FitFuel: Designed for Returns

FitFuel is a fictional AI-enhanced fitness app built as the capstone project for Design Lab's AI for UX Design course. I led this as a solo design exercise run to a real four-week production cadence: research direction, stakeholder research, a decision-ready design package, and this case study. My role covered research synthesis, persona and journey work, opportunity prioritization, one prototyped flow, and an accessibility audit, with Claude used throughout as a structured working partner rather than a content generator.

FitFuel: Designed for Returns

FitFuel is a fictional AI-enhanced fitness app built as the capstone project for Design Lab's AI for UX Design course. I led this as a solo design exercise run to a real four-week production cadence: research direction, stakeholder research, a decision-ready design package, and this case study. My role covered research synthesis, persona and journey work, opportunity prioritization, one prototyped flow, and an accessibility audit, with Claude used throughout as a structured working partner rather than a content generator.

FitFuel: Designed for Returns

FitFuel is a fictional AI-enhanced fitness app built as the capstone project for Design Lab's AI for UX Design course. I led this as a solo design exercise run to a real four-week production cadence: research direction, stakeholder research, a decision-ready design package, and this case study. My role covered research synthesis, persona and journey work, opportunity prioritization, one prototyped flow, and an accessibility audit, with Claude used throughout as a structured working partner rather than a content generator.

FitFuel: Designed for Returns

FitFuel is a fictional AI-enhanced fitness app built as the capstone project for Design Lab's AI for UX Design course. I led this as a solo design exercise run to a real four-week production cadence: research direction, stakeholder research, a decision-ready design package, and this case study. My role covered research synthesis, persona and journey work, opportunity prioritization, one prototyped flow, and an accessibility audit, with Claude used throughout as a structured working partner rather than a content generator.

The Goal

Feature Engagment = Downstream Impact

Fitness apps have two working models: cheap self-tracking tools that log what you did, or expensive human coaching that tells you what to do next. Neither is where FitFuel started. The open question was narrower and less comfortable: not "what feature should we build," but "who is actually underserved, and can we prove it before we design for them." Three persona hypotheses came out of the design brief: a time-constrained professional, a returning routine builder, and an edge-case user with access needs. Only two survived contact with the interview data. The professional persona was cut for thin evidentiary support, a finding I documented rather than quietly patched over. What the surviving research pointed to instead was sharper than any of the three original hypotheses: across the journey map, the single highest-risk moment for the Returning Routine Builder wasn't a missing feature anywhere in the app. It was how the product handled the act of coming back after a lapse. No competitor in the landscape scan designs for that specific moment. That gap, not a feature request, became the central design problem.

The Goal

Feature Engagment = Downstream Impact

Fitness apps have two working models: cheap self-tracking tools that log what you did, or expensive human coaching that tells you what to do next. Neither is where FitFuel started. The open question was narrower and less comfortable: not "what feature should we build," but "who is actually underserved, and can we prove it before we design for them." Three persona hypotheses came out of the design brief: a time-constrained professional, a returning routine builder, and an edge-case user with access needs. Only two survived contact with the interview data. The professional persona was cut for thin evidentiary support, a finding I documented rather than quietly patched over. What the surviving research pointed to instead was sharper than any of the three original hypotheses: across the journey map, the single highest-risk moment for the Returning Routine Builder wasn't a missing feature anywhere in the app. It was how the product handled the act of coming back after a lapse. No competitor in the landscape scan designs for that specific moment. That gap, not a feature request, became the central design problem.

The Goal

Feature Engagment = Downstream Impact

Fitness apps have two working models: cheap self-tracking tools that log what you did, or expensive human coaching that tells you what to do next. Neither is where FitFuel started. The open question was narrower and less comfortable: not "what feature should we build," but "who is actually underserved, and can we prove it before we design for them." Three persona hypotheses came out of the design brief: a time-constrained professional, a returning routine builder, and an edge-case user with access needs. Only two survived contact with the interview data. The professional persona was cut for thin evidentiary support, a finding I documented rather than quietly patched over. What the surviving research pointed to instead was sharper than any of the three original hypotheses: across the journey map, the single highest-risk moment for the Returning Routine Builder wasn't a missing feature anywhere in the app. It was how the product handled the act of coming back after a lapse. No competitor in the landscape scan designs for that specific moment. That gap, not a feature request, became the central design problem.

