UI / UX Design
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
New customers waited days for manual setup before seeing any value. I designed a self serve onboarding flow powered by AI co creation reducing time to value from days to under 10 minutes.
Year :
2025
2025
Industry :
Finance
Finance
Client :
Evergrowth
Evergrowth
Project Duration :
4 weeks
Introduction
Introduction
Evergrowth is a B2B AI platform that automates sales research, account enrichment, and personalised outreach at scale. As a freelance designer working closely with engineering and PM, I joined at a critical moment: the product was moving from a manual, high touch tool to a fully scalable AI platform. The core problem was trust, not features. Users were left guessing how the AI made decisions with no sources, no reasoning, no feedback. The onboarding was a fragmented wizard that demanded effort before delivering any value. Before I could fix the experience, I needed to fix three things:
Blank Page Paralysis : users had to define complex strategy from scratch with zero AI scaffolding. The Black Box : the AI processed data silently, killing confidence at the exact moment it needed to be built. Fragmented Mental Models : every team had a different understanding of how Agents worked, causing inconsistencies across screens.
My goal was to fix all three within five core screens, building something that felt intelligent and transparent from the very first click.
My key objectives were to:
Reduce Time to Value (TTV): Cut onboarding time from hours to under 10 minutes.
Build Trust: Move from a "Black Box" AI to a transparent system where users can see and verify data sources.
Scale Self Serve: Eliminate the need for customer success managers to hand hold new users.
The starting point: original product (2024)
Research & Discovery
Research & Discovery
I mapped the end to end journey and ran interviews with existing users. Three patterns came up every time: users couldn't see where the AI's data came from, the setup asked for expert strategy before delivering any value, and the agent kept asking for information it should have already known from the URL scrape.
A Revenue Operations Manager summed it up best during testing: "I didn't know if the AI was still running or had just broken." That one line became the brief for the entire State Matrix.
Re-architecting the setup flow into a 3 layer system: Shell, Agentic Brain, and Artifact Stage.
Ideation & Usability tetsing
Ideation & Usability tetsing
Before landing on the split pane model, I explored a pure chat interface and an improved wizard. Chat tested poorly users lost track of decisions. The wizard was too close to the original problem. The split pane won because it separated process from result, giving users control and visibility at the same time. To bridge the logic gaps identified during the discovery phase, I re architected the platform around a split pane mental model. This allowed the "Chat Shell" and the "Artifact Stage" to work in together. By separating the conversational instructions from the structured strategic data, we created a shared source of truth that was missing in the legacy version.
This new architecture introduced a much clearer hierarchy and a more efficient way to "steer" the AI:
Persistent Artifact Stage: A dedicated right-hand pane that keeps strategic business models visible, even as the chat scrolls.
Sequential Logic: A 3 step breadcrumb system that prevents users from feeling lost in complex agent configurations.
Dynamic Visibility: Using focused side panels for Refinement tasks rather than jumping to new pages.
Evidence Linked Cards: Every AI generated output is now anchored to a specific source link, ensuring the data is auditable at a glance.
These changes reduced cognitive load, removed the Black Box frustration, and made the transition from Setup to Active Growth Plays noticeably smoother for users.
Watching real users complete onboarding revealed three things we changed immediately:
Users couldn't tell if the AI was working. Silent processing felt identical to a crash. This directly led to the Thinking, Success, Low Confidence, and Error state system.
Users questioned every AI output. "Where did this come from?" came up in every session. This made source citations non-negotiable, not a nice-to-have.
Users lost their place mid-setup. Switching between steps broke context entirely. This validated the split-pane chat on the left, artifact always visible on the right.
Figma design process walkthrough
Workspace Experience
Workspace Experience
The refined workspace experience focused on turning raw AI processing into a manageable daily workflow. By introducing easier to scan tables with status colored agent states (Idle, Thinking, Success) and an improved filtering system for Leads and Personas, users could finally manage their AI workforce at scale. The redesigned Growth Play creation flow using a focused side panel allowed users to tweak agent logic without losing their place in the dashboard. This transformed the platform from a one time setup tool into a reliable daily command center.
