Every few weeks a small business owner decides they need "an AI system" and starts pricing up a build. Most of them do not need one. The build versus buy question is the fork that decides whether AI saves you money or quietly becomes a project that eats it. This guide walks through the real costs, the honest trade-offs, and a way to decide that does not require a technical background.
What does "build vs buy AI" actually mean for a small business?
The words hide a lot of range. "Buy" spans a general assistant you use through a browser, a specialist app for your trade, and a paid feature bolted onto software you already run. "Build" spans a light automation wired between two apps and a full custom application with its own database. The decision is not one switch, it is a spectrum, and the skill is placing each task at the right point on it rather than treating your whole business as one project.
Should a UK small business build or buy its AI?
The reason buying wins so often is that the hard, expensive part of AI, the underlying model, is already built and shared across millions of users. When you subscribe, you rent a fraction of something that cost billions to make. When you build, you are not rebuilding the model, but you are paying to wrap it in software, connect it to your data, test it, and keep it alive. That wrapper is where small-business budgets disappear. Start by assuming you will buy, and make any build earn its place against that default.
How much does building custom AI really cost?
The Office for National Statistics reports that few UK adopters yet describe their AI use as extensive, which tells you most businesses are still learning what they need. Committing to a build before you know that is how requirements stay vague, and vague requirements are what blow budgets. The McKinsey and University of Oxford study of large IT projects, drawn from more than 5,400 projects, found they run 45% over budget and 7% over schedule on average while delivering 56% less value than predicted. Those were multi-million-pound projects, but the causes, scope creep and unclear goals, do not need scale to bite a small build. For a fuller breakdown of what AI costs at small-business scale, the guide to AI costs for UK small businesses in 2026 is worth reading alongside this.
When does buying off-the-shelf AI make more sense?
The ONS data makes the case plainly: large language models are among the most widely adopted AI technologies in UK firms, because a general assistant handles a wide spread of everyday work without any custom software. If your problem is "I spend too long writing quotes, replies and updates," a subscription solves it this week. The same pattern covers customer contact, where the mechanics of a fast, human-checked first reply are set out in the guide to AI for customer service in UK small businesses. Buying also means the vendor carries the upgrade and security burden, so your tool quietly improves without a new invoice.
When is building your own AI worth it?
These cases are real but rarer than they feel. A logistics firm with a pricing method no competitor uses, a clinic with a data-handling rule no vendor supports, a manufacturer whose machines speak a format nothing off-the-shelf reads, these can justify a build. The test is differentiation: does owning this make you meaningfully harder to copy, or are you just recreating something you could rent. If it is the latter, buy it. Before any build, write the process down so the requirement is tested, not assumed. The guide to AI business process documentation covers doing that properly.
What hidden costs catch people out on both sides?
Bought tools waste money quietly, through licences nobody uses and features that duplicate across three apps. The fix is a fortnightly review of what each subscription actually earns. Builds waste money loudly and later, when the person who made it leaves, the model it relied on changes, or a small tweak needs a developer you no longer have on call. The ONS picture, most adopters using only a couple of AI tools, is a useful discipline: fewer tools, used properly, beat a sprawling stack or a bespoke system nobody maintains.
How do you decide without a technical background?
You do not need to understand the technology to make this call, you need to understand your own work. List the jobs you want AI to help with, then sort each into buy or maybe-build using that single question. Most will land firmly in buy. For the handful that do not, run them manually through a bought assistant such as Claude for a month, measure what they are worth, and let that number decide whether a build is justified. A decision made against real usage beats one made against a sales pitch.
Is there a middle path between build and buy?
This middle path is often mislabelled as "building." Wiring a bought assistant into your inbox, setting up a shared prompt library, or connecting two apps so information flows automatically is configuration, not development. It carries a fraction of the risk, needs no ongoing developer, and moves with you if you change tools later. Treat true from-scratch building as the last resort, reached only when configuration has been tried and genuinely falls short.
How do you avoid the most common build-vs-buy mistake?
Owners talk themselves into builds for reasons that are emotional, not commercial: the sense that owning software is safer, or that a custom system signals ambition. The McKinsey and Oxford overrun figures are the antidote. If projects at that scale, with professional teams, still average 45% over budget and deliver 56% less value than predicted, a small business with no in-house engineering should treat a build as a serious commitment, not a default. Prove the need first, in real use, before anyone writes bespoke code.
What should you do first this month?
Start small and measured rather than big and speculative. Choose the jobs you dread, apply a bought tool with a clear prompt and a human check, and track two numbers: time saved and quality. Most owners find the results settle the question before a build is ever on the table. The point of build versus buy is not to pick a side in the abstract, it is to spend the least money for the most useful, reliable help, and to keep the risk where a vendor can carry it rather than where you have to.
Sources: ONS, Artificial intelligence in UK businesses: 2023 to 2026 and McKinsey and University of Oxford, Delivering large-scale IT projects on time, on budget, and on value.
AI Build vs Buy — FAQ
What does build vs buy AI mean?
It is the choice between building a custom AI system for your business, usually with developers and a bespoke integration, or buying an existing tool and configuring it to your needs. For most UK small businesses, buying means subscribing to a general assistant or a specialist app, while building means paying someone to create something new around your data and processes. The honest starting position is that buying wins far more often than owners expect, because the off-the-shelf tools are now capable and cheap, and a custom build carries cost and risk that only pay off in narrow cases. Build for genuine differentiation, buy for everything else.
Is it cheaper to build or buy AI for a small business?
Buying is almost always cheaper to start and cheaper to run at small-business scale. A subscription to a capable assistant costs a fixed monthly fee with no development bill, while a custom build front-loads design, engineering and testing before it earns anything. The McKinsey and University of Oxford study of large IT projects found they run 45% over budget on average, which is a caution about how build costs escalate. Small builds are smaller, but the same forces apply. Buy first, prove the value, and only build when a bought tool genuinely cannot do the job.
When should a small business build its own AI?
Build when the capability is core to how you compete and no bought tool fits, when you have a genuinely unusual process or dataset, or when integration with your existing systems is the whole point and off-the-shelf connectors do not reach it. Even then, start by proving the value with a bought tool used manually, so you build against a tested requirement rather than a guess. Most owners who think they need a build actually need better configuration of something that already exists. Building for a task a subscription handles well is the most expensive mistake in this decision.
How much AI does the average UK business actually use?
Less than the headlines suggest. According to the Office for National Statistics, adoption among UK businesses is still wide but shallow: many firms have tried at least one AI tool, yet most use only a couple and few describe their use as extensive. Large language models, the kind behind general assistants, are among the most widely adopted types. That pattern, broad experimentation rather than deep commitment, is a strong argument for buying and learning first. It lets you find out what genuinely helps your business before you spend anything on a custom build you might not need. Most owners discover a subscription already covers the work they had in mind.
What is the biggest risk in a custom AI build?
Cost and time overrunning while the value fails to arrive. The McKinsey and Oxford research on large IT projects found they run 45% over budget and 7% over schedule on average, and deliver 56% less value than predicted. Those are large projects, but the causes, unclear requirements, scope creep and thin ownership, scale straight down to a small build. The second risk is maintenance: a build is not finished when it ships, it needs someone to keep it running as models and your business change. A bought tool moves that burden to the vendor.
