Build, buy, or automate: a simple test
Before paying for another tool or a custom build, decide which of the three your problem actually needs.
Every operational problem has three candidate fixes: buy an off-the-shelf tool, build custom software, or automate the existing workflow with a thin AI layer over the systems you already run. The right choice follows from one measurement, the monthly cost of the manual process, set against how well each fix removes it.
Each wrong choice is expensive in its own way: buy wrongly and the process bends around someone else's product; build wrongly and a long software project replaces a problem a subscription could have solved; automate wrongly and AI gets wrapped around a workflow that should simply have been retired.
When should you buy an off-the-shelf tool?
Buy when a product fits the job as your process actually runs, without contortion. The monthly fee for a fitting tool is almost always cheaper than any build, and an audit that always concludes "hire us" is a brochure, so we make the buy recommendation whenever the numbers point that way, even when it ends the conversation.
The trap is the tool that almost fits. The gap between what it does and what the job needs gets filled with manual work, and that manual work never appears on an invoice, so the tool looks cheap while the process stays expensive.
When is automating the existing workflow the right answer?
Automate when the work is repetitive, spans tools you already pay for, and needs judgement at only a few steps: reading a document, classifying a request, drafting a reply. This is where most SMEs find their fastest payback, because the systems stay, the process stays, and the build is a thin layer that removes the copying and chasing between them. One of our builds did this for a UK professional-services firm whose staff moved the same fields between four systems every week; manual re-keying fell by 78% and the first shared workflow was live in 2 weeks.
When does a custom build earn its cost?
Build when the workflow, the data or the compliance requirements are specific enough that no product will ever quite fit, and the manual cost is large enough to justify the price. Pricing logic is a common example: one of our builds encoded a Yorkshire construction services firm's own approved pricing rules into its quoting workflow, and first quote drafts came out 63% faster, work no generic product could have carried because the rules were the firm's own.
The discipline is running the three options in the right order. Price the manual process first, check for a tool that fits, check whether a thin automation over existing systems removes the cost, and reserve custom software for the cases that survive both filters. Most problems fall to the second or third option, which is why the honest answer is usually also the cheapest one.
What AI governance actually means for a board
The real test of AI governance is whether you can reconstruct what a system did and why, months later, for someone with the power to penalise you.
Why we measure every AI build against one number
Most AI programmes stall on the choice of first project. Scoring every candidate on monthly payback settles the choice with arithmetic instead of opinion.