CASE STUDY
MAPIA
From Problem to Product in 6 Weeks
To complete the Product Leadership Program taught by Íñigo Medina at Instituto Tramontana, we had to build a digital product in six weeks. Together with Paula Carrasco (Mercado Libre) and Asier García (Chubbyapps), we decided to solve a problem Asier experienced daily: manually personalizing dozens of press emails. Asier knew personalization worked — it improved his response rates — but it wasn’t scalable. That’s how MAPIA was born.
Summary
Duration: 6 weeks
Team: Paula Carrasco, Asier García, and me
Result: Functional MVP of AI-powered email personalization platform
The Problem
Email marketing tools (Mailchimp, etc.) only insert basic tags — name, company — that are obvious and ineffective. Nobody wants to feel like just another number on a distribution list.
Asier validated that manually personalizing each email (mentioning last contact, connecting news with journalist’s interests) significantly improved response. But the process exhausted him and couldn’t scale.
The challenge: Could we automate deep personalization without losing authenticity?
The Solution
MAPIA (Mails Automáticos Personalizados por IA / AI-Personalized Automated Mails): platform that generates mass personalized emails using AI, feeding on real context from each recipient.
Scope decision: Only mass personalization. CRM, sending, analytics were deliberately left out of the MVP.
My Role
In a small, multidisciplinary team, roles blur and joint work is what matters — everyone contributed where they could. For my part, I focused especially on research, design, and MCP technological development.
Process (6 Weeks)
Week 1: Definition
Initial prototype in Lovable. Concept validation without investing too much effort. Lovable generated more features than we asked for, forcing us to be more explicit about what we did NOT want to build.
Week 2: User Research
Conversations with potential users revealed critical insights:
Alfredo Lores (marketing):
Already did similar processes manually. Needed CRM integration. His concern: “That it doesn’t seem like a robot.”
Emma González (communications):
Sent non-personalized notes due to lack of time. Adjusting tone (formal/informal) was enough for her. Her fear: “I’m terrified to delegate something so personal to AI.”
Key conclusions:
- Yes, there’s demand for authentic personalization
- Trust in AI is a barrier to address
- Personal data is “in people’s heads,” not in structured databases
Week 3: Strategic Pivot
Technical problems with Gmail OAuth forced us to pivot to Mailchimp. Turned out to be a better decision: ~70% market share, users with already-created audiences, more validatable product, publishable on their marketplace.
Final stack: OAuth2 for Mailchimp + transactional API + contact notes as personalization source.
Week 4: Technical Blockages
Mandrill API wasn’t working. Days debugging. Perplexity gave me the answer: required paid account (poorly documented by Mailchimp). I bought a domain (mapia.tech), configured DNS (SPF, DKIM), migrated to Resend.
These frustrating days turned me into a vibe coding enthusiast — I learned to build software with AI through prompts.
Week 5: Execution
Resend migration completed, Mailchimp connection, audience selection, basic personalization, functional mass sending.
I added an integrated contact notes manager (simpler than native Mailchimp interface) to generate adherence.
Week 6: Polish
Landing page working and focus on branding: I added bananas as a kind of logo, giving the product personality while making a direct nod to Mailchimp’s monkey. We discovered unexpected “functionality”: automatic adaptation to recipient’s language if indicated in notes.
Results
Technical validation: Mailchimp + OpenAI + Resend integrated
Feedback: Positive from multiple profiles (PR, marketing, sales)
New processes: from conceptualizing and designing to directly implementing a functional application
Identified Risks
We identified several risks requiring validation:
- Limited adherence: For occasional users (occasional press releases), ROI of configuring the system could be low
- Technical dependency: Lovable is a “black box” that makes scaling or migrating to own infrastructure difficult
- Market size: The niche might not easily expand to broader user base
- No moat: Mailchimp could add this functionality. We don’t have “hard-to-copy”
- Trust in AI: Psychological barrier expressed by users like Emma: “I’m terrified to delegate something so personal”
Key Learnings
Vibe coding
- Don’t speculate, ask technical questions to AI to understand how things work
- Copy errors directly into prompts
- Build in parts, validate before integrating
- Rollback without fear — sometimes it’s faster than debugging
Product
- Best products are born from personal frustrations
- Without clear limits, the product becomes a feature amalgamation
- Real feedback > speculation — ship early
- Pivots = adjustments with new information, not failures
Research
- “Terror of delegating to AI” signals real product barrier
- Data in heads vs structured systems = critical for architecture
Takeaways
MAPIA demonstrated it’s possible to build functional products in weeks with modern tools. But the real value was the learnings about decision-making under pressure, effective pivots, and maintaining focus on solving real problems.
The process transformed me: from conceptualizing products and coordinating developers, to directly prototyping by conversing with AI. This ability to “think in code” is increasingly valuable for product managers.
Email me you want to chat about this case or are looking for a product manager for your team.
–