The Goal

Feature Engagment = Downstream Impact

Fitness apps have two working models: cheap self-tracking tools that log what you did, or expensive human coaching that tells you what to do next. Neither is where FitFuel started. The open question was narrower and less comfortable: not "what feature should we build," but "who is actually underserved, and can we prove it before we design for them." Three persona hypotheses came out of the design brief: a time-constrained professional, a returning routine builder, and an edge-case user with access needs. Only two survived contact with the interview data. The professional persona was cut for thin evidentiary support, a finding I documented rather than quietly patched over. What the surviving research pointed to instead was sharper than any of the three original hypotheses: across the journey map, the single highest-risk moment for the Returning Routine Builder wasn't a missing feature anywhere in the app. It was how the product handled the act of coming back after a lapse. No competitor in the landscape scan designs for that specific moment. That gap, not a feature request, became the central design problem.

Product Fit

Discovery Moments

I ran this as an AI-augmented workflow with a fixed division of labor: Claude for synthesis speed, breadth, and structure; me for framing, evidence interpretation, and every claim that shipped in a stakeholder-facing document. Concretely: Every fact-shaped claim (market, competitive, or behavioral) was tagged Validated, Partially Validated, Hypothesis, or Synthesis, and no claim was called Validated without two independent sources. This wasn't a formatting choice; it changed which opportunity areas survived. Two opportunity areas that looked promising on first pass got cut on the evidence, not on gut feel: an accountability middle tier, because the competitive scan challenged the premise it was solving for and the interview data showed aggressive accountability notifications as a documented deletion trigger rather than a retention lever; and a GLP-1 fitness bridge, for having zero primary-transcript support. Claude drafted first-pass syntheses, personas, and journey map content from interview transcripts and market research. I reviewed every synthesis-tagged conclusion against the source transcripts before it was allowed into a document, and re-framed at least one finding directly: research on AI trust showed users don't need algorithmic explainability, they need override ability and a default-to-gentler posture under uncertainty, which is a different (and more actionable) design target than "make the AI's reasoning legible."

Product Fit

Discovery Moments

I ran this as an AI-augmented workflow with a fixed division of labor: Claude for synthesis speed, breadth, and structure; me for framing, evidence interpretation, and every claim that shipped in a stakeholder-facing document. Concretely: Every fact-shaped claim (market, competitive, or behavioral) was tagged Validated, Partially Validated, Hypothesis, or Synthesis, and no claim was called Validated without two independent sources. This wasn't a formatting choice; it changed which opportunity areas survived. Two opportunity areas that looked promising on first pass got cut on the evidence, not on gut feel: an accountability middle tier, because the competitive scan challenged the premise it was solving for and the interview data showed aggressive accountability notifications as a documented deletion trigger rather than a retention lever; and a GLP-1 fitness bridge, for having zero primary-transcript support. Claude drafted first-pass syntheses, personas, and journey map content from interview transcripts and market research. I reviewed every synthesis-tagged conclusion against the source transcripts before it was allowed into a document, and re-framed at least one finding directly: research on AI trust showed users don't need algorithmic explainability, they need override ability and a default-to-gentler posture under uncertainty, which is a different (and more actionable) design target than "make the AI's reasoning legible."

Product Fit

Discovery Moments

I ran this as an AI-augmented workflow with a fixed division of labor: Claude for synthesis speed, breadth, and structure; me for framing, evidence interpretation, and every claim that shipped in a stakeholder-facing document. Concretely: Every fact-shaped claim (market, competitive, or behavioral) was tagged Validated, Partially Validated, Hypothesis, or Synthesis, and no claim was called Validated without two independent sources. This wasn't a formatting choice; it changed which opportunity areas survived. Two opportunity areas that looked promising on first pass got cut on the evidence, not on gut feel: an accountability middle tier, because the competitive scan challenged the premise it was solving for and the interview data showed aggressive accountability notifications as a documented deletion trigger rather than a retention lever; and a GLP-1 fitness bridge, for having zero primary-transcript support. Claude drafted first-pass syntheses, personas, and journey map content from interview transcripts and market research. I reviewed every synthesis-tagged conclusion against the source transcripts before it was allowed into a document, and re-framed at least one finding directly: research on AI trust showed users don't need algorithmic explainability, they need override ability and a default-to-gentler posture under uncertainty, which is a different (and more actionable) design target than "make the AI's reasoning legible."

