Most advice about AI is written for companies with data science teams and seven-figure budgets. A firm with $5 million in revenue or $300 million does not have either, and does not need them. What it needs is a short list of places where AI changes the economics of the business, a sensible way to acquire the tools, and a disciplined way to find out whether they work.
Treat AI as an operating lever
For a small or mid-sized business, AI is not a technology program. It changes what a person can get done in a day, how many transactions a team can handle, and how much of a larger competitor’s capability you can reach without their overhead. Judge every use by a business measure: hours per transaction, margin, errors, capacity or how quickly you win customers.
A 30-person firm and a 1,000-person company are different problems
| Small business | Mid-sized business | |
|---|---|---|
| What AI is for | Operating like a much larger competitor without the overhead | Removing labor bottlenecks and adding capacity |
| How it is rolled out | Owner-led, quick, using the AI built into software you already pay for | Cross-functional, integrated into governed workflows |
| Technical support | No data science team; you rely on your vendors | IT coordination, integrations and data security |
| Governance | Simple written rules and direct owner oversight | Formal policy, role-based permissions, audit logs and exception paths |
A small firm can use AI for market research, round-the-clock customer questions, variance analysis and personalized outreach that once took a team ten times its size. A mid-sized company can use it to get past labor shortages in routine work such as payables and receivables, claims, scheduling and procurement, as long as its controls keep up.
Buy, configure, integrate or build
The common mistake is to assume that more sophisticated AI means custom software. In most cases the opposite is true. Sophistication is choosing the right way to acquire each capability.
| Capability | Best approach | Why |
|---|---|---|
| Writing, summarizing, research | Buy off the shelf | A commodity. There is no advantage in building it |
| Assistants inside the tools your people already use | Buy from the platform you already have | Security, user management and scale come with it |
| Reading and extracting from documents | Buy and configure | Pre-trained tools need your templates and rules, not new code |
| Workflow automation | Configure and integrate | Uses the interfaces of software you already own |
| Software agents that carry out bounded tasks | Integrate, with some custom work | They must connect to your own systems and rules |
| Something only you can do, built on your own data | Build, deliberately | This is where a defensible advantage comes from |
The default is to buy before you build. Custom development earns its cost only where your data, your expertise or your customer relationships make the result hard for anyone else to copy.
Do not climb a ladder
AI creates value in several different ways, and a business does not have to master the simple ones before it tries the harder ones. A mid-sized company can give every employee a licensed assistant this quarter while it builds an AI-enabled product in its fastest-growing unit. We make the case in Your AI strategy is not a ladder.
Run each pilot as a bounded experiment
An AI pilot should be small, owned and measured. Eight steps keep it honest:
- Name a specific point of operational friction.
- Decide which kind of AI use it is, and how you will acquire it.
- Set the scope and say what is excluded.
- Assign one accountable business owner.
- Confirm what data the tool may use, and under which security rules.
- Measure today’s time, error rate or cost, so there is a baseline.
- Set approval limits and the conditions under which you stop.
- Set a decision date, usually 30 to 90 days out, to scale, revise or stop.
At the review, ask three questions:
- Does it pay for itself? Do the saved hours, added volume or new revenue exceed the license and integration costs?
- Is it reliable? Did errors stay within your standard without creating hidden review work for supervisors?
- Can you live with it? Can your own staff keep it running, without a new dependency on one vendor or an unmanaged security gap?
Where governance fits
Each of these steps carries risk: what data goes into a tool, which suppliers are adding AI to products you already use, and what a person must check before a result is used. Small firms need a page of clear rules. Mid-sized companies need approval tiers, permissions and records. Either way, AI governance is part of the plan, not a step you add after something goes wrong.