How I leveraged AI as a strategic co-pilot to deliver a complete product blueprint—without compromising on ownership

In the competitive world of iGaming affiliates, a casino review page is often the end of the user journey, not the beginning of discovery. This case study tackles that exact challenge: how to design a Dynamic Recommendation Engine that keeps users engaged by intelligently suggesting “what to see next.”

As a Product Owner, I was tasked with creating a feature from the ground up: defining the problem, designing the logic, wireframing the UX, and outlining the full Agile roadmap for an MVP. This article details my end-to-end process, with a particular focus on how I leveraged Artificial Intelligence as a strategic co-pilot to enhance research, stress-test ideas, and accelerate prototyping—without compromising on strategic ownership.

📥 Download the Full Deliverables

Full 23-Page Project Document – Complete PRD with problem statement, recommendation logic, user stories, MVP roadmap, and risk assessment
Interactive HTML Prototype – Mobile-first working demo with horizontal scroll interface

---

AI as Co-Pilot, Not Author

Let me be transparent upfront: this project was AI-augmented, not AI-generated.

I used multiple AI tools (Claude, ChatGPT, DeepSeek, and Gemini) across a week of intensive work, through ten major iterations of refinement. But here’s what that actually means:

What AI did for me:

• Structured competitive research findings into comparison frameworks
• Provided industry benchmark data and market context
• Generated technical pseudo-code for similarity algorithms
• Built the responsive HTML/CSS prototype from my wireframe specifications
• Helped calculate risk scenarios and mitigation strategies

What I owned entirely:

• The problem diagnosis and market opportunity identification
• All strategic decisions (what to build, what to exclude, and why)
• The two-stage hybrid recommendation logic design
• MVP scoping and kill metric definitions
• SEO/performance constraint balancing
• Every trade-off, every priority call, every “no”

The AI didn’t design the product. I did. The AI helped me execute faster and think more comprehensively—like having a research assistant, a technical reviewer, and a documentation specialist available 24/7.

---

The Assignment: 10 Deliverables, 4-6 Week Timeline

The brief from the client was comprehensive and demanding. They needed a complete product definition for a dynamic recommendation module to be added to casino review pages. The requirements were:

Required Deliverables:

1. Problem Statement – Clear articulation of the user/business problem
2. Testable Hypotheses and Goals – Specific, measurable success criteria
3. Recommendation Logic Overview – How the system decides what to suggest
4. Mobile-First Wireframes – Visual design specification
5. Epic Definition – High-level feature description in Agile format
6. User Stories – Granular development tasks with acceptance criteria
7. MVP Definition – What’s in scope, what’s explicitly out
8. 4-6 Week Timeline – Sprint-based delivery plan
9. Data & Tracking Plan – Metrics and measurement strategy
10. Risk Assessment – Potential blockers and mitigation plans

Key Constraints:

• Module cannot significantly shift core content positions (SEO stability)
• Must maintain or improve current Core Web Vitals scores
• Server-side rendering required for search engine crawlability
• Mobile-first design mandatory
• Must use MVP approach with clear incremental delivery

This wasn’t just a design task—it was a complete product ownership exercise from discovery through execution planning.

---

My Process: From Understanding to Delivery

Stage 1: Understanding the Real Problem (Days 1-2)

The first critical insight came from not taking the brief at face value.

The stated requirement was “add a recommendation module.” But I needed to understand why—what user behavior gap was this solving? What business problem justified the investment?

My analytical approach (with AI assistance):

1. Technical audit of the reference page (LeoVegas review on Casino Hawks)
   • I used standard SEO tools (Lighthouse, Screaming Frog) to assess performance
   • AI helped: Structure the findings into a diagnostic framework
   • I discovered: Great Core Web Vitals (LCP 1.3s, CLS 0.0), but zero intelligent cross-linking

Fig. 1 – Current state: Static text list with minimal engagement

2. Competitive landscape research
   • I manually reviewed 12 major affiliate sites (Gambling.com, AskGamblers, Casino.org, etc.)
   • AI helped: Organize findings into comparison tables highlighting patterns
   • I identified: A market-wide “relevance gap”—everyone uses static lists or basic filters, nobody does contextual algorithmic recommendations
3. User journey mapping
   • I mapped two key personas: “The Researcher” (arrives via Google, evaluates one casino) and “The Comparer” (arrives via top-10 list, wants to cross-reference)
   • AI helped: Visualize the journey flow and identify friction points
   • I pinpointed: The exact moment of disengagement—when the review ends and there’s no contextual “what next?”

The real problem wasn’t missing features. It was missing continuity.

---

Stage 2: Designing the Hybrid Logic (Days 2-3)

This is where the strategic thinking happened—and where AI moved from assistant to sparring partner.

My core innovation: A two-stage recommendation system that combines objective similarity with behavioral intelligence.