Product Fit

Discovery Moments

I ran this as an AI-augmented workflow with a fixed division of labor: Claude for synthesis speed, breadth, and structure; me for framing, evidence interpretation, and every claim that shipped in a stakeholder-facing document. Concretely: Every fact-shaped claim (market, competitive, or behavioral) was tagged Validated, Partially Validated, Hypothesis, or Synthesis, and no claim was called Validated without two independent sources. This wasn't a formatting choice; it changed which opportunity areas survived. Two opportunity areas that looked promising on first pass got cut on the evidence, not on gut feel: an accountability middle tier, because the competitive scan challenged the premise it was solving for and the interview data showed aggressive accountability notifications as a documented deletion trigger rather than a retention lever; and a GLP-1 fitness bridge, for having zero primary-transcript support. Claude drafted first-pass syntheses, personas, and journey map content from interview transcripts and market research. I reviewed every synthesis-tagged conclusion against the source transcripts before it was allowed into a document, and re-framed at least one finding directly: research on AI trust showed users don't need algorithmic explainability, they need override ability and a default-to-gentler posture under uncertainty, which is a different (and more actionable) design target than "make the AI's reasoning legible."

Approach

Understanding the Pain Points

I ran a design sprint with users ranging from daily active to once a month, advanced to beginner. I brought the engineering team along to hear the feedback directly. As we walked through campaign creation, I had the participants explain their workflow, end goal, and how they thought Odyssey worked. This revealed a general misunderstanding of some of the UI. As the engineers clarified terms and features, light bulbs went off as participants began to understand. With valuable insights uncovered, we saw a clear path to improvement. The whole activity took about an hour and a half.

Approach

Understanding the Pain Points

I ran a design sprint with users ranging from daily active to once a month, advanced to beginner. I brought the engineering team along to hear the feedback directly. As we walked through campaign creation, I had the participants explain their workflow, end goal, and how they thought Odyssey worked. This revealed a general misunderstanding of some of the UI. As the engineers clarified terms and features, light bulbs went off as participants began to understand. With valuable insights uncovered, we saw a clear path to improvement. The whole activity took about an hour and a half.

Approach

Understanding the Pain Points

I ran a design sprint with users ranging from daily active to once a month, advanced to beginner. I brought the engineering team along to hear the feedback directly. As we walked through campaign creation, I had the participants explain their workflow, end goal, and how they thought Odyssey worked. This revealed a general misunderstanding of some of the UI. As the engineers clarified terms and features, light bulbs went off as participants began to understand. With valuable insights uncovered, we saw a clear path to improvement. The whole activity took about an hour and a half.

Approach

Understanding the Pain Points

I ran a design sprint with users ranging from daily active to once a month, advanced to beginner. I brought the engineering team along to hear the feedback directly. As we walked through campaign creation, I had the participants explain their workflow, end goal, and how they thought Odyssey worked. This revealed a general misunderstanding of some of the UI. As the engineers clarified terms and features, light bulbs went off as participants began to understand. With valuable insights uncovered, we saw a clear path to improvement. The whole activity took about an hour and a half.

Design Process

Simplifying Authoring

Every field and dropdown in the UI was needed, so cutting options to simplify would break live campaigns. Also, I couldn't reshuffle the layout, forcing advanced users to relearn it. So rather than revamping the original creation path entirely, I decided to put a simplified version alongside it as an easier means to the same end. For the CIF channel (aka "by the way, have you heard of {feature name}?"), this alternate was called Basic CIF, which was stripped of all optional fields. An unexpected benefit of this approach: new authors could use this path as training wheels.

Design Process

Simplifying Authoring

Every field and dropdown in the UI was needed, so cutting options to simplify would break live campaigns. Also, I couldn't reshuffle the layout, forcing advanced users to relearn it. So rather than revamping the original creation path entirely, I decided to put a simplified version alongside it as an easier means to the same end. For the CIF channel (aka "by the way, have you heard of {feature name}?"), this alternate was called Basic CIF, which was stripped of all optional fields. An unexpected benefit of this approach: new authors could use this path as training wheels.

Design Process

Simplifying Authoring

Every field and dropdown in the UI was needed, so cutting options to simplify would break live campaigns. Also, I couldn't reshuffle the layout, forcing advanced users to relearn it. So rather than revamping the original creation path entirely, I decided to put a simplified version alongside it as an easier means to the same end. For the CIF channel (aka "by the way, have you heard of {feature name}?"), this alternate was called Basic CIF, which was stripped of all optional fields. An unexpected benefit of this approach: new authors could use this path as training wheels.

Design Process

Simplifying Authoring

Every field and dropdown in the UI was needed, so cutting options to simplify would break live campaigns. Also, I couldn't reshuffle the layout, forcing advanced users to relearn it. So rather than revamping the original creation path entirely, I decided to put a simplified version alongside it as an easier means to the same end. For the CIF channel (aka "by the way, have you heard of {feature name}?"), this alternate was called Basic CIF, which was stripped of all optional fields. An unexpected benefit of this approach: new authors could use this path as training wheels.