Old vs new evergrowth agentic platform
Redesigned platform workflow for clarity and speed
Client Experience - SaaS First Redesign
Client Experience - SaaS First Redesign
Our users are growth leaders who need to move quickly without losing strategic depth. To fix the Black Box issues identified in my initial mapping, I shifted the experience from a simple chatbot to a Strategic Command Center. This layout ensures that as the AI thinks and fetches, the client always feels in total control of the output.
The redesign:
Organizes AI outputs into Artifacts: Instead of losing data in a scrolling chat, complex outputs like Business Models are now clean, expandable cards that stay anchored on the right.
Refined the Workspace Hierarchy: I used a split-pane layout so users can talk to their data. This allows for real-time refinement without ever losing the context of the strategic draft.
Prioritizes Next-Step Actions: High leverage actions like Approve Segment or Launch Play are placed consistently at the bottom of the artifact to maintain momentum.
Reduces Cognitive Noise: I stripped away the unnecessary sidebars and navigation during onboarding, focusing the client entirely on one strategic decision at a time.
Every decision was validated with engineering before going high fidelity. API response times and scraper limitations directly shaped the interaction model the agent narrating its progress, for example, exists because URL scraping takes 8 to 15 seconds and a static loader would feel broken.
Redesigned Agentic training center screens
Design Strategy (The Agentic Philosophy)
Design Strategy (The Agentic Philosophy)
Designing Evergrowth wasn’t just about the UI it was about defining how humans and AI work together. I moved away from the Magic Button (black box) approach and toward an Agentic Philosophy: the AI acts as a junior analyst doing the heavy lifting, while the user stays in control as the Editor in Chief.
I focused on three main pillars:
1. The Split Screen Thinking Model
Standard chatbots bury results in a long scrolling history. I introduced a split layout to separate the Process from the Result.
Left Side (The Chat): Where you give instructions and see status updates.
Right Side (The Artifact): A persistent canvas where the actual work (Business Models, Personas) lives.
The Why: It keeps the Work in Progress visible so the user never feels lost.
2. Human in the Loop (HITL) Intervention
The AI never makes the final call. Every output is presented as a Draft that the user can hover over to Refine, Edit, or Delete.
The Why: It’s much less intimidating to fix a 90% complete draft than to start from a blank page. This lowers anxiety about AI mistakes.
3. Trust through Traceability
Trust comes from honesty, not perfection. I added Source Citations throughout the interface. If the AI identifies a target audience, it provides a direct link to the specific webpage or document where it found that info.
The Why: This moves the user from "blindly trusting" to "verifying." It turns the product into a professional tool for experts.
Evergrowth journey mapped end-to-end
The 3 Step Journey
The 3 Step Journey
I broke the onboarding into three clear milestones. This transformed a daunting setup into a guided ladder of value.
Step 1: In testing, users described this as the first moment they felt the product was genuinely intelligent.
Step 2: A 90% complete draft is far less intimidating than a blank page.
Step 3: This ensures agents are pointed in exactly the right direction before any automated activity begins.
The 3 Step Journey
The State Matrix
The State Matrix
In an agentic system, the UI needs to feel alive so users don't think it’s broken while the AI is thinking. I designed a matrix of states to manage this communication
Thinking: A subtle, high energy pulse that signals the agent is processing.
Success: A clear green state that confirms the data is saved and learned.
Low Confidence (The Warning): An amber highlight that tells the user: I’m not 100% sure about this fact please double check it.
Error: A friendly recovery state that explains why a fetch failed (e.g., a firewall) and offers a workaround.
Technical Considerations (Prompt Snippets)
I didn't just design the UI; I designed the logic behind the agent. By writing the System Prompts, I controlled how the AI talks and where it looks for data.
