It's not like they had a meeting where a consultant said "We reviewed all your requirements, and you could build this with normal software engineering practices, or we could build this in spreadsheets with associated deleterious effects on privacy, correctness, durability, extensibility, readability, and interoperability, likely resulting in an estimated $N to $NN million shortfall in 5 years. What do you want? Yeah, let's do the spreadsheets."

It's a two-pronged attack of using tools that you have figured out as you go (not everyone is an educated, practiced, professional programmer, or even good at Excel) and a sunk cost fallacy/resistance to changing what "works".

Spreadsheets are like the ubiquity of motorcycles in 3rd world countries. Jumping on one and blasting off without any safety equipment is cheap, fast, everyone uses them, and they get 80% of the job done. And they will kill you, but that's life.

The thing about spreadsheets is you don't have to commission bespoke software. And just the value of that swamps everything else in the vast majority of cases.

In small businesses I doubt the success rate is anywhere near 50% if a non-technical person recognizes a business need for a software solution, hires someone to build it for them, and then it actually reaches a useful conclusion. But if you are numerate at all you can drive a spreadsheet. Not having to swim with the sharks for months or years while you hemorrhage cash is hard to appreciate for the sharks (us, the people who build technical things). Most small businesses and many large ones don't have anyone who can manage even contracting something like this out.

If things have gone on long enough that the problem is stable, clearly defined, and the costs of 3rd order effects like lack of revision control are losing enough money to justify the risk of a custom software project, sure, it needs to be done. But for many companies that risk will always be bigger than the ones that come with spreadsheets.

ERP systems are supposedly finished software that you deploy and tweak for particular business needs. The success rate on deploying those is waaaay below 100%, and the expenses are huge. If your company or department is non-technical the time for custom business software is usually when your hair is on fire and it's the only option.

Very true, especially for small business. But large multi national enterprises are also doing this.

I worked at a company that did product testing from spreadsheets. And not statistically modelling them. Like physical products. The spreadsheet logic was controlling power supplies on a physical test bench.

Impressive.

I worked at a financial software firm written in python that used Excel behind the scenes to run calculations and modelling.

> written in python that used Excel behind the scenes

That's from my worst nightmares :(