Why we waited 2 years to build our most requested feature

For the past five years, our Spanish members have asked for one thing: "Let me book my medical appointments through the app." Not some… All of them.

It was the #1 request in every satisfaction survey. But instead of immediately throwing engineers at the problem, we spent two years waiting, testing, and learning. This is the story of how Operations turned a seemingly impossible ask into reality—and what that reveals about building products in a market that's not ready for you yet.

🏥 The problem: when being the second-best isn't good enough

In Spain, health insurance is about accessing a medical network. When you visit a clinic you don't pay anything upfront—no copays, no reimbursement hassles. You show your card and walk out. The clinic bills your insurer directly. Yet having access to clinics wasn't enough for the type of customer we attract: people who expect a fully seamless digital experience for their health insurance.

Early on, we offered what most insurers provide: a searchable map of clinics instead of a 500-page PDF directory. Nice, but not enough. 

So we integrated with Spain's dominant online booking provider, allowing members to book appointments directly in the app—like many other insurers do. We became the second-best insurer in Spain by number of agendas available to book in-app.. 

Our members' response? "This is great, but I still have to call on 80% of clinics to book."

💡 The manual solution: Concierge service

Rather than accept defeat, we launched a magical service that doesn't scale: Concierge.

When members contacted support asking about coverage, we didn't just answer their questions—we offered to book the appointment for them. "I see you need a dermatology appointment. Tell me where and when works best, and we'll call for you."

The delight was immediate. Members who used this white-glove service reported 10-15% higher satisfaction than those who didn't. In reviews, they called it their favorite Alan feature. Adoption skyrocketed.

But each appointment request took 30-60 minutes of agent time.Many members had specific requirements: early morning or late afternoon slots to avoid missing work, English or French-speaking doctors for our international members...

This was delightful, but not scalable.

We needed to either reduce usage at the same costs, or to reduce cost and expand usage. We went for the harder problem.

🧪 First attempts: when technology isn't ready

When ChatGPT launched in November 2022, we immediately saw the potential for automating these calls.

As someone in Operations, not a Software Engineer, I started experimenting.

March 2023: The Zapier phone call experiment

 Using off-the-shelf-tools, I built my first prototype. It was... impressively terrible.

The latency was 10-15 seconds between each exchange. Imagine calling a clinic, saying "hello," and waiting 15 seconds for a response.

April 2023: The email approach

Since real-time conversations weren't viable, we tried email. AWe ran several tests—it worked! The AI successfully booked appointments. The problem: average response time was 4-5 hours, sometimes stretching to 2 days for a single appointment. This was not good enough.

We paused. The technology wasn't ready yet.

⏰ The waiting game: what Ops did for two years

While waiting for AI technology to mature, we:

We built conviction about the solution, but waited for the right moment to execute.

🚀 Early 2025: technology catches up

By early 2025, AI voice providers with acceptable latency (under 1 second) started emerging. We decided to prototype to prove it worked! 

The no-code prototype

I built a working prototype using:

The testing process

After months of iteration—fixing hallucinations, preventing the AI from confusing its role, refining the prompts—we reached a stable version and launched a pilot: Intercom Integration for Members

Support agents could trigger the automation directly in Intercom when members requested appointments, integrating seamlessly with our existing workflow.

📊 The results: feasibility risk cleared!

Here's what we learned: we can reach a 90% probability of successfully booking an appointment our member accepts  by calling 7-10 centers.

Why don’t all calls work?

But the ones that worked? They worked brilliantly. While it's clearly an AI voice, clinics are patient with it because we are bringing them new patients.

🎯 What Operations does at Alan (and why it matters)

This journey reveals the role of Ops in building 0-1 products at Alan:

1. Refusing to settle

Being second-best in online booking wasn't acceptable when our members demand excellence. Operations excellence drives the ambition for 10x better solutions.

2. Deep problem understanding

We spent time understanding   exactly in what members want , what “good enough” looks like (hint: not 20% of medical network available with online booking), and what excellence requires.

3. Risk reduction through prototyping 

We resisted the pressure to build immediately. Testing every few months saved us from wasting engineering resources on premature solutions.

AI and no-code tools expanded what Ops can do. I designed the member experience in Figma, built working prototypes with Zapier, measured results, and gathered conviction. 

Before a single engineer wrote production code, we proved:

🔮 What's next

Our engineering team is now building the robust, secure, scalable version of what started as a prototype. We're launching to members in early 2026.

But this is just one example. Spain's health insurance market has countless problems waiting for innovative solutions:

At Alan, we have  unlimited problems to solve. That's exactly what makes Operations exciting and essential.

Want to see more ?