Every software vendor has bolted “AI” onto their product this year. Your CRM has it. Your accounts package has it. There’s a chatbot on half the websites you visit. So when someone in your business says “we should be using AI,” the honest first question is: to do what, exactly?
That question matters, because the answer usually points you down one of two very different routes. You either buy AI that already exists and fit your business around it, or you build something that fits your business. Most companies end up doing a bit of both. The trick is knowing which job belongs in which bucket, so you don’t overpay for a bespoke build you didn’t need, or force a generic tool to do work it was never designed for.
Here’s how we think about it.
When off-the-shelf is the right call
If the problem you’re solving is a common one, buy the common solution. Summarising documents, drafting first-pass emails, transcribing meetings, writing marketing copy, answering routine questions: thousands of businesses need these, so there are mature, cheap, well-supported tools that already do them well. Building your own version would be slower, more expensive and worse.
Off-the-shelf wins when:
- The task isn’t specific to your business. Anyone in any industry would recognise it.
- Your own data isn’t the point. You’re using the tool’s intelligence, not teaching it about your world.
- You can live with the tool’s way of working. You adapt to it, not the other way round.
- You want it running next week, not next quarter.
The catch is that everyone else has the same tool. It won’t give you an edge. It just stops you falling behind, which is a perfectly good reason to use it.
When custom is worth it
Custom AI earns its cost when the value is locked up in something only your business has: your data, your process, your particular way of making a decision.
Say you’re a housebuilder and you want to spot which plot reservations are most likely to fall through, based on years of your own sales history. No off-the-shelf tool has your data or understands your pipeline. Or you run a field service operation and you want to route jobs using rules that currently live in three experienced schedulers’ heads. That knowledge is yours, and a generic tool can’t learn it.
Custom is the right route when:
- The problem is specific to how your business works.
- Your own data is the asset, and the results get better because the system learns from it.
- The decision it supports is valuable enough that a better answer pays for itself.
- It needs to sit inside your existing systems, not as yet another separate login.
Custom costs more up front and takes longer. What you get is something a competitor can’t simply go and buy.
The bit people skip: total cost of ownership
An off-the-shelf subscription looks cheap next to a custom build. But “cheap” is the monthly figure, not the whole cost. Add the per-seat licences as you grow, the workarounds for when the tool almost-but-not-quite fits, the data you can’t easily get back out, and the fact that the price is set by the vendor and only ever moves one way.
A custom build is the opposite shape: more to begin with, then yours. No per-seat tax, no feature you’re waiting on someone else to ship, and it does exactly the job you built it for.
Neither is automatically cheaper. It depends on how many people use it, how long you’ll run it, and how central it is to what you do. We’d rather help you work that out honestly than sell you the more expensive option by default.
A simple way to decide
Before you commit either way, three questions get you most of the answer:
- Is this problem unique to us, or does every business have it? Common problem, buy it. Unique problem, consider building.
- Is our data doing the work? If the value comes from your data, off-the-shelf can’t capture it.
- How much is a better decision worth? If it’s marginal, don’t build. If it changes real money or real risk, a custom system pays back.
If you’d like a straight answer for a specific problem you’ve got in mind, that’s the kind of conversation we have all the time.