Stage 2.1: Objective Similarity Scoring

I designed an algorithm that matches casinos on verifiable attributes:

• Bonus Type/Value (30% weight)
• Payment Methods (25%)
• Game Focus (25%)
• Licenses (20%)

AI’s contribution here:

• I asked it to brainstorm initial weighting models and challenge my percentages
• It helped draft pseudo-code logic for the scoring mechanism
• Critical human decision: I chose these specific weights based on internal search query analysis showing what users actually prioritize (e.g., “high bonus” appears 2.3x more than “game variety” in searches)

Example AI interaction:

Stage 2.2: Behavioral Signal Boost

Here’s where I went beyond standard similarity engines.

My insight: Use aggregated search behavior (not individual tracking) to surface what users actually want when reading about a specific casino.

If users reading about “LeoVegas” frequently search for “fast withdrawals,” the system boosts casinos that excel at withdrawal speed—and adds a Motivational Badge (“⚡ Fast Withdrawals”) explaining why the recommendation appears.

AI’s contribution:

• Helped me think through edge cases (“What if search volume is too low?”)
• Suggested fallback strategies (use Google Search Console data if internal logs insufficient)
• Critical human decision: I chose transparency over optimization—the badge system makes recommendations explainable, even if it sacrifices some conversion potential

AI also warned me about potential risks I hadn’t considered:

• Insufficient search data on long-tail casino pages
• Seasonality skewing behavioral signals
• The need for data normalization (comma-separated strings vs. structured fields)

I built all of these into my risk mitigation plan.

---

Stage 3: From Logic to Interface (Days 3-4)

This is where AI became a true accelerator.

I had a clear vision for the UX: horizontal-scroll cards, mobile-first, performance-obsessed. But turning that vision into a working prototype would normally take days of HTML/CSS iteration.

My workflow:

1. I sketched the wireframe (pen and paper, basic layout)
2. I wrote detailed specifications:
   • Card dimensions, spacing, typography
   • Touch target sizes (minimum 44×44px)
   • Performance constraints (<80KB total, <0.7s LCP impact)
   • Accessibility requirements (ARIA labels, keyboard navigation)
3. I prompted AI (ChatGPT specifically): “Generate responsive HTML/CSS for a mobile-first horizontal scroll card component with these exact specs: [detailed list]. Use semantic HTML, include ARIA labels, optimize for Core Web Vitals.”
4. I iterated 4 times:
   • First version: Basic structure, needed refinement on swipe affordance
   • Second version: Added visual “peek” (partial card visible on right) to signal scrollability
   • Third version: Optimized image loading (lazy-load, WebP format)
   • Fourth version: Accessibility audit corrections (contrast ratios, focus states)

The result: A production-ready prototype in ~6 hours instead of ~12 hours.

But here’s the key: AI couldn’t have built this without my specifications. It didn’t decide to use horizontal scroll (I did, based on mobile UX best practices). It didn’t know to prioritize the badge visibility (I specified that based on user testing research). It executed my design vision efficiently.

---

Stage 4: Structuring for Agile Delivery (Days 4-5)

Translating a product vision into actionable development work requires a different skill set—and this is where the 23-page document took shape.

My approach (with heavy AI collaboration):

1. Epic and User Stories: I broke down the feature into granular, testable stories
   • “As a backend developer, I want to implement the Stage 1 similarity algorithm…”
   • “As a user on mobile, I want to see up to 4 casino suggestions with clear reasons why…”
   AI helped: Structure stories in proper Agile format, suggest acceptance criteria I decided: Which stories were MVP vs. future, story point estimates, dependencies
2. Timeline and Resource Planning:
   • AI provided: Industry benchmark data on similar WordPress projects
   • I calculated: 140-190 dev hours based on component complexity, added ±30% buffer
   • I scoped: 4-6 weeks with specific sprint goals and deliverables per week
3. Data and Tracking Plan:
   • AI suggested: Standard GA4 event structures and metric frameworks
   • I defined: The specific events to track (recommendation_impression, recommendation_click), parameter schema, and kill metrics
   • Critical human call: Setting the kill threshold at 3% CTR (below this = fundamental failure, not just need for optimization)
4. Risk Assessment:
   • AI helped: Brainstorm potential failure modes I hadn’t considered
   • I prioritized: Which risks were acceptable (manual badge assignment initially) vs. non-negotiable (SEO stability, performance budget)

Example of AI catching something I missed:

This kind of technical review accelerated my thinking and made the final document more robust.

---

Stage 5: Refinement and Quality Control (Day 5)

The final day was about tightening every detail through multiple review passes.

My iteration process:

• Revisions 1-3: Core content development (problem, logic, stories)
• Revisions 4-6: Adding depth (competitive analysis, fallback plans, cost estimates)
• Revisions 7-8: Clarity pass (removing jargon, improving flow, adding executive summary)
• Revisions 9-10: Final polish (consistency check, formatting, screenshot integration)

AI’s role in refinement:

• Caught inconsistencies (e.g., timeline estimates in different sections didn’t align)
• Suggested clearer phrasing for complex technical concepts
• Helped maintain professional tone throughout

My role:

• Made every editorial decision
• Ensured authentic voice (not “AI-ese”)
• Added personal insights and judgment calls that only domain experience provides

The hardest part wasn’t writing—it was deciding what NOT to include. The document could easily have been 40+ pages. Keeping it focused at 23 pages while covering all 10 deliverables required constant prioritization.

