You can't solve what you can't see.

Thankfully, there are tools available to do just that. I wanted to speed things up and process a more diverse set of bank statements, so I turned to the data for guidance.

If you have never heard of SimilarWeb get ready to have your mind blown.

Being able to see your competitors traffic sounds like some sort of voodoo right? The TLDR is that they buy web traffic data from ISPs other sources, they offer a unique view of the web. You can see which keywords your competitors get traffic from.

The leading sites by country list helps to narrow down the messaging for different markets.

SimilarWeb shows which countries your competitors are getting most of their traffic from.

I had assumed my initial customers would be accountants in Australia, UK, Canada and the US, maybe not.

I have been running a highly tuned set of Google Ads with a $50 / day budget and a max CPC of $1. My goal was to get users who are looking to convert their bank statement pdfs into excel. Initially I had restricted the ads to Australia, UK, Canada and the US.

This had been getting a couple of clicks a day, not enough to draw any conclusions, but enough to test UX changes. One thing that stood out was that people did read the privacy policy before uploading.

After analyzing SimilarWeb, I made the decision to remove the location restriction. I saved the changes and took a break for dinner. When I came back two hours later my budget for the day had been spent. Maybe 60% of the uploads where actually bank statements. So I dropped my max bid down to $0.50 and my average CPC came down to $0.35, the uploads continued.

Surprisingly, hardly anyone read the privacy policy, yet nearly everyone uploaded something. A lot of random stuff got uploaded. Aside from bank statements it was mostly invoices and lists of stuff.

The most random upload was a bunch of passport scans, I wish I could talk to that user ... like, what did you expect to happen?

Having a diverse set of statements run through the software was eye opening. I had never heard of most of the banks, which included:

Some of the bank statements looked like they where designed for a dot matrix printer. So I turned off the ads and am currently adjusting the detection to make it more robust.

A handful of the statements had 50+ pages of transactions for a single months statement. So I have refactored the code to be a little more thoughtful in how I am using memory. There is something oddly satisfying about a document with over a thousand transactions reconciling flawlessly. Maybe that's just me.

My Trello board is still full of tickets with edge cases to account for. Lots of them simple others, not so much.

In total there have been 250 ish uploads, but only 10 user registrations. None the signups proceeded past the adding a card stage. Once I have worked through my backlog of edge cases i'll switch the ads back on and take the next step. I can see there is demand for a more generic table extraction product. Overall i'm happy with the progress.

Running a small ad campaign with a few hundred dollars was a cheap way to validate the UX changes I made.

I over optimised my ads too early and should have let the data drive restricting the ads by location.

If you haven't tried SimilarWeb already, I strongly recommend spending an hour or two with your competitors data.