Snippet Example: "Act as a RevOps Expert. Scrape URL but ignore generic 'About Us' fluff. Focus on pricing tiers and 'Jobs to be Done.' If data is missing, ask the user do not guess."
The Result is by setting these guardrails, we cut down hallucinations and made sure the AI felt like a pro, not a chatbot.
Visual state indicators that keep the user informed during asynchronous AI processing.
Impact & Outcomes
Impact & Outcomes
The redesign turned a high friction process into a growth engine. By focusing on trust and speed, I saw
Time to Value : setup dropped from days to under 10 minutes.
Trust : 75% of users clicked Source links, making transparency a measurable retention signal.
Scale : 200+ organisations onboarded, £10m+ volume processed, 20,000+ active users reached without CS intervention.
Market : contributed to ~10% market share in target segment.
A strategic command center giving sales leaders a bird's eye view of their AI workforce
Final Agentic Platform experience
Reflection: Systems vs. Screens
Reflection: Systems vs. Screens
This project taught me that in agentic systems, transparency is a feature. The most important thing I designed wasn't a screen it was an edit loop. Users don't want to be replaced by AI, they want to be amplified by it. If I had more time, I'd have tracked whether users who completed all three onboarding steps showed better 30 day retention that data would have sharpened the business case considerably.
More Projects
UI / UX Design
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
New customers waited days for manual setup before seeing any value. I designed a self serve onboarding flow powered by AI co creation reducing time to value from days to under 10 minutes.
Year :
2025
2025
Industry :
Finance
Finance
Client :
Evergrowth
Evergrowth
Project Duration :
4 weeks
Introduction
Introduction
Evergrowth is a B2B AI platform that automates sales research, account enrichment, and personalised outreach at scale. As a freelance designer working closely with engineering and PM, I joined at a critical moment: the product was moving from a manual, high touch tool to a fully scalable AI platform. The core problem was trust, not features. Users were left guessing how the AI made decisions with no sources, no reasoning, no feedback. The onboarding was a fragmented wizard that demanded effort before delivering any value. Before I could fix the experience, I needed to fix three things:
Blank Page Paralysis : users had to define complex strategy from scratch with zero AI scaffolding. The Black Box : the AI processed data silently, killing confidence at the exact moment it needed to be built. Fragmented Mental Models : every team had a different understanding of how Agents worked, causing inconsistencies across screens.
My goal was to fix all three within five core screens, building something that felt intelligent and transparent from the very first click.
My key objectives were to:
Reduce Time to Value (TTV): Cut onboarding time from hours to under 10 minutes.
Build Trust: Move from a "Black Box" AI to a transparent system where users can see and verify data sources.
Scale Self Serve: Eliminate the need for customer success managers to hand hold new users.
The starting point: original product (2024)
Research & Discovery
Research & Discovery
I mapped the end to end journey and ran interviews with existing users. Three patterns came up every time: users couldn't see where the AI's data came from, the setup asked for expert strategy before delivering any value, and the agent kept asking for information it should have already known from the URL scrape.
A Revenue Operations Manager summed it up best during testing: "I didn't know if the AI was still running or had just broken." That one line became the brief for the entire State Matrix.
Re-architecting the setup flow into a 3 layer system: Shell, Agentic Brain, and Artifact Stage.
Ideation & Usability tetsing
Ideation & Usability tetsing
Before landing on the split pane model, I explored a pure chat interface and an improved wizard. Chat tested poorly users lost track of decisions. The wizard was too close to the original problem. The split pane won because it separated process from result, giving users control and visibility at the same time. To bridge the logic gaps identified during the discovery phase, I re architected the platform around a split pane mental model. This allowed the "Chat Shell" and the "Artifact Stage" to work in together. By separating the conversational instructions from the structured strategic data, we created a shared source of truth that was missing in the legacy version.
This new architecture introduced a much clearer hierarchy and a more efficient way to "steer" the AI:
Persistent Artifact Stage: A dedicated right-hand pane that keeps strategic business models visible, even as the chat scrolls.
