Can AI build a custom tool cheaply enough for a small organisation to afford?
Yes for small internal tools: dashboards, data entry, reporting, and glue between systems that used to take months now take days. What has not become cheaper is deciding what to build, keeping the data safe, and maintaining the thing afterwards, and those are what determine whether it was worth building.
What actually got cheaper
Typing. The part of software work that consists of writing known patterns correctly, wiring a form to a database, parsing a file format, producing a chart from a table, has collapsed in cost. That part used to be most of the invoice on a small project, which is why small projects were rarely worth commissioning.
The website audit tool behind my own health check is a working example: 85 checks across 18 separate tools, written in Python, a language I had not used before writing it. Twenty years of knowing how software should be structured did the part the AI could not.
What did not get cheaper
- Deciding what to build. A tool that automates a process nobody has agreed on produces disagreement faster.
- The data model. Records outlive the software around them. Getting the shape wrong is the expensive mistake, and it is not one an AI will notice on your behalf.
- Security and data protection. Anything holding names, addresses or household data carries obligations under UK GDPR whatever wrote the code.
- Running it. Hosting, backups, updates and somebody to call when it stops.
Where a custom tool now beats a subscription
| Situation | Why custom wins |
|---|---|
| The process is unusual to your organisation | Off-the-shelf products serve the average case, and you are paying to work around the difference |
| Few users, long life | Per-seat pricing is cheap at three users and punitive at thirty, forever |
| Two systems that will not talk | A small piece of glue is days of work and removes a recurring manual job |
| The data must stay yours | Export is a feature you control rather than one you request |
Off-the-shelf still wins for anything standardised and regulated: payroll, accounts, email. Nobody should be commissioning those.
The risk worth naming
Code generated quickly by someone who cannot read it is a liability rather than an asset. It works until it does not, and then nobody can say why. The test of an AI-built tool is not whether it runs on the day it is delivered, but whether a competent person can pick it up in two years, understand it, and change it. That is a question about how it was structured and documented, and it is entirely a human decision.
How to tell whether yours is worth building
Count the hours currently spent on the manual version, each month, by everyone who touches it. If the answer is a day a month or more, and the process is stable enough to describe in a page, a custom tool is now cheap enough to be the obvious answer. Below that, a spreadsheet and a clear convention will serve you better.
Describe the manual job in a paragraph and I will tell you whether it is a two-day tool, a two-week one, or something you should not build at all.
More answers
- How does a small organisation work out its carbon footprint without buying software?
- What does scope 3 actually require a small organisation to measure?
- What does a net zero roadmap cost when nobody in-house does sustainability?
- How do you collect carbon data from households or members over years, not once?
- What should a website accessibility audit cover, and what does it cost?
DON'T PANIC
cyberspaceroad.com