My bank statement converter application is dying, and AI killed it.
Six months ago, Monthly Recurring Revenue peaked in February 2026. Today it’s down 12 percent, and the trend is clear: new subscriber acquisition has flatlined since April. The cancellation rate hasn’t changed much, so the problem isn’t retention. People just stopped signing up.
I’ve been here before. Girlfriend Plus, an app I built in 2017, followed the same trajectory. Revenue spiked, I got complacent, did minimal development, and watched it collapse into irrelevance. The difference this time is I’m not going to repeat that mistake.
The Commoditization Problem
The core issue is simple: generative AI has made my specialized tool obsolete for a large segment of users. People are using free-tier ChatGPT to convert their bank statements from PDF to Excel. It works fine if you have a few documents. For most casual users, that’s enough.
I stopped posting revenue updates on Twitter in September 2025 partly because of this. I didn’t want to advertise the opportunity to a community that’s already figured out they can use AI to solve the same problem for free. But the real insight here is that building in public doesn’t translate into sales for every business model. Twitter attention is great for awareness, but it’s not a customer acquisition channel for infrastructure tools.
The uncomfortable truth is that I’ve created a market that made my own product easier to replace. By documenting and sharing my process, I accelerated the timeline for competitors and the AI community to understand the problem space.
Why This Matters for Indie Developers
This isn’t unique to me. Any developer-focused SaaS that solves a problem explainable in a few sentences is vulnerable to AI disruption. The smaller your feature set, the more commoditized your solution becomes. That’s a hard constraint on certain business models.
I have legitimate competitors now, some of them much better than me. When you build in public, you signal both opportunity and execution, and you attract smart people to that space. Some of them will outcompete you.
The path forward isn’t to give up. It’s to make the product harder to commoditize. More features, better UX, faster processing, deeper integrations. I’m working on exactly that. Users report that scanned documents feel slow, with no visibility into progress. “Processing” for five minutes is anxiety-inducing. So I’m adding progress indicators and optimizing the pipeline. These aren’t revolutionary changes, but they move the product further from what a generic AI tool can provide.
Content Marketing as a Long-Term Bet
I’m doubling down on content marketing because it’s the only marketing channel that compounds. You pay once, and if the content is good, it generates value forever. Paid advertising is a treadmill. Stop paying and your leads disappear.
I’ve also been exploring unconventional channels: transit ads in Hong Kong, sponsorships on niche podcasts, YouTube video partnerships. These have high friction (nobody replies to cold outreach), but the upside is real. A single YouTube sponsor spot on an evergreen video about personal finance could run for years.
The challenge is that traditional content marketing takes time, and my revenue is declining now. There’s a mismatch between the timeline of strategic improvements and the immediate pressure of churn. This is where discipline matters. I can’t panic-spend on CPC advertising because the economics are terrible. I have to stay committed to the long-term approach even when the short-term metrics are red.
Learning from Customer Support
Here’s something I neglected for too long: Stripe captures cancellation reasons when users leave, and I’ve never actually analyzed that data until now. It’s a goldmine for understanding what’s broken. When a user says “switched services,” I can follow up and ask why. That conversation teaches me what I’m missing.
I’m also paying close attention to support tickets. A user recently thought their scanned document upload was broken because the status showed “Processing” for an eternity. That’s two product improvements right there: faster processing and transparency into what’s happening. Small details like this compound.
The Non-Negotiable: Keep Building
I’m not going back to working at a company. Last employment was November 2020, and corporate engineering in 2026 sounds like pure management theater. Friends tell me the job is basically babysitting AI agents and reviewing pull requests. That’s not for me.
The only path that makes sense is to keep iterating on this product, to understand the market better through customer feedback, and to accept that some businesses are fundamentally more vulnerable to disruption than others. The question isn’t whether AI will replace my tool for casual users. It already has. The question is whether I can build something defensible for power users and enterprise customers before I run out of runway.
Sometimes the best indicator that you should persist is that everyone else already quit.