---

Key Lessons: When AI Helps (and When It Doesn’t)

After five intensive days of AI collaboration, here’s what I learned:

✅ Where AI Excelled

1. Structured thinking acceleration: Turning rough notes into organized frameworks
2. Technical prototyping: Converting specifications into working code
3. Data contextualization: Providing market benchmarks and industry comparisons (e.g., “CTR for text-only navigation typically ranges 1-2% based on industry benchmarks”)
4. Risk brainstorming: Identifying edge cases and potential failure modes
5. Documentation formatting: Maintaining consistency across 23 pages

❌ Where AI Failed (and Human Judgment Was Critical)

1. Problem framing: AI couldn’t identify the “relevance gap” in the market—that required competitive analysis and pattern recognition
2. Strategic trade-offs: Deciding to exclude ML/personalization from MVP wasn’t a technical decision, it was a business one
3. Weight calibration: The 30/25/25/20 split for attributes came from understanding user search behavior, not algorithmic optimization
4. Kill metric setting: The 3% CTR threshold required product intuition about what’s “needs tuning” vs. “fundamentally broken”
5. Authentic voice: AI-generated content sounds polished but generic—every section needed rewriting to sound like me

🎯 The Critical Skill: Knowing When to Accept vs. Reject AI Output

Example of rejection: AI suggested using user ratings as a similarity factor. I explicitly excluded this because affiliate casino ratings are often manipulated—search behavior is more honest. That judgment call required domain knowledge.

Example of acceptance: AI generated the HTML prototype structure, which I accepted with minor refinements. The output was technically sound and saved significant time.

The pattern: Accept AI output for execution tasks (coding, structuring, formatting). Reject or heavily edit for strategic decisions (what to build, how to prioritize, what trade-offs to make).

---

The Bottom Line: AI as Force Multiplier, Not Replacement

This project took approximately 40 hours of focused work over five days.

Without AI: I estimate this would have taken 55-60 hours—roughly 35-40% longer.

But the real value wasn’t time savings. It was thinking quality.

AI functioned like having:

• A research assistant who never sleeps
• A technical reviewer who catches edge cases
• A documentation specialist who maintains consistency
• A sparring partner who challenges assumptions

What it couldn’t do: Make the decisions that define product success.

The strategic choices—what problem to solve, what logic to implement, what metrics define success, what risks to accept—those all required human product ownership.

---

Why This Matters for Product Managers

The iGaming affiliate space is competitive and data-driven. This case study demonstrates:

✅ End-to-end product thinking: From problem diagnosis through execution planning
✅ Data-informed decision making: Using search behavior, competitive analysis, and performance metrics
✅ Pragmatic MVP scoping: Shipping value quickly with clear kill criteria
✅ Cross-functional awareness: Balancing UX, SEO, performance, and business goals
✅ Modern workflow integration: Leveraging AI tools without losing strategic ownership

Most importantly, it shows transparent methodology. I’m not hiding AI involvement—I’m showcasing how to use it responsibly and effectively.

---

📥 Explore the Complete Work

Download: 23-Page Product Document
Includes: Problem statement, competitive analysis, two-stage recommendation logic, mobile-first wireframes, Epic with 5 user stories, MVP definition, 4-6 week timeline, data/tracking plan, and comprehensive risk assessment.

Download: Interactive HTML Prototype
A working mobile-first demo featuring:

• Horizontal scroll card interface
• Responsive design (tested on iPhone/Android)
• Performance-optimized (Lighthouse 95+ score)
• Accessibility compliant (WCAG 2.1 AA)

---

💬 Let’s Discuss

• For Product Managers: Have you integrated AI into your product development workflow? What worked? What backfired?
• For Hiring Managers: Does this level of AI transparency in a portfolio piece build confidence or raise concerns about independent capability?
• For Anyone Curious: Where do you see the line between “AI-assisted” and “AI-dependent” in product work?

I’m actively refining this approach and genuinely curious about different perspectives. Drop your thoughts in the comments—let’s have an honest conversation about the future of AI-augmented product work.

---

About This Case Study

This project was completed in five intensive days as a product ownership exercise demonstrating comprehensive PM capabilities from discovery through delivery planning. The final deliverables represent production-ready documentation that could be handed to a development team and shipped within the proposed 4-6 week timeline.

AI Tools Used: Claude (Anthropic), ChatGPT (OpenAI), DeepSeek, Gemini (Google)
Time Investment: ~40 hours across 10 major revisions over five days
Strategic Decisions: 100% human
Tactical Execution: AI-accelerated, human-reviewed

All downloadable materials are available above. Feel free to explore, critique, and reach out with questions.