A profitable PDF-to-Excel SaaS published its own numbers on the way down

Revenue down 24% from the February peak. New subscribers from 191 a month in January to 45 in August. MRR growth negative since April.

The founder of Bank Statement Converter posted the whole thing publicly on August 28 (Bank Statement Converter). The named cause is free-tier chatbots being good enough at small-batch statement parsing, even though they're slower and more expensive at volume than a purpose-built converter. Being technically worse doesn't matter when the customer already has the tab open and the marginal cost is zero.

Most disruption coverage is an analyst projecting a category decline. This is first-party accounting from the company being eaten, with dated monthly numbers. That makes it usable evidence instead of narrative.

The failure mode it identifies is narrow and testable. Any product whose entire value proposition is a file-format transformation, PDF to Excel, CSV to JSON, DOCX to Markdown, image to text, is now competing against a general model that does the transformation adequately as a side effect of existing. The transformation isn't the moat. It never was; it was just expensive enough to build that nobody bothered.

What survives that test? Volume, integration and governance. Bank Statement Converter's own post concedes the chatbots are worse at volume. A product that handles ten thousand statements a night with deterministic output, error handling, an audit trail and a support contract has something a chat window doesn't. A product that handles four statements does not.

Set this next to the Lovable CTO's argument from the same week. Fabian Hedin told Latent Space that SaaS companies now have to "provide the shovel for AI to use their capabilities," meaning every capability you ship has to be reusable agentically rather than only through your own UI (Latent Space). Menlo's Deedy Das, whose firm led Lovable's $400M Series C at $13.3B, put the company at a $500M annualized run rate with over 900 million monthly visits to Lovable-built apps.

And then look at what launched on Product Hunt yesterday. PageIndex took #1 with reasoning-based retrieval over document folders, Firecrawl published a developer index citing 63% recall@10 across 1,179 real queries, and Colrows launched a semantic execution layer compiling natural language to governed SQL (Product Hunt). Three launches in 24 hours, none selling a dashboard, each competing on a published retrieval number and naming an incumbent viewer as what it displaces.

The index is becoming the product. The interface is becoming a commodity. If your product is a viewer over a corpus, the viewer is the part that gets eaten, and the structured map of the corpus is what you should have been selling.