The five levels of AI in a small business (and how to find yours)

Ask ten small businesses whether they "use AI" and ten say yes. Ask what that means and you get ten wildly different answers, from "I asked ChatGPT to write a birthday post once" to "our support desk drafts its own replies." Both count as using AI. They are not remotely the same thing.

There is a ladder here, and knowing which rung you are on matters more than any tool list, because the right next move is completely different at each level. Here are the five levels we see, the tell for each one, and the single move that takes you up.

Level 1: Dabbler

The tell: one or two people paste prompts into ChatGPT or Claude when they think of it. Results live and die in the chat window. Nothing is shared, nothing is written down.

This is where most small businesses and nonprofits are right now, and it is a fine place to be. Surveys of nonprofits put roughly four in ten organizations exactly here: a person or two, learning on their own.

What to do at this level: pick one repetitive writing task and do it with AI every time for two weeks. Emails, proposals, meeting summaries. Nothing else.

The move up: make it deliberate. Agree as a team that AI is something the business uses, write one page of rules (what never gets pasted in), and turn on the AI features in the tools you already pay for.

Level 2: Adopter

The tell: the org has decided AI is a thing it does. Copilot features are on in your email or CRM, there is a policy, maybe there was a training lunch. But everyone still works alone, and nobody saves what works.

What to do at this level: map each role to its two best AI tasks, and start measuring roughly what time it saves. Keep it honest; self-reported hours are fine.

The move up: reuse. The level boundary is one question: does anyone reuse saved prompts, or a custom assistant someone set up? The day your best prompt becomes a shared, named thing instead of one person's private trick, you have changed levels.

Level 3: Systematizer

The tell: saved prompt libraries, Claude Projects or custom GPTs set up for recurring jobs, prompts paired with your how-we-do-it docs. New hires inherit the good prompts instead of reinventing them.

This is the fastest-growing level, and the least talked about. It is also where AI stops being a personal productivity trick and starts being a business capability.

What to do at this level: build assistants for your two or three highest-volume jobs, and pair every standard procedure with the prompt that does it.

The move up: connect your information. Everything below this line runs on what someone pastes in. The next level runs on your actual data, which means your data has to live somewhere organized enough to connect. That single unlock, getting your information AI-ready, is the most valuable move on the whole ladder.

Level 4: Automator

The tell: AI runs as a step inside a workflow without a person kicking it off each time. A form comes in, a draft goes out, a human checks a queue. Your tools talk to each other, and AI works on your real records, not pasted fragments.

Genuinely few small organizations are here. The ones that are did not get here with fancier tools; they got here with organized data and standardized workflows, which is unglamorous and decisive.

What to do at this level: automate whole workflows with a human checkpoint, measure results per workflow, and upgrade your data practices as you go. This is also the level where hiring outside help starts paying for itself.

The move up: custom capability. When an off-the-shelf tool no longer fits the job and the volume justifies it, you are asking build-versus-buy questions. That is level 5 territory.

Level 5: Builder

The tell: someone technical on the team connects AI directly to your systems through APIs, builds tools with coding assistants like Claude Code, and evaluates quality before trusting anything at scale.

Be skeptical of anyone who claims this level is common. Even among large enterprises, only a small fraction of AI deployments are genuinely autonomous systems. For a small business, level 5 is a choice about a specific high-volume problem, not a destination everyone should chase.

What to do at this level: scope before you build. The most expensive mistake at level 5 is building something a $50-a-month tool already does.

Three honest notes about the ladder

Higher is not automatically better. The right level is the one your actual work justifies. A five-person firm at a well-run level 3 is often getting more value per person than an enterprise at level 4.

Tool names do not set your level. Using Claude Code does not make you a level 5 business, and using plain ChatGPT does not cap you at level 1. What sets your level is how deeply the work is woven in: reuse, shared practice, connected data, standardized workflows.

Skipping levels usually fails. The pattern behind most stalled AI efforts is a level-1 organization buying a level-4 tool. The tool was fine. The prompts, data, and habits it needed did not exist yet.

Find your level

You can probably place yourself within one level just from the tells above. If you want it done properly, our assessment figures out your level in about ten minutes of conversation, scores your readiness across four dimensions, and gives you a roadmap matched to where you actually are, including the specific moves that take you up one rung. Not five.