I replaced Google Analytics, Sentry, and Intercom with AI in a few hours
Three SaaS products replaced. Under $20 a month. All built with AI in about 15 hours total across a few weeks. Not because I have a grudge against SaaS, but because each time I signed up for the “proper” tool, I spent more time fighting the interface than actually getting what I needed.
This is part of a bigger experiment. I’m building Inselnova, a browser-based multiplayer strategy game, and I wanted to see how much of the surrounding tooling one person can build themselves with AI.
Error tracking (replacing Sentry)
I signed up for Sentry. Had a look. Thought, “I could just build this myself.”
I asked Claude Code to build a basic error tracker for the frontend (Vite) and backend, then send everything to a new API system I called Helm. Frontend error boundary wraps the app. Backend wraps key handlers. Both POST to Helm on error.
When a player hits a bug, I get who hit it, what happened, the stack trace, browser info. I copy that into my editor, fix it, mark it done. About 2 hours total.
If you’re considering this: it only works if you’re a solo dev or a very small team. Sentry’s value is in shared dashboards, alerting rules, and team workflows. If it’s just you, a list of errors with enough context to fix them is all you need.
Analytics (and why Google Analytics lost me)
A week after launch I realised I had no frontend analytics. Threw in Google Analytics. Regretted it a week later. It’s unrecognisable from 10 years ago. The UI is hostile. I somehow ended up with two accounts. Finding basic funnel data takes three clicks through six menus.
I was already sending telemetry for bug tracking. So I asked Claude Code to adjust the same code to capture what GA does: sessions, events, user IDs, tags, timestamps. One hour. Drop-in replacement for GA’s JavaScript.
The frontend is static Astro. Telemetry pings a small service in Helm. Stored in PostgreSQL. I even exported my GA data as CSV, gave it to the AI, and it backfilled a week of traffic.
If you’re considering this: GA does one thing that matters, record where people click and how long they stay. Custom code does that too. If you don’t need segmentation, attribution models, or their “insights” feature, you don’t need GA.
How I built a funnel in an hour
I always start bottom-up. Where do I want people to be? Work backwards from there. Once you know your endpoint, you know how much traffic you need at the top.
Mine has four stops: landing page, signup page, account created, enters the game. I added 14, 30, and 90 day filters because retention looks different week to week.
The trick is tying telemetry to game data. When someone lands on the site they get a session token. When they register, the token passes to the backend. When they enter the game, same token. End-to-end tracking without any third-party tools.
Built it in about an hour while doing other work.
If you’re trying this: don’t measure things because other products measure them. Define what success looks like for your product and instrument backwards. For a game, I needed to know: do new players stay? The funnel tells me that. GA tells me 47 different things nobody asked for.
Re-engagement (replacing Intercom)
Early on I had drop-off. Product instinct says reach out to everyone who joins, personally. I did that for the first few batches. Sent emails. A couple of people came back. But I couldn’t track if my email brought them back or if they’d have returned anyway.
I didn’t want to spend every morning sending emails manually. I also didn’t need a full Intercom setup for what’s basically two things: send a message when a condition is met, track if they responded.
So I built it. Extracted the data, asked Codex to help figure out what conditions should trigger a message, asked ChatGPT to shape the copy. Then pasted the spec into Claude Code.
An hour later I had something rough. Not finished. I didn’t want it finished. I wanted something I could play with. After a few more hours in the evenings, it was sending automated in-game messages and emails with full tracking.
People have replied to these messages. Some don’t realise it’s automated.
Every campaign is versioned, so later I can run analysis on what converted better. Too early for that data yet.
The real reason I built this: I had a card in Trello that said “talk to the last batch of signed up players” and I kept putting it off. So I automated it.
User behaviour tracking
Early on I found something useful: players who finish the first achievement set almost always become long-term players. So the backend runs an hourly job that tags each user: likely long-term, churning, or churned.
The re-engagement system uses these tags. Churning players get nudged. Long-term players get left alone. None of this works if the data lives in different systems. Bug data, telemetry, game data, and user tags all need to be in the same database.
If you’re doing something similar: this is the bit that makes it all worth it. Buying four separate SaaS products means your data is in four places. Building it yourself means one database, one query, one view of each player. That’s the actual advantage.
What it costs
PostgreSQL and game server on Render.com. Total: under $20 a month.
Game server: $7/month, running at less than 20% capacity. Database: $6/month. Frontend: free, static files. The rest is build minutes.
Load tested it. Can run 100+ concurrent players on that setup.
Should you do this?
Three questions to ask yourself.
Do you own the product? If you work on a team where analytics goes to a data team and errors go to DevOps, this doesn’t make sense. You need shared dashboards. If you’re solo or the owner, custom tooling is faster.
Do you know what you need? I knew I needed errors logged and a funnel. If you’re still figuring out what matters, SaaS forces you to think about it. Custom tools let you skip that step and potentially regret it later.
Can you afford the time? I spent about 15 hours total across all four systems. That’s time I didn’t spend on the game. For me it was worth it because the SaaS would’ve cost the same hours in mental overhead, fighting interfaces and waiting for features. Your situation might be different.
The tools I built are boring. No animations. No machine learning insights. Just the data I need, formatted the way I think about it. That’s the point.
The game is Inselnova, a free browser-based multiplayer island strategy game. It’s in early access and I’d love for you to try it.
If you’ve built your own tooling to replace SaaS products, I’d love to hear what worked and what wasn’t worth the effort.