Sequential Logic: A 3 step breadcrumb system that prevents users from feeling lost in complex agent configurations.
Dynamic Visibility: Using focused side panels for Refinement tasks rather than jumping to new pages.
Evidence Linked Cards: Every AI generated output is now anchored to a specific source link, ensuring the data is auditable at a glance.
These changes reduced cognitive load, removed the Black Box frustration, and made the transition from Setup to Active Growth Plays noticeably smoother for users.
Watching real users complete onboarding revealed three things we changed immediately:
Users couldn't tell if the AI was working. Silent processing felt identical to a crash. This directly led to the Thinking, Success, Low Confidence, and Error state system.
Users questioned every AI output. "Where did this come from?" came up in every session. This made source citations non-negotiable, not a nice-to-have.
Users lost their place mid-setup. Switching between steps broke context entirely. This validated the split-pane chat on the left, artifact always visible on the right.
Figma design process walkthrough
Workspace Experience
Workspace Experience
The refined workspace experience focused on turning raw AI processing into a manageable daily workflow. By introducing easier to scan tables with status colored agent states (Idle, Thinking, Success) and an improved filtering system for Leads and Personas, users could finally manage their AI workforce at scale. The redesigned Growth Play creation flow using a focused side panel allowed users to tweak agent logic without losing their place in the dashboard. This transformed the platform from a one time setup tool into a reliable daily command center.
Old vs new evergrowth agentic platform
Redesigned platform workflow for clarity and speed
Client Experience - SaaS First Redesign
Client Experience - SaaS First Redesign
Our users are growth leaders who need to move quickly without losing strategic depth. To fix the Black Box issues identified in my initial mapping, I shifted the experience from a simple chatbot to a Strategic Command Center. This layout ensures that as the AI thinks and fetches, the client always feels in total control of the output.
The redesign:
Organizes AI outputs into Artifacts: Instead of losing data in a scrolling chat, complex outputs like Business Models are now clean, expandable cards that stay anchored on the right.
Refined the Workspace Hierarchy: I used a split-pane layout so users can talk to their data. This allows for real-time refinement without ever losing the context of the strategic draft.
Prioritizes Next-Step Actions: High leverage actions like Approve Segment or Launch Play are placed consistently at the bottom of the artifact to maintain momentum.
Reduces Cognitive Noise: I stripped away the unnecessary sidebars and navigation during onboarding, focusing the client entirely on one strategic decision at a time.
Every decision was validated with engineering before going high fidelity. API response times and scraper limitations directly shaped the interaction model the agent narrating its progress, for example, exists because URL scraping takes 8 to 15 seconds and a static loader would feel broken.
Redesigned Agentic training center screens
Design Strategy (The Agentic Philosophy)
Design Strategy (The Agentic Philosophy)
Designing Evergrowth wasn’t just about the UI it was about defining how humans and AI work together. I moved away from the Magic Button (black box) approach and toward an Agentic Philosophy: the AI acts as a junior analyst doing the heavy lifting, while the user stays in control as the Editor in Chief.
I focused on three main pillars:
1. The Split Screen Thinking Model
Standard chatbots bury results in a long scrolling history. I introduced a split layout to separate the Process from the Result.
Left Side (The Chat): Where you give instructions and see status updates.
Right Side (The Artifact): A persistent canvas where the actual work (Business Models, Personas) lives.
The Why: It keeps the Work in Progress visible so the user never feels lost.
2. Human in the Loop (HITL) Intervention
The AI never makes the final call. Every output is presented as a Draft that the user can hover over to Refine, Edit, or Delete.
The Why: It’s much less intimidating to fix a 90% complete draft than to start from a blank page. This lowers anxiety about AI mistakes.
3. Trust through Traceability
Trust comes from honesty, not perfection. I added Source Citations throughout the interface. If the AI identifies a target audience, it provides a direct link to the specific webpage or document where it found that info.
