# The AI-Native Readiness Assessment for Small Businesses

*Published 2026-07-22 · Dmitry Shirokov · shirokoff.ca/blog/ai-native-readiness*

The question I get from small-business owners has changed. Two years ago it was "should we do AI?" Now it's "are we *behind* on AI?" — usually asked in the slightly panicked tone of someone whose competitor just put "AI-powered" on their homepage. It's the wrong question, and the fear behind it leads to the wrong spending: a chatbot nobody asked for, a copilot licence per seat that gets used twice, a "let's pilot something" with no way to tell whether it worked.

Readiness for AI has almost nothing to do with ambition, budget, or how many times your competitor said "AI." It comes down to six unglamorous things, and you can score yourself on them in an afternoon. This is that assessment — aimed at the business with ten to a few hundred people and no data-science team.

## What "AI-native" actually means for a small business

**AI-native means AI sits inside the workflow that makes or saves you money — not bolted onto the side of it.**

Three postures, and only one is what people mean by AI-native:

| Posture | What it looks like | Where the value is |
|---|---|---|
| AI-curious | Staff use ChatGPT on their own to draft emails | Real but invisible, unmanaged, ungoverned |
| AI-enabled | You bought a tool with an "AI" feature and turned it on | Whatever the vendor built — you rent someone else's use case |
| AI-native | A core task is done by AI with a person supervising, measured, improving | Yours, compounding, defensible |

AI-native means: you picked a task that happens often and costs real money, put a model in the middle of it, kept a human on the outcomes that matter, and can point at a number that moved. It's a higher bar than a subscription.

## The six dimensions that predict readiness

None of them is about the model — you can rent a frontier model by the token.

1. **A use case that pays.** A task that happens *often*, costs *real money/time* today, and whose success is *measurable*. If you can't name the task and value an hour saved on it, you're not ready.
2. **Data you can reach.** The bar is *reachable*, not *perfect*. The data behind the use case must be exportable/queryable without a six-month integration. Clean-enough and accessible beats pristine and locked.
3. **A process worth augmenting.** AI augments a process; it can't invent one. The signal is that the process is repeatable enough to describe on one page. If nobody can say how the work is done, automating it is premature.
4. **Someone who owns it.** One accountable, empowered, willing owner — usually a sharp operator, not a data scientist, often with a partner filling the engineering gap. "IT will handle it" (when IT is one person) is the anti-signal.
5. **Governance you can live with.** A few decisions made on purpose: where the data goes, and who catches the model when it's confidently wrong. Keep a human in the loop on risky, customer-facing, or regulated decisions. The EU AI Act's risk tiers are a decent free framework to borrow.
6. **Economics that close.** Value of the task must clear the running cost *with* an iteration budget — the first version will be mediocre and you'll want three more passes. If it only pencils out at zero iteration and best-case pricing, it doesn't pencil out.

## Score yourself

Rate your chosen use case 0–3 on each dimension (0 = not true, 3 = clearly true with evidence). Total 0–18.

```yaml
use_case: "Answer repeat customer questions from our docs"
dimensions:
  use_case_that_pays:       0   # frequent? measurable? $ per hour saved?
  reachable_data:           0   # gettable without a big integration?
  process_worth_augmenting: 0   # steps fit on one page?
  an_owner:                 0   # one accountable, empowered, willing person?
  livable_governance:       0   # data-handling + human-in-the-loop on purpose?
  economics_that_close:     0   # value > running cost, with iteration budget?
# 0-6 Curious | 7-12 Pilot-ready | 13-18 Scale-ready
```

## Reading your score

| Score | Tier | What to do |
|---|---|---|
| 0–6 | Curious | Not yet. Don't buy a platform. Fix your lowest dimension (usually use case or data), re-score. Build intuition with off-the-shelf tools meanwhile. |
| 7–12 | Pilot-ready | Run *one* narrow, measured pilot on your strongest use case. Buy before build. Define the metric first. Human in the loop. |
| 13–18 | Scale-ready | Pilot with a pre-agreed plan to widen if the metric moves. The risk now is over-building — stay boring, instrument everything. |

## Where to start

**Start with the smallest supervised version of your highest-scoring use case — crawl before you walk.** The first project's job is to prove the loop (data in → model does the task → human checks risky outcomes → you measure) works at all, on one task, for real.

- **Crawl (weeks):** one task, one team, human checks every output, measure vs the old way.
- **Walk (months):** widen to the team, human checks the *risky* outputs only, track cost per outcome.
- **Run:** the task lives in the workflow; the human owns exceptions and the quality bar; it compounds to the next task.

On the build ladder, climb as little as possible: buy/prompt a hosted model first, add retrieval over your own documents when answers need your specifics, and go no further unless forced. Fine-tuning and training are almost never where an SMB should start.

**Traps, in order of frequency:** buying the platform before the use case; the AI-strategy deck with no reachable data behind it; automating a broken process (the AI reproduces the chaos faster); the pilot with no metric; and chasing data perfection so you never start.

## What to carry away

AI-native is a task in your core workflow done by a model with a person supervising, measured, and improving. Readiness comes down to six things — a use case that pays, data you can reach, a process worth augmenting, an owner, livable governance, and economics that close. Score your best use case honestly; the number tells you the next move. Buy the intelligence by the token and put your effort into the boring readiness that lets it land — one paying, measured, well-run task at a time.

*Related: [AI Strategy](https://shirokoff.ca/blog/ai-strategy) · [Data Strategy](https://shirokoff.ca/blog/data-strategy) · [Fine-tuning vs RAG vs Prompting](https://shirokoff.ca/blog/fine-tuning-vs-rag-vs-prompting) · [Feasibility Matrix](https://shirokoff.ca/assessment/feasibility-matrix)*