Build vs buy vs integrate: how should an SME add AI to an existing tool?

Most SMEs frame adding AI as "build it ourselves or buy a tool", and quietly ignore the option that usually wins: integrate - glue AI into the systems you already run. Building is slow and risky, buying bolts on yet another silo, but integrating lets you add intelligence to your existing tools without a rebuild or a migration. Here is how to choose.

I connect AI to systems for a living, so I have no stake in selling you a build or a licence. But the "build vs buy" framing itself is the problem: it hides the third option, and the third option is the right one more often than not.

Build vs buy vs integrate: what does each actually mean?

• Build means commissioning custom software - your own models, or bespoke applications wrapped around someone else's - owned and maintained by you.
• Buy means licensing a finished product: an AI assistant, a vertical SaaS tool, a point solution that does one job out of the box.
• Integrate means wiring AI into the tools you already have - your CRM, inbox, spreadsheets and database - through a thin orchestration or gateway layer, so intelligence is added to your existing workflow rather than moved into a new one.

The first two are the options every vendor frames the decision around. The third is the one that leaves your data where it is and your team where they are.

How do the three compare?

	Build	Buy	Integrate
Time to value	Months	Days	Days to weeks
Upfront cost	High	Low (recurring licence)	Low to moderate
Fit to your process	Exact - you built it	Whatever the product assumes	High - shaped around your tools
Maintenance burden	Entirely yours	The vendor's	The glue layer only
Data location	Yours	Their ecosystem	Stays in your systems
Lock-in risk	Low, but you own the upkeep	High - another silo	Low - swap components freely
Best when	The process is your core edge	A tool already nails the job	You have good tools, no join

No column is wrong. Build when the process is genuinely your competitive advantage and nothing off the shelf fits. Buy when a product already does the job well and you are happy to live inside it. Integrate when you have decent tools that simply do not talk to each other - which describes most SMEs.

Why is "build vs buy" the wrong question for most SMEs?

Because it presents a false choice between two expensive extremes while ignoring the assets you already own. Building AI in-house is beyond most small teams: the 2025 Stanford AI Index reports the median cost of training a frontier model now runs into the tens of millions of dollars, and even a modest bespoke build carries real engineering and maintenance load (Stanford HAI). Buying looks like the safe default, but each new tool is another place your data lives and another subscription to leave - and the purchase on its own rarely turns into value. McKinsey's 2025 survey of nearly 2,000 organisations across 105 countries found that 88% now use AI in at least one business function, yet only around a third have begun to scale it - most are still experimenting or piloting, without material benefit at the enterprise level (McKinsey). Buying AI is the easy part. Getting it to reach the way the business actually runs is where it stalls. Integration sidesteps both failure modes: you are not funding a moonshot, and you are not buying a fourth system that does not know about the other three.

Doesn't buying a tool already "integrate" AI for me?

Only inside its own walls. A bought AI tool integrates AI with itself - its data, its interface, its workflow. What it does not do is integrate with the CRM your sales team actually uses, the inbox your operations run from, or the spreadsheet finance closes the month in. That gap is where value leaks: the AI is smart about the data it can see, and blind to everything sitting in your other systems. True integration means the intelligence reaches across all of them, not just the one you bought.

When is building genuinely the right call?

When the process you are automating is the business, and doing it better than anyone else is your edge. If your differentiator is a proprietary matching algorithm, a unique underwriting model, or a workflow no competitor can copy, a bought tool commoditises you and integration alone will not express it. That is the moment a build pays back. For everything else - the internal, operational, "we just need these systems to talk" work that fills most SME backlogs - a custom build is usually cost and risk you do not need to take on.

How does integration avoid the lock-in that buying creates?

By keeping the pieces loosely coupled, so no single vendor owns your workflow. This is the same principle behind avoiding vendor lock-in with single-platform AI assistants: your system of record stays yours, a thin layer in the middle routes each task to whichever model or tool is best today, and you can swap any component tomorrow without touching the rest. Buy a monolithic tool and its data, prompts and workflows accrete inside one supplier's ecosystem until leaving means a rebuild. Integrate, and the AI is a component you can replace - not a foundation you are stuck on.

How do you actually run the integrate route?

You do not need a platform migration or a big-bang project. You need a small amount of glue, added where it earns its keep:

• Start from a real process, not a tool. Pick one workflow that spans systems today - a lead that should flow from inbox to CRM, an invoice that should reconcile itself - and automate that, not "AI in general". If you are unsure a process is ready, run it through this readiness test first.
• Keep your systems of record. Data stays in your CRM, database or warehouse. The AI reads and writes to them; it does not become the new home for your data.
• Put an orchestration layer in the middle. A tool such as n8n, or an AI gateway, connects your apps and routes each task to the right model - so you can change providers without rewiring the workflow.
• Add intelligence one step at a time. Classify here, extract there, draft a reply somewhere else. Each addition is small, testable and reversible.

Done this way, integration gives you much of the fit of a build at a fraction of the cost, without the extra silo a purchase adds.

The bottom line

"Build vs buy" is a choice between the most expensive option and the one that adds another silo. For most SMEs the answer is neither: you already own good tools, and what you are missing is the join between them. Integrating AI into your existing stack - adding intelligence without moving your data or retraining your team - is faster than building, stickier than buying, and it compounds on the systems you have rather than replacing them.

That join is exactly what I build. If you have the tools but not the connections between them, that is the kind of thing a thirty-minute discovery call sorts out - I wire your systems together so AI works across all of them, on your terms. See process automation and AI consulting for how I approach it.