — visibly demonstrating warehouse-agnostic positioning;
→
→
→
Highlighted MCP-native compatibility with Claude, Cursor, and Codex in marketing surfaces and integration setup — reassuring prospects that Motley fits inside their existing agent stack rather than replacing it;
Highlighted MCP-native compatibility with Claude, Cursor, and Codex in marketing surfaces and integration setup — reassuring prospects that Motley fits inside their existing agent stack rather than replacing it;
→
→
→
Surfaced open-source provenance throughout the product — every governed query traces back to a SLayer definition, with a direct link to the underlying repo. Transparency about how the layer works became a credibility signal in itself.
Surfaced open-source provenance throughout the product — every governed query traces back to a SLayer definition, with a direct link to the underlying repo. Transparency about how the layer works became a credibility signal in itself.
Overall, the design evolution for Konvi reflects a balanced approach—melding a trust-inspiring visual identity with an engaging, user-centric experience. The carefully chosen color scheme, interactive elements, and strategic user flow all work together to create a platform that is both sophisticated and accessible.
Overall, the design evolution for Konvi reflects a balanced approach—melding a trust-inspiring visual identity with an engaging, user-centric experience. The carefully chosen color scheme, interactive elements, and strategic user flow all work together to create a platform that is both sophisticated and accessible.
Impact
Impact
Impact
Motley was built from zero — no legacy product, no prior interface, no existing users to migrate. The work was measured against three design-attributable outcomes: how fast users reached value, how much of what the product generated was actually usable, and how much trust the interface earned in sales conversations.
Motley was built from zero — no legacy product, no prior interface, no existing users to migrate. The work was measured against three design-attributable outcomes: how fast users reached value, how much of what the product generated was actually usable, and how much trust the interface earned in sales conversations.
Time to First Governed Query
Time to First Governed Query
→
→
→
New users reach a working governed query in under 30 minutes from sign-up — driven by surfacing Data Sources as the first onboarding step and structuring the Master editor around typed slide blocks mapped to the semantic layer, rather than free-form prompts.
New users reach a working governed query in under 30 minutes from sign-up — driven by surfacing Data Sources as the first onboarding step and structuring the Master editor around typed slide blocks mapped to the semantic layer, rather than free-form prompts.
Document Acceptance Rate
Document Acceptance Rate
→
→
→
87% of generated documents are shipped to end customers without manual editing — the metric the team optimized for above all others, since a generated report is only valuable if it can be sent without a human review pass;
87% of generated documents are shipped to end customers without manual editing — the metric the team optimized for above all others, since a generated report is only valuable if it can be sent without a human review pass;
→
→
→
The audit panel — showing which queries fired and which metrics resolved — was the single largest contributor, confirmed in qualitative interviews.
The audit panel — showing which queries fired and which metrics resolved — was the single largest contributor, confirmed in qualitative interviews.
Trust Signals in Sales & Support
Trust Signals in Sales & Support
→
→
→
Security review cycles close in the same week for design partners, attributed to surfacing audit trail, access controls, and data residency as first-class UI rather than buried legal copy
Security review cycles close in the same week for design partners, attributed to surfacing audit trail, access controls, and data residency as first-class UI rather than buried legal copy
→
→
→
Support tickets about "the AI gave the wrong number" stay near zero — users self-diagnose value differences against their dashboards via the audit panel, without escalating to the team.
Support tickets about "the AI gave the wrong number" stay near zero — users self-diagnose value differences against their dashboards via the audit panel, without escalating to the team.
The reframing matters more than any single number. Motley isn't described as another AI reporting tool, but as the layer agents query when the answer has to be defensible — a position the design earned by treating governance, audit, and metric integrity as primary UI, not secondary copy.
The reframing matters more than any single number. Motley isn't described as another AI reporting tool, but as the layer agents query when the answer has to be defensible — a position the design earned by treating governance, audit, and metric integrity as primary UI, not secondary copy.
Copyright © 2026 Monochromatic Ltd. All rights reserved.
Workhub, 77 Lower Camden Street, Dublin, D02 XE80, Ireland
Copyright © 2026 Monochromatic Ltd. All rights reserved.
Workhub, 77 Lower Camden Street, Dublin, D02 XE80, Ireland
Copyright © 2026 Monochromatic Ltd. All rights reserved.
Workhub, 77 Lower Camden Street, Dublin, D02 XE80, Ireland