The Why: This moves the user from "blindly trusting" to "verifying." It turns the product into a professional tool for experts.
Evergrowth journey mapped end-to-end
The 3 Step Journey
The 3 Step Journey
I broke the onboarding into three clear milestones. This transformed a daunting setup into a guided ladder of value.
Step 1: In testing, users described this as the first moment they felt the product was genuinely intelligent.
Step 2: A 90% complete draft is far less intimidating than a blank page.
Step 3: This ensures agents are pointed in exactly the right direction before any automated activity begins.
The 3 Step Journey
The State Matrix
The State Matrix
In an agentic system, the UI needs to feel alive so users don't think it’s broken while the AI is thinking. I designed a matrix of states to manage this communication
Thinking: A subtle, high energy pulse that signals the agent is processing.
Success: A clear green state that confirms the data is saved and learned.
Low Confidence (The Warning): An amber highlight that tells the user: I’m not 100% sure about this fact please double check it.
Error: A friendly recovery state that explains why a fetch failed (e.g., a firewall) and offers a workaround.
Technical Considerations (Prompt Snippets)
I didn't just design the UI; I designed the logic behind the agent. By writing the System Prompts, I controlled how the AI talks and where it looks for data.
Snippet Example: "Act as a RevOps Expert. Scrape URL but ignore generic 'About Us' fluff. Focus on pricing tiers and 'Jobs to be Done.' If data is missing, ask the user do not guess."
The Result is by setting these guardrails, we cut down hallucinations and made sure the AI felt like a pro, not a chatbot.
Visual state indicators that keep the user informed during asynchronous AI processing.
Impact & Outcomes
Impact & Outcomes
The redesign turned a high friction process into a growth engine. By focusing on trust and speed, I saw
Time to Value : setup dropped from days to under 10 minutes.
Trust : 75% of users clicked Source links, making transparency a measurable retention signal.
Scale : 200+ organisations onboarded, £10m+ volume processed, 20,000+ active users reached without CS intervention.
Market : contributed to ~10% market share in target segment.
A strategic command center giving sales leaders a bird's eye view of their AI workforce
Final Agentic Platform experience
Reflection: Systems vs. Screens
Reflection: Systems vs. Screens
This project taught me that in agentic systems, transparency is a feature. The most important thing I designed wasn't a screen it was an edit loop. Users don't want to be replaced by AI, they want to be amplified by it. If I had more time, I'd have tracked whether users who completed all three onboarding steps showed better 30 day retention that data would have sharpened the business case considerably.
More Projects
UI / UX Design
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
Evergrowth: Designing an AI Onboarding That Turns Setup into Strategy
New customers waited days for manual setup before seeing any value. I designed a self serve onboarding flow powered by AI co creation reducing time to value from days to under 10 minutes.
Year :
2025
2025
Industry :
Finance
Finance
Client :
Evergrowth
Evergrowth
Project Duration :
4 weeks
Introduction
Introduction
Evergrowth is a B2B AI platform that automates sales research, account enrichment, and personalised outreach at scale. As a freelance designer working closely with engineering and PM, I joined at a critical moment: the product was moving from a manual, high touch tool to a fully scalable AI platform. The core problem was trust, not features. Users were left guessing how the AI made decisions with no sources, no reasoning, no feedback. The onboarding was a fragmented wizard that demanded effort before delivering any value. Before I could fix the experience, I needed to fix three things:
Blank Page Paralysis : users had to define complex strategy from scratch with zero AI scaffolding. The Black Box : the AI processed data silently, killing confidence at the exact moment it needed to be built. Fragmented Mental Models : every team had a different understanding of how Agents worked, causing inconsistencies across screens.
My goal was to fix all three within five core screens, building something that felt intelligent and transparent from the very first click.
My key objectives were to:
Reduce Time to Value (TTV): Cut onboarding time from hours to under 10 minutes.
