Marius Schäffer

Building @copyhero_app in public: E-Commerce Scaling AI Platform. €10.2k MRR → €50k. 10+ yrs dev.

[Marius Schäffer @marsc_hb]

Decided to take a break from X until we hit 10k MRR with our SaaS.

Didn't expect it to take more than a year 😂

But this month we finally did it. Honestly, it doesn't feel any different. The work just keeps growing with it.

When I stopped posting we were at 2.5k MRR. So roughly a 4x in a year.

Next goal: 50k MRR, another 5x.

This time I'll keep posting along the way.

[Marius Schäffer @marsc_hb]

We are focused on ecom businesses basically doing product listing and research.

Haha my god I am way to deep in my bubble. I though of Google Tag Manager. Our customers list 1000s of products a day to different ecom stores. from shopify they get synced to their GMC. We do have the revenue analytics from shopify but I want the google metrics as well so we can display ROAS etc on a product level and iterate on the listing/research process.

But still nice to meet you!

[Marius Schäffer @marsc_hb]

@rauchg Oh i thought this was engagement bait. What?

[Marius Schäffer @marsc_hb]

The audience for the app is totally different from my audience on X. Therefore not posting didn't hurt in that regard.

There was no "hidden" secret or single channel we could benefit from. It was always about improving the product and delivering customer value.

Regarding the next 5x I think the biggest challenge is Customer Acquisition.

[Marius Schäffer @marsc_hb]

Decided to take a break from X until we hit 10k MRR with our SaaS.

Didn't expect it to take more than a year 😂

But this month we finally did it. Honestly, it doesn't feel any different. The work just keeps growing with it.

When I stopped posting we were at 2.5k MRR. So roughly a 4x in a year.

Next goal: 50k MRR, another 5x.

This time I'll keep posting along the way.

[Marius Schäffer @marsc_hb]

@DylanLaCom @ShopifyDevs I don't think so:

https://t.co/FjhM2j9GRQ

[Marius Schäffer @marsc_hb]

@lngmnn8 thats true :D

first thing we are trying for the next step is a simpler and more "flashy" offer to improve our CAC. That's basically what is holding us back rn.

[Marius Schäffer @marsc_hb]

No quite the opposite actually. I always identified as a backend engineer. This was mainly driven by me not being able to come up with something that looks good.

But I can tell from my feeling if something looks good or not. So I let Claude do the heavily lifting and just pick what looks best.

[Marius Schäffer @marsc_hb]

@jayjoshuamedia Oh where my DMs closed? Should be working now.

[Marius Schäffer @marsc_hb]

Yeah. That was a hard realization.

I hit a mental hard stop when I was about to approve Claude building two new database tables and inserting millions of rows for a one-off migration. Combined with our pgBackRest WAL archive, that could easily have added up to terabytes of unnecessary writes.

What I do now is discuss every feature with Claude on three layers:

1. What I actually want to achieve.

When you let Claude run off on its own, it often builds things it wasn’t even asked to.

2. Architecture and public API contracts.

Claude likes to bloat these for backwards compatibility and who knows what other reasons.

3. Module level.

I don’t really care much here. Unit tests can cover it, and Claude is really good at that.

But you have to keep an eye on 1 and 2 as early as possible, so I have Claude write them out in clear sentences for every issue I spec.

I also have it write actual diff lines for the public interfaces. You can catch so much bad design this way. When you push back on those changes, Claude usually finds the elegant solution on its own in the second iteration. I haven’t managed to get that reliably on the first try yet.

There was a great post on this by @addyosmani

https://t.co/NC4T9x8FJl

[Addy Osmani @addyosmani]

https://t.co/2gyp2Erpxx

[Marius Schäffer @marsc_hb]

@victorinpublic Monster white is a good start! 🫡

[Marius Schäffer @marsc_hb]

@taylanhq @rakibxzs are mac apps hard? I am trying out stuff for copyhero as well with tauri and it looks promising so far. We do have all the backend stuff in place tbf

[Marius Schäffer @marsc_hb]

@ClaudeDevs i do not really understand this.

was this status line visible to the user before? Great improvement though!

What I noticed that usage drains much slower (could also be placebo).

[Marius Schäffer @marsc_hb]

@SchaefferV12416 Thank you :)

[Marius Schäffer @marsc_hb]

@VictorKay97 Unfortunately this is in german but I basically was at a football game when this happend. Was a false positive but could have been bad.

The PM Bot is basically handling the GH side and can issue kickoffs of Claude Code/Codex Agents.

[Marius Schäffer @marsc_hb]

@build_with_kush Nothing crazy. Product got a lot better. And the customers using it are using it more and more.

Almost feels like a big enterprise solution rn.

[Marius Schäffer @marsc_hb]

@livesanches Thank you!

[Marius Schäffer @marsc_hb]

ye it wasn't deliberate for a year. But 10k was a goal I wanted to hit.

I think when you are building something where the audience on X overlaps with the product audience there is no harm in doing X as well. But since the target audience for my X and my SaaS are totally different, I decided to focus on the product first.

[Marius Schäffer @marsc_hb]

That's a really good question, and honestly, we didn't stop anything. I still handle all of our customer support myself. But with the way the system works, it's basically nonexistent right now.

All the CS I do is setup and integration work. The rest runs more or less on autopilot.

1. Service uptime

I think the only real outage we've had was when Hetzner had an issue with their vSwitches and our app servers couldn't reach the database cluster. It was resolved within about an hour, and luckily it wasn't during a busy period.

2. Architecture

When we moved away from AWS, we rebuilt the infrastructure and services from scratch for ultra-high availability (multi-region PostgreSQL with Patroni and HA Kubernetes with 3 replicas). The only thing that can really bite us is networking issues in the private network, but that's a risk we accept here. So far we've only experienced it once.

3. Product analytics / logs / metrics

Basically everything lands in either PostHog or Dash0. So all our logs, alerts and product analytics are stored separately and are AI-native.

4. Grok bot & Claude

With Claude, I can browse the entire Dash0/PostHog stack and trace errors or issues a customer is facing without doing anything myself. Most of the time it comes down to a misconfiguration. And because we have very sensitive alerts in Dash0, most errors land on my desk before a customer even notices them.

I also have a Grok bot connected to PostHog and Dash0 via webhooks that trigger on every alert. The bot investigates, and if something is urgent, it can page me through ilert. So it's basically a 24/7 on-call engineer, which is really, really cool! The bot can also start working on a fix while I'm getting to my PC. That has only happened once, though, because of a database migration that held a row lock for too long.

5. AI-native

Our app also has an AI assistant that has the full API surface exposed as an MCP server. So it can also trace misconfigurations and issues from the user's perspective, on the application layer. The same MCP server is available to Claude and other clients as well.

Since we're focused on large B2B accounts, we don't cut corners on customer support, and I think the system we have in place now should scale pretty easily without degrading CS quality.

[Marius Schäffer @marsc_hb]

@ChalithaWi Thats the real question. What's more valuable?

I couldn't have done both at the same time because of focus.