Custom AI software or off the shelf? When building wins
Most businesses should buy, not build. How to tell when you're the exception, from Yorkshire developers who build bespoke AI tools and will say when you don't need one.
We’re custom AI developers. Our business is building things that don’t exist yet. So you’d expect this piece to argue that your business needs bespoke software.
It probably doesn’t. Most businesses, most of the time, should buy something off the shelf, configure it, and get on with their lives. Bespoke software is expensive, slow and yours to maintain forever. The case for it is narrow, and when it applies it’s overwhelming. Here’s how to tell which side of the line you’re on.
Buy it if someone’s already built it
Accounting, payroll, CRM, email marketing, project management, HR, booking, e-commerce, helpdesk. These are solved problems with mature products, most of which now have AI features built in that are better than anything a small firm could commission. If your need is one of these, buy the product, pay for the tier that doesn’t train on your data, and spend your money on getting your people to use it well.
The same goes for most “AI tools”. A general-purpose assistant on a business plan covers drafting, summarising, research and the long tail of daily tasks. The ONS found in July 2026 that the typical adopting business uses 1.6 AI technologies (ONS, 20 July 2026), and for a lot of firms that’s genuinely enough. Shallow but sensible.
Automate it if the problem is between the products
The middle ground, and where most of the value sits for a small business, is not building software at all. It’s connecting the software you’ve already got so information stops being re-typed between it.
An enquiry from the website into the CRM with a draft reply. An approved quote into the job system and the accounts package. A supplier invoice read, matched and queued. These are automations built on tools like Make or n8n with an AI model doing the reading, and they cost a few thousand pounds rather than tens of thousands. We’ve compared the tools and set out the costs separately. Exhaust this option before you consider building.
Build it when one of these is true
The job is your business, not a department of it. If the process is the thing you sell, or the thing that makes you different, off-the-shelf software will fit it the way a suit from a catalogue fits: roughly, and everyone can tell. Our care home client’s entire operation runs on the governance process that Guardian Care encodes. No generic compliance tool would have fitted, because the process is the product.
You’ve tried the products and you’re fighting all of them. Three subscriptions, two spreadsheets and a person whose job is reconciling them is a sign. If you’re paying for five tools and using 20% of each, a single thing built for your workflow can be cheaper to run and far cheaper in time.
The data is sensitive and the generic tools want it in their cloud. Care, legal, health, finance, anything with children or vulnerable adults in it. Bespoke means you decide where it lives, which models see it and under what terms. The ICO’s guidance on AI and data protection is the baseline either way, but it’s much easier to meet when you control the architecture. Note that the guidance is marked as under review following the Data (Use and Access) Act 2025, so check for updates.
The AI needs to know things only your business knows. Your price book, your forty years of past jobs, your rules about which customer gets which terms. A generic tool can’t be taught this well. A bespoke one is built around it.
You want to sell it. Sometimes the tool you need is the tool your whole sector needs, and the thing you’re building is a product. That’s a different conversation and a much bigger one; we’ve had it ourselves with VinylPeek, which started as a side project and became a subscription business.
What building actually involves
Being straight about this, because the pitch usually isn’t.
It takes longer than the quote says. Not because builders lie, but because the first version reveals what the second version needs to be. Budget for iteration. A fixed-scope first phase with a fixed price is the right shape; a fixed price for “the whole thing” is a fiction.
It needs someone on your side. A person in your business who knows the process cold, is available to answer questions, and has the authority to say “no, that’s not how we do it”. Projects without that person stall.
It’s yours to maintain. Models get retired. Browsers change. Security patches need applying. Either budget a retainer or accept you’ll pay for fixes as they come. Software that nobody looks after stops working, and bespoke software has nobody but you to look after it.
It should start small. Even a platform should ship something usable in weeks, not months, so you find out early whether the idea survives contact with the people who’ll use it.
The honest test
Ask: if a competitor had this exact tool, would it worry me? If the answer is no, buy something. If the answer is “yes, a lot”, that’s the signal that it’s worth building, because it means the thing you’d be building is an advantage, not a cost.
We’d rather you got this right than hired us for the wrong thing. If you’re not sure which side of the line your idea sits on, describe it to us and we’ll tell you honestly. About a third of the time, the answer is “don’t build it”, and that’s a good afternoon’s work for everyone.