Build Trust: Move from a "Black Box" AI to a transparent system where users can see and verify data sources.
Scale Self Serve: Eliminate the need for customer success managers to hand hold new users.
The starting point: original product (2024)
Research & Discovery
Research & Discovery
I mapped the end to end journey and ran interviews with existing users. Three patterns came up every time: users couldn't see where the AI's data came from, the setup asked for expert strategy before delivering any value, and the agent kept asking for information it should have already known from the URL scrape.
A Revenue Operations Manager summed it up best during testing: "I didn't know if the AI was still running or had just broken." That one line became the brief for the entire State Matrix.
Re-architecting the setup flow into a 3 layer system: Shell, Agentic Brain, and Artifact Stage.
Ideation & Usability tetsing
Ideation & Usability tetsing
Before landing on the split pane model, I explored a pure chat interface and an improved wizard. Chat tested poorly users lost track of decisions. The wizard was too close to the original problem. The split pane won because it separated process from result, giving users control and visibility at the same time. To bridge the logic gaps identified during the discovery phase, I re architected the platform around a split pane mental model. This allowed the "Chat Shell" and the "Artifact Stage" to work in together. By separating the conversational instructions from the structured strategic data, we created a shared source of truth that was missing in the legacy version.
This new architecture introduced a much clearer hierarchy and a more efficient way to "steer" the AI:
Persistent Artifact Stage: A dedicated right-hand pane that keeps strategic business models visible, even as the chat scrolls.
Sequential Logic: A 3 step breadcrumb system that prevents users from feeling lost in complex agent configurations.
Dynamic Visibility: Using focused side panels for Refinement tasks rather than jumping to new pages.
Evidence Linked Cards: Every AI generated output is now anchored to a specific source link, ensuring the data is auditable at a glance.
These changes reduced cognitive load, removed the Black Box frustration, and made the transition from Setup to Active Growth Plays noticeably smoother for users.
Watching real users complete onboarding revealed three things we changed immediately:
Users couldn't tell if the AI was working. Silent processing felt identical to a crash. This directly led to the Thinking, Success, Low Confidence, and Error state system.
Users questioned every AI output. "Where did this come from?" came up in every session. This made source citations non-negotiable, not a nice-to-have.
Users lost their place mid-setup. Switching between steps broke context entirely. This validated the split-pane chat on the left, artifact always visible on the right.
Figma design process walkthrough
Workspace Experience
Workspace Experience
The refined workspace experience focused on turning raw AI processing into a manageable daily workflow. By introducing easier to scan tables with status colored agent states (Idle, Thinking, Success) and an improved filtering system for Leads and Personas, users could finally manage their AI workforce at scale. The redesigned Growth Play creation flow using a focused side panel allowed users to tweak agent logic without losing their place in the dashboard. This transformed the platform from a one time setup tool into a reliable daily command center.
Old vs new evergrowth agentic platform
Redesigned platform workflow for clarity and speed
Client Experience - SaaS First Redesign
Client Experience - SaaS First Redesign
Our users are growth leaders who need to move quickly without losing strategic depth. To fix the Black Box issues identified in my initial mapping, I shifted the experience from a simple chatbot to a Strategic Command Center. This layout ensures that as the AI thinks and fetches, the client always feels in total control of the output.
The redesign:
Organizes AI outputs into Artifacts: Instead of losing data in a scrolling chat, complex outputs like Business Models are now clean, expandable cards that stay anchored on the right.
Refined the Workspace Hierarchy: I used a split-pane layout so users can talk to their data. This allows for real-time refinement without ever losing the context of the strategic draft.
Prioritizes Next-Step Actions: High leverage actions like Approve Segment or Launch Play are placed consistently at the bottom of the artifact to maintain momentum.
Reduces Cognitive Noise: I stripped away the unnecessary sidebars and navigation during onboarding, focusing the client entirely on one strategic decision at a time.
