The AI tools market is loud, but for a small business it collapses into three tiers, and the right tier follows a simple rule: buy AI where your work already lives. Here's the landscape without the hype, and the decision path we walk clients through.
Tier one: the AI built into what you own
Microsoft 365 Copilot (around $30 per user per month on top of your plan) and Google's Gemini features sit inside the mail, documents, and spreadsheets your team already uses. Drafting in Word, meeting summaries in Teams, formula help in Sheets, all without new logins, new training, or new data agreements beyond the vendor you've already trusted with the data.
The honest tradeoff: convenience over peak quality. Built-in assistants are rarely the best model at any given task, and per-user pricing across a whole company adds up fast for features some staff won't touch. Start with licenses for your five heaviest document-and-email people, measure, then expand or don't.
Tier two: standalone assistants
Claude, ChatGPT, and their business plans, roughly $20 to $30 per user per month, used through their own apps. Stronger at everything in the five jobs than the embedded features: longer documents, better reasoning, better drafts. The business tiers matter for a specific reason: they come with admin controls and, critically, terms where your inputs aren't used for model training. Consumer free tiers are where client data goes to leak, which is why the tier decision belongs in your AI policy, not in each employee's browser habits.
The tradeoff is the copy-paste seam: work leaves the document, visits the assistant, comes back. Mild friction, real value. Most businesses getting daily AI value live in this tier.
Tier three: custom, through the API
When a workflow repeats hundreds of times a month, the seam becomes the cost, and it's worth wiring AI directly into your systems: every inbound invoice extracted into the accounting queue automatically, every support email pre-triaged, every contract summarized on arrival. This is a small software project, not a subscription: modest build cost, then API usage billing that's often cheaper per task than seat licenses. If the workflow came out of a successful tier-two pilot, the build has already proven its math. What it looks like in practice: a first Claude API feature, and the sober version of adding AI to your app.
The decision path
Start at tier two with a pilot group; it's the fastest way to learn what your business actually uses AI for. Add tier one where the embedded convenience clearly beats the seam. Graduate specific high-volume workflows to tier three when the measurements justify a build. And ignore, categorically, the industry-specific "AI platform" priced at 10x a general assistant with a thin wrapper on top. Ask any vendor what model runs underneath and what happens to your data. The pause tells you plenty.
Want this handled instead of homeworked? That's the job.
Email us →