Advanced Authoring

For advanced authoring, I was limited by what I could touch, so as not to break existing experiences. Even so, I used a handful of design tactics to melt away confusion across the journey. I used chunking to make the number of fields seem less overwhelming. I suggested autosave, allowing users to walk away at any time. I also audited copy to remove engineer-y terms and make the language more consistent.

Choose Campaign Type

Choose Campaign Type

Configure Triggering Mechanism

Configure Triggering Mechanism

Select User Target

Select User Target

Author the Experience

Author the Experience

Choose Campaign Type

Choose Campaign Type

Configure Triggering Mechanism

Configure Triggering Mechanism

Select User Target

Select User Target

Author the Experience

Author the Experience

Choose Campaign Type

Choose Campaign Type

Configure Triggering Mechanism

Configure Triggering Mechanism

Select User Target

Select User Target

Author the Experience

Author the Experience

Choose Campaign Type

Choose Campaign Type

Configure Triggering Mechanism

Configure Triggering Mechanism

Select User Target

Select User Target

Author the Experience

Author the Experience

Process Continued

Further Improvements

Automating

The workflow UI was very form-field intensive. Instead of showing a sea of empty fields each time a campaign was created, pre-loading common inputs delivered meaningful time savings.

Templates

Marketers struggled with too many style options. Instead of making them configure everything manually, I defined a set of simple templates with pre-loaded content tailored to common goals.

Copy Audit

Copy throughout the tool was filled with developer-speak. I replaced these with plain language and clear questions, so marketers quickly understood what the tool does and what results to expect.

Process Continued

Further Improvements

Automating

The workflow UI was very form-field intensive. Instead of showing a sea of empty fields each time a campaign was created, pre-loading common inputs delivered meaningful time savings.

Templates

Marketers struggled with too many style options. Instead of making them configure everything manually, I defined a set of simple templates with pre-loaded content tailored to common goals.

Copy Audit

Copy throughout the tool was filled with developer-speak. I replaced these with plain language and clear questions, so marketers quickly understood what the tool does and what results to expect.

Process Continued

Further Improvements

Automating

The workflow UI was very form-field intensive. Instead of showing a sea of empty fields each time a campaign was created, pre-loading common inputs delivered meaningful time savings.

Templates

Marketers struggled with too many style options. Instead of making them configure everything manually, I defined a set of simple templates with pre-loaded content tailored to common goals.

Copy Audit

Copy throughout the tool was filled with developer-speak. I replaced these with plain language and clear questions, so marketers quickly understood what the tool does and what results to expect.

Process Continued

Further Improvements

Automating

The workflow UI was very form-field intensive. Instead of showing a sea of empty fields each time a campaign was created, pre-loading common inputs delivered meaningful time savings.

Templates

Marketers struggled with too many style options. Instead of making them configure everything manually, I defined a set of simple templates with pre-loaded content tailored to common goals.

Copy Audit

Copy throughout the tool was filled with developer-speak. I replaced these with plain language and clear questions, so marketers quickly understood what the tool does and what results to expect.

Impact

These internal tools at Amazon that drove a 160% increase in global campaigns, 208% YoY growth, and cut campaign creation time from months to as little as 2 weeks.

Impact

These internal tools at Amazon that drove a 160% increase in global campaigns, 208% YoY growth, and cut campaign creation time from months to as little as 2 weeks.

Impact

These internal tools at Amazon that drove a 160% increase in global campaigns, 208% YoY growth, and cut campaign creation time from months to as little as 2 weeks.

Copyright © 2025 Jesse Bones. All Rights Reserved. All content, designs, and images on this website are the property of Jesse Bones and respective clients. Unauthorized use, reproduction, or distribution of any materials without explicit permission is prohibited.

Copyright © 2025 Jesse Bones. All Rights Reserved. All content, designs, and images on this website are the property of Jesse Bones and respective clients. Unauthorized use, reproduction, or distribution of any materials without explicit permission is prohibited.

Copyright © 2025 Jesse Bones. All Rights Reserved. All content, designs, and images on this website are the property of Jesse Bones and respective clients. Unauthorized use, reproduction, or distribution of any materials without explicit permission is prohibited.

Copyright © 2025 Jesse Bones. All Rights Reserved. All content, designs, and images on this website are the property of Jesse Bones and respective clients. Unauthorized use, reproduction, or distribution of any materials without explicit permission is prohibited.