The Website Came Last

Marketa Heidelberg originally just needed to move her website somewhere else. It turned into a clean-up of her whole online business, because the move alone would not have solved anything.

The new website had to fit how Marketa sells, where her customers come from and how much time she can give it. I built it in Claude Code as the very last step, in spring 2026. AI cut the build down to one evening. But for it to build the right thing, someone had to steer it.

Where things stood

Marketa is a personal trainer. She sells online movement programs and does 1:1 coaching. Most new prospects find her on Instagram.

On the old website, even the contact form did not work. She sold her programs through Ablefy, a platform where a content creator can run a whole store: a page for every product, a checkout and the content behind a paywall. But the offer was a mess and the path to each program was different. For the best-selling program, Morning Moves, it went from an Instagram story to a landing page in Canva, then through two pages on the sales platform, and only then to the checkout.

There was no single place where someone could look through the whole offer and choose. And sales depended on whatever Marketa happened to be promoting on Instagram. When she promoted nothing, nothing sold.

1. Map how it all works

Before changing anything, in 2025 I wrote down what Marketa uses and how it all connects. I counted seven tools: website, domain and email on Jimdo, the Ablefy sales platform, bookings through Calendly, email marketing in Brevo, a Linktree link hub on Instagram, calls on Zoom, and Zapier moving contacts between the sales platform and the email marketing.

Then I mapped the customer journeys: how someone gets from Instagram to a program, to booking a call or to the newsletter. And finally the contacts: where they come from (the form, bookings, the newsletter, email, Instagram messages, purchases) and where they end up.

2. Fix what hurts most

Before any big changes, we fixed what was doing damage right away. It was not a planned project, more a series of fixes along the way.

The biggest problem was the domain. Jimdo was also the domain registrar, and its DNS settings allowed almost nothing. That is why many of Marketa's marketing emails landed in spam. We left Jimdo completely, moved the domain to Websupport and set up Google Workspace.

In Brevo, we split the contacts by how close someone is to buying, from followers to prospects to customers.

And on the sales platform I found out that the checkout could be changed with CSS. So Morning Moves got a better checkout before the new website even existed.

3. Understand the business model

At the end of 2025 the plan was still just to move the website. In spring 2026, though, came the idea of a subscription: one access to all programs instead of one-time purchases. That changed the question. It was no longer about where the website would run, but about what the whole business should do.

The analysis showed that Marketa actually runs two businesses. 1:1 coaching is tailored to each client and runs into one limit: how much time Marketa can give it. Online programs do not have that limit, but they sold only when Marketa was promoting them. Every extra sale cost more time and more content.

So the goal was not a nicer website. Sales of online programs had to grow without Marketa adding her time at the same rate. And because she funds everything herself, digital sales have to pay for themselves.

I also wrote down what we did not know. Whether people move from online programs to coaching. Whether the problem is in the offer, in how it reaches people or in who it is aimed at. Whether to stay on the sales platform or build her own portal, which would cost money and need maintaining. We took the subscription as an assumption, not a certainty: it will not fix the problems with one-time purchases on its own, but it starts a new way of earning from the content.

This led to the main decision: everything from Instagram would lead to the website, not straight to individual pages on the sales platform. That turned the website from a business card into the foundation of the whole business.

4. Let the data decide what to build

I went through the sales data from the platform, and it directly shaped the website. Free samples converted at 35–46%, so they have their place on the website as the first step. The program trial converted at 17.7%, the full program at only 7%. Coaching never sold through the platform once, so on the website it leads straight to an intro call. Whatever did not sell, we removed from the offer.

On that I built the structure of the website: from free samples through the first programs to coaching. Then came the design system, meaning the logo, colours, fonts and the rules for using them. I made the first version as a plain HTML page, before Claude Design existed.

5. Build

I started building the website in Framer. In March 2026, though, I switched to Claude Code with the Opus 4.6 model and built the whole website there.

I am not a developer, I am a product manager. Claude wrote the code and I decided what gets built and why. According to the git history, it went like this:

• 11 March: first commit.
• 15 March: all the main pages in one evening. The pages for three programs and the page about Marketa took about twenty minutes together. It went that fast because Claude got a finished design system, the site structure and the copy.
• 16 March: the website connected to Brevo, so newsletter sign-ups go straight into the right list.
• 29 March: double opt-in added for the newsletter and for the waiting list for a new program.

The whole website is in German, which I speak only at a beginner level. That was not an obstacle either. The copy was written in German with an English version alongside, which I used to check it, and Marketa had the final word.

After launch, I used data from Google Search Console and Semrush to add what people actually search for and what the website was missing. On top of that came traffic analytics and redirects from the old addresses, so nothing Google already knew got lost. The old website ended the moment the new one went live, and we cancelled Framer completely.

The result and what comes next

Analytics shows that the path through the website works. Instagram is now the biggest source of visits, with Google second. The most visited program page is Morning Moves, the one people used to reach through Canva and two more pages.

The subscription is not live yet, and programs are still sold as one-time purchases. The portal is the next step.

You can see the website at marketaheidelberg.com.

Before and after

What follows from it

The website pages themselves were built in one evening. Before that came the mapping, the fixes and above all the decision about what to build.

Had we stuck to the original plan and just moved the website to Framer, it would not have solved any of what was holding Marketa back. What mattered was the approach: analysis first, which in the product world we call discovery, and working out what to build. Claude Code then took care of the execution and made it much faster.

On a project this size, thanks to AI, nobody has to master engineering, design, marketing or sales in depth anymore. AI covers much of those fields. What it does not cover is understanding how a whole online business works, knowing which tools are available today, and the judgment of what is good and what is not. That is where a person is still irreplaceable, and Claude will not do it for them.

AI multiplies whatever a person brings to it. Those who know what they are doing will do more with it, and faster. Those who do not will just make their mistakes faster.