Build, Buy or Switch On: Deciding Where Your AI Comes From

Most "build vs buy" decisions about AI skip a third option: switching on what you already pay for. Check that first, because the AI features inside your existing software are often enough. Buy a specialist tool when they are not. Build only where your own knowledge and methods are what make the work valuable.

What does build, buy or switch on mean?

Switching on means using AI features inside software you already license, such as your CRM, office suite or help desk. Buying means paying for a new specialist AI product. Building means assembling something of your own, usually on top of a bought model. Each has a different cost, speed and level of control.

Almost nobody builds from scratch any more. "Build" now usually means writing down your method, giving the AI your business knowledge and connecting it to your systems. The model underneath is bought.

Why start with what you already pay for?

Because it costs nothing extra to test and often does the job. Much business software now includes AI features, and teams often have not tried them. Checking them first avoids adding another subscription for something you already own, and gives you a quick baseline to judge anything else against.

Many businesses already pay for more AI than they use. A team might hold an AI assistant subscription, a CRM with AI features and an office suite with a built-in copilot. Yet they still shop for a new tool to summarise calls. Sometimes the existing ones can do it once switched on and set up.

This is how AI tool sprawl starts: each team buys its own answer to a problem a licensed tool already solved. A quick audit of what is already paid for usually pays for itself.

Switching on has limits. A built-in feature knows only what is in that one tool. So if the job needs knowledge spread across your CRM, documents and calls, the feature will produce generic output.

When is buying the right answer?

Buy when a specialist product does the job well, the job is common across businesses, and your own knowledge is not what makes the output valuable. Transcription, scheduling and spam filtering are good examples. Buying gets you something proven, supported and improving, without anyone in your business having to maintain it.

The evidence favours buying over building alone. According to MIT NANDA's 2025 research, as reported by Fortune, buying AI tools from specialised vendors and building partnerships succeeded about 67% of the time. Internal builds succeeded far less often. Specialists have already solved problems an internal team would meet for the first time.

Buy carefully, though. According to Gartner, only about 130 of the thousands of vendors calling their products "agentic" are real, a practice it calls "agent washing". Ask every vendor to show how the product was tested, where your data goes and how you would leave. The AI vendor due diligence guide sets out the questions.

When is building worth it?

Build when your own knowledge or method is what makes the work good. Proposals written in your voice from your past wins, sales follow-ups that use your objection handling, reports in your house format: no off-the-shelf product knows these things. Build that layer, and build it on bought models and tools.

Building badly is common. MIT NANDA's study, as reported by THE Journal, found only 5% of custom enterprise AI tools reach production. So the build should be small and specific: one workflow, its method written down and its knowledge supplied.

Start simple. According to Anthropic's engineering guidance, the best approach is "finding the simplest solution possible, and only increasing complexity when needed". A written method plus the right business knowledge often beats an elaborate agent.

Ownership is the other reason to build. A method and knowledge base held in your own files stay with you if you change tools or suppliers. The guide to who owns your company's AI explains why that matters.

What does each option really cost?

Switching on costs setup time and little else. Buying adds a subscription and some setup, plus the cost of moving if you leave. Building costs the most time upfront and needs an owner to keep it current. The cheapest option to start is not always the cheapest to keep, so compare all three over a year.

Upkeep is where builds go wrong. A built workflow that nobody owns drifts out of date as the business changes. Before building, name the person who will keep it current.

How do you decide for one workflow?

Take one workflow and ask three questions in order. Can a tool we already pay for do this well enough? If not, does a specialist product do it well, without needing our own knowledge? If not, is the job valuable enough to build, and who will own it? Stop at the first yes.

1. Switch on? Test the existing feature on real work for two weeks.
2. Buy? Trial a specialist product against the same measure.
3. Build? Only if neither works and your knowledge is what makes the output good.
4. Wait? If nothing passes, the honest answer may be not yet.

The fourth answer matters. Some jobs are not worth AI work today, because the tools are immature, the data is missing or the job is too rare.

Skein works in this order: if a tool the client already pays for can do the job, it is switched on rather than built, and Skein says so when the answer is to wait; the AI engineering page shows what gets built when building is right. That suits a growing B2B business deciding where to spend on AI. A business that needs one specialist product configured may be better served by that vendor's own implementation partner.

Frequently asked questions

Should a small business build or buy AI?

Most should first check AI features in software they already pay for, then buy specialist tools for common jobs. Building suits work where the business's own knowledge and methods make the output valuable, such as proposals or follow-ups. Even then, build on bought models rather than from scratch.

What does switching on AI mean?

It means using AI features already included in software you license, such as a CRM, office suite or AI assistant, rather than buying or building something new. Some may be switched off or never set up. Testing them first is the cheapest way to find out what AI can do.

Is it cheaper to build or buy AI?

Buying is usually cheaper to start and building can be cheaper to keep, but only if someone maintains it. Compare costs over a year: setup, subscriptions or usage, upkeep and the cost of leaving. Switching on existing features is often cheapest of all.

Why do internal AI builds fail more often?

Internal teams meet problems for the first time that specialist vendors have already solved, and many builds are too broad. Builds also decay when nobody owns them. MIT NANDA's research, as reported by Fortune, found internal builds succeeded far less often than buying from specialists or partnering.

When should you build your own AI?

Build when your own knowledge or method is what makes the work good, and no product can know it: your voice, your past wins, your objection handling. Keep the build small, write the method down, supply the knowledge it needs and name an owner before you start.

Related reading