Should you build, buy or wait on AI?
For most SMEs, buying and configuring existing AI tools is the sensible default. Build only where the workflow is a real differentiator you can maintain.
Why buying is usually the right default
Off-the-shelf AI tools improve quickly and someone else maintains them. That includes assistants built into software you already use. For most SMEs, buying and configuring these tools delivers most of the available value at a fraction of the cost and risk of building.
Buying makes sense when:
The task is common: Many businesses share needs like drafting, summarising, scheduling, transcription and customer service triage, so mature products already exist.
Speed matters: You can test a bought tool within days and drop it if it does not fit.
You lack in-house technical capacity: Someone has to fix, update and secure anything you build. If that person does not exist, buy.
Your data fits standard tools: Your information does not need unusual handling or specialist models.
The trade-off is control. You accept the vendor's roadmap, pricing changes and data terms. Before you commit, check where data is stored and whether it is used to train models.
When building is worth it
Build only where the workflow gives you a genuine edge over competitors and you can maintain what you build. Good reasons to build include:
A real differentiator: The process is how you win work, such as a proprietary pricing method or a specialised way of assessing clients.
No suitable product exists: You have tested the market and nothing handles your workflow without heavy workarounds.
You can own it long term: You have, or will fund, the skills to monitor, update and secure the tool as the underlying models change.
The value justifies the effort: The expected benefit clearly exceeds the build cost plus years of maintenance.
A middle route often works best. Buy a capable platform and configure it with your own templates, prompts, knowledge and workflow automation. You get a tailored result without owning the underlying technology.
The hidden cost of waiting
Waiting can look prudent while tools change so quickly, but it has costs of its own:
Staff move anyway: If the business offers nothing, people use AI on personal accounts. That creates shadow AI, with no oversight of what data goes where.
Competitors build capability: The advantage rarely comes from the tool itself. It comes from teams that have learned to use it well, and that learning takes time to catch up on.
Skills gaps widen: The longer you delay, the larger the training effort when you do start.
Waiting is reasonable in a few narrow cases. Your regulator may not yet have set clear expectations for a specific use, a vendor you rely on may be about to release a relevant feature, or the business may have more pressing priorities. Even then, set a review date and do not leave the decision open.
Decision criteria to apply
For each candidate use case, put these questions to your leadership team:
Is this how you compete? If not, buy.
Does a product already do most of it? If so, buy and configure.
Can you maintain what you build? If not, do not build.
What happens if you do nothing for a year? If the answer is staff working around you, waiting is the riskier choice.
The answer depends on your sector, how sensitive your data is and your technical capacity. It can also differ between use cases in the same business. If you are still deciding where to begin, where a small business should start with AI sets out a low-risk first step.