Every decision was validated with engineering before going high fidelity. API response times and scraper limitations directly shaped the interaction model the agent narrating its progress, for example, exists because URL scraping takes 8 to 15 seconds and a static loader would feel broken.
Redesigned Agentic training center screens
Design Strategy (The Agentic Philosophy)
Design Strategy (The Agentic Philosophy)
Designing Evergrowth wasn’t just about the UI it was about defining how humans and AI work together. I moved away from the Magic Button (black box) approach and toward an Agentic Philosophy: the AI acts as a junior analyst doing the heavy lifting, while the user stays in control as the Editor in Chief.
I focused on three main pillars:
1. The Split Screen Thinking Model
Standard chatbots bury results in a long scrolling history. I introduced a split layout to separate the Process from the Result.
Left Side (The Chat): Where you give instructions and see status updates.
Right Side (The Artifact): A persistent canvas where the actual work (Business Models, Personas) lives.
The Why: It keeps the Work in Progress visible so the user never feels lost.
2. Human in the Loop (HITL) Intervention
The AI never makes the final call. Every output is presented as a Draft that the user can hover over to Refine, Edit, or Delete.
The Why: It’s much less intimidating to fix a 90% complete draft than to start from a blank page. This lowers anxiety about AI mistakes.
3. Trust through Traceability
Trust comes from honesty, not perfection. I added Source Citations throughout the interface. If the AI identifies a target audience, it provides a direct link to the specific webpage or document where it found that info.
The Why: This moves the user from "blindly trusting" to "verifying." It turns the product into a professional tool for experts.
Evergrowth journey mapped end-to-end
The 3 Step Journey
The 3 Step Journey
I broke the onboarding into three clear milestones. This transformed a daunting setup into a guided ladder of value.
Step 1: In testing, users described this as the first moment they felt the product was genuinely intelligent.
Step 2: A 90% complete draft is far less intimidating than a blank page.
Step 3: This ensures agents are pointed in exactly the right direction before any automated activity begins.
The 3 Step Journey
The State Matrix
The State Matrix
In an agentic system, the UI needs to feel alive so users don't think it’s broken while the AI is thinking. I designed a matrix of states to manage this communication
Thinking: A subtle, high energy pulse that signals the agent is processing.
Success: A clear green state that confirms the data is saved and learned.
Low Confidence (The Warning): An amber highlight that tells the user: I’m not 100% sure about this fact please double check it.
Error: A friendly recovery state that explains why a fetch failed (e.g., a firewall) and offers a workaround.
Technical Considerations (Prompt Snippets)
I didn't just design the UI; I designed the logic behind the agent. By writing the System Prompts, I controlled how the AI talks and where it looks for data.
Snippet Example: "Act as a RevOps Expert. Scrape URL but ignore generic 'About Us' fluff. Focus on pricing tiers and 'Jobs to be Done.' If data is missing, ask the user do not guess."
The Result is by setting these guardrails, we cut down hallucinations and made sure the AI felt like a pro, not a chatbot.
Visual state indicators that keep the user informed during asynchronous AI processing.
Impact & Outcomes
Impact & Outcomes
The redesign turned a high friction process into a growth engine. By focusing on trust and speed, I saw
Time to Value : setup dropped from days to under 10 minutes.
Trust : 75% of users clicked Source links, making transparency a measurable retention signal.
Scale : 200+ organisations onboarded, £10m+ volume processed, 20,000+ active users reached without CS intervention.
Market : contributed to ~10% market share in target segment.
A strategic command center giving sales leaders a bird's eye view of their AI workforce
Final Agentic Platform experience
Reflection: Systems vs. Screens
Reflection: Systems vs. Screens
This project taught me that in agentic systems, transparency is a feature. The most important thing I designed wasn't a screen it was an edit loop. Users don't want to be replaced by AI, they want to be amplified by it. If I had more time, I'd have tracked whether users who completed all three onboarding steps showed better 30 day retention that data would have sharpened the business case considerably.