Build or buy AI tools? How small businesses should decide
By Keith Hamilton · · 4 min read
Build or buy AI is the question that follows once you know which task you want AI to help with. You can buy a ready-made tool, switch on a feature in software you already use, or have something built for you. For most small and medium businesses the answer is to buy far more than you build, with a bespoke build kept for the rare process that no product handles. This guide sets out the options and the questions that decide it.
What are the realistic options when choosing AI software?
There are four realistic options, from least to most effort: use what you already have, buy a specialist tool, configure a platform, or commission a bespoke build. Build or buy is rarely a straight choice between two things.
Use what you already have
Many systems you already pay for, such as accounting, customer relationship and office software, now include AI features. Switching these on is the quickest and cheapest route. The limits are that they only work within that system, and you have less say over how they behave.
Buy a specialist tool
Buying AI tools built for one job, such as invoice capture, call answering or inbox triage, gets you something quick to set up and maintained by the supplier. You pay a subscription, and you accept the supplier's way of doing things.
Configure a platform
Some platforms let you assemble your own workflows using AI models and connections to your systems, with little or no code. This gives more flexibility than a specialist tool, but someone has to design, test and look after what is built.
Commission bespoke AI development
Bespoke AI development is software written for your business alone. It fits your process exactly, but it costs the most to create and you are responsible for keeping it running, secure and up to date.
Which questions decide whether to build or buy AI?
Six questions decide it: how standard the process is, whether it sets you apart, what it connects to, who looks after it, whether you can leave, and what happens to your data.
How standard is the process?
If many businesses do the task in much the same way, buy. For a common job such as processing supplier invoices, a product already exists and is cheaper than building one. Building makes sense only when the process is genuinely unusual.
Does it set you apart from competitors?
If the task is routine back-office work, buy. If it is part of what customers choose you for, there is a case for something more tailored. Even then, configuring a platform is usually enough.
What does it need to connect to?
The tool has to work with your existing systems, so check that before anything else. A cheap product that needs staff to copy information between screens costs more in time than it saves. Ask suppliers to show the connection working, not just to describe it.
Who will look after it?
Everything you build needs an owner after launch: someone to fix problems, make changes and keep it secure. If nobody in the business can do that, you depend on whoever built it. Factor that into the cost.
Can you leave?
Find out how you would get your data out, and what would happen if the supplier raised prices, changed its terms or closed. A tool you cannot leave gives the supplier a lot of power over you.
What happens to your data?
You remain responsible for customer and staff data whichever option you choose. Check where it is processed, whether it is used to train the supplier's models, and what contract terms apply. Our UK GDPR and AI checklist covers the main points.
How do you compare the cost of building and buying AI?
Compare the full cost over three years, including internal time, not the upfront price. Upfront price is a poor guide on its own.
A worked example
Here are two ways of automating one task, costed over three years:
- Buy: set-up of £2,000 and a subscription of £300 a month. Over three years that is £2,000 plus £10,800, a total of £12,800.
- Build: a build cost of £15,000, plus support and changes of £3,000 a year. Over three years that is £15,000 plus £9,000, a total of £24,000.
On these figures, buying costs just over half as much as building. Building only makes sense if it delivers a clearly larger benefit, or if no product does the job. Add the internal staff time for each option too, since a build needs more of it.
Use your own quotes and set the result against the benefit worked out in your AI business case.
Which option should you try first?
Try the simplest option that works, and go through the options in order:
- Can a feature you already pay for do the job well enough?
- If not, is there a specialist tool?
- Only then look at configuring a platform.
- Consider a bespoke build last.
Whatever you choose, start with a small pilot on one task and measure it against a baseline before committing further. It is much easier to change direction after a three-month trial than after a large build.
How do you keep AI supplier selection neutral?
Write down your requirements before you speak to suppliers, then compare every option against them. Suppliers are naturally keen on their own approach: a software house recommends a build, and a product company recommends its product.
Neither is necessarily wrong, but neither is neutral. Fixing your requirements first stops the first demonstration from setting the agenda.
Where Forwardcycle fits
Helping you set requirements, compare options and manage suppliers on your behalf is part of our commercial leadership service. If you are not yet sure which tasks are worth automating, start with the free AI Readiness Assessment.