Home · Insights · Fundamentals
Do you need custom AI, or will off-the-shelf tools work?
"Build, don't buy" is the loudest AI advice going around right now, and it comes from companies with an engineering department. A business with five people and no engineers is answering a different question, and the advice that gets given usually skips straight past that difference.
The trend that is actually happening, and who it is actually about
McKinsey's State of AI survey, published November 2025, found that 32% of organisations have decided against buying a software product or feature because it could be built internally with agentic coding tools instead. In tech companies specifically, that figure is 41%.
That is a real and fairly sudden shift, and it is worth understanding why it happened before deciding it applies to you. Agentic coding tools have made a category of software cheap to build for the first time, but only for organisations that already have engineers who can direct those tools, review the output, and own what gets shipped. The survey population is enterprises. The finding is about engineering capacity that already exists, being redirected toward building instead of procuring.
What the same period looked like for small businesses
The U.S. Census Bureau's Business Trends and Outlook Survey, published May 2026 from data collected through May 3, 2026, measured AI use across firms of every size. Firms with 250 or more employees: 37% use AI. Firms with 4 or fewer employees: under 20%. The gap did not hold steady, it widened, because AI use kept climbing at firms with 20 or more employees while it stayed flat at firms smaller than that.
Put the two findings next to each other and the picture is not "everyone is building now." It is two different worlds moving in different directions: large organisations with engineering teams building more of their own software, and almost everyone below 20 employees not moving much at all. A five-person business reading "stop buying, start building" headlines is being handed advice measured on a population it does not belong to.
The actual question for a small business
"Custom AI" in the enterprise conversation usually means a model, an agent framework, or a software product built and maintained by an internal engineering team. Almost no business the size of a typical Mallorca operation needs that, and almost none should try to build it. The real choice sitting in front of a small business owner is narrower and more practical than the headlines suggest:
| Off-the-shelf, as-is | A wired system |
|---|---|
| A tool with an AI feature bolted on, used the way it ships | The same tools, connected to each other and to how you actually work |
| Fast to start, cheap up front | Costs more to set up, keeps the value once it is running |
| Works for a generic version of your process | Works for your actual process, including the messy parts |
| You do the connecting yourself, or nobody does | Enquiries, follow-ups, and handoffs move without someone remembering to |
That right-hand column is not "build custom AI." It is implementation: taking tools that already exist, most of which a small business is already paying for in some form, and wiring them so information moves between them without a person carrying it by hand. That distinction is the one we go into in full in what AI implementation actually means, and it is the difference between a tool sitting on top of a business and a system running underneath it.
A quick way to tell which one you actually need
- Does an off-the-shelf tool already do this well on its own? Transcribing a call, drafting a first-pass reply, summarising a document: buy it. This is a solved problem and building it yourself is wasted effort.
- Is the gap between tools, not inside one of them? A lead fills in a form, and nobody tells the CRM, and nobody messages them back for six hours. No single tool "solves" that, because the problem lives in the handoff. That is implementation work, not a purchase.
- Does it depend on something specific to your business that no vendor could ship generically? Your own pricing logic, your own qualification criteria, your own follow-up cadence. That is also implementation: configuring and connecting real tools around your specifics, not commissioning a model from scratch.
Almost every real gap we find when we look at a business's funnel and back office falls into #2 or #3. Genuine "we need a model built from nothing" situations are rare enough that if you are unsure whether you are in one, you almost certainly are not.
Why this matters more than it looks like it should
Getting this wrong costs money in a specific way. Buying five separate AI-branded tools because each one looked good in a demo, then never connecting them, produces five subscriptions and the same handoff gaps you started with. Trying to commission a custom build for a problem an existing tool already solves burns a budget on something you could have had working the same week. The expensive mistake is not picking the wrong side of build-versus-buy. It is answering a question that was never actually yours to answer, because the real work in the small-business version of this is neither one.