Is My Business Ready for AI? An AI Readiness Assessment for Small Business

Where AI helps in an owner-led company

The useful uses of AI in a small or mid-sized business are less glamorous than the demos. They cluster around reading and writing — work where a person currently has to read something messy before anything can move. A 2026 Census Bureau working paper, built on the Bureau’s Business Trends and Outlook Survey, found that “writing, document analysis, and information search are the leading Generative AI use in tasks,” and that 66% of firms using AI relied on it solely to augment tasks. That matches the jobs below.

• Drafting. First drafts of follow-up emails, quote cover letters, job summaries and review requests, written from the record and sent by a person who reads them first.
• Triage. Reading each new inquiry or email, sorting it by type and urgency, and putting it in the right person’s queue with the key details pulled out.
• Call answering and after-hours inquiries. Answering when nobody can, taking the caller’s details and reason for calling, booking into open slots where the rules are clear, and handing everything else to a named person first thing the next morning.
• Document intake. Reading intake forms, purchase orders, supplier invoices or certificates and filling the fields in your system, flagged for a person to confirm.
• Summarizing. Turning a call recording, a long email thread or a month of job notes into a short summary someone can act on.

Notice what each of these has in common: AI moves information from an unstructured form into a structured one, or drafts something a person approves. The decision — the price, the promise, the yes or no — stays with someone whose name is on it.

Where AI does not help

AI does not fix a hand-off that has no owner. If new inquiries sit unanswered because nobody is responsible for them, an AI that answers instantly just produces faster replies that still lead to nobody. Name the owner first; then decide whether AI helps that person.

AI does not help much where the answer has to be exactly right and a wrong one is hard to spot. The National Institute of Standards and Technology lists “confabulation” among the main risks of generative AI: “the production of confidently stated but erroneous or false content.” A confident wrong price, a wrong policy term or a wrong date in a customer message is worse than a slow right one. Pricing, contract terms, refunds, compliance answers and anything legal, medical or financial stay with a qualified person, with AI at most preparing the file.

And AI does not replace the first real conversation with a new customer. It can make sure the conversation happens quickly; it should not be the conversation. Low-volume work, work where every case is different, and processes where the records are not kept anywhere also fall outside what AI can do well today.

Why AI pilots fail: what the evidence says

There is no credible, independent study of AI failure rates specific to small US businesses, so be wary of anyone quoting one. The evidence that does exist comes from larger organizations, and it points in one direction. In KPMG’s Global AI Pulse survey of 2,145 senior leaders at companies with $100 million or more in revenue, fielded in April and May 2026, returns remained concentrated among “only 7 percent of leaders who report establishing ROI.” Leaders with strong visibility into what AI costs were five times as likely to report established ROI: 15 percent against 3 percent.

Why it goes wrong is better documented than how often. RAND interviewed 65 experienced data scientists and engineers in 2024 and identified five root causes. The first is that stakeholders “misunderstand — or miscommunicate — what problem needs to be solved using AI.” The others: the organization lacks the data the project needs; it focuses “more on using the latest and greatest technology than on solving real problems for its intended users”; it lacks the infrastructure to manage its data and deploy the result; and the problem is too hard for AI to solve. RAND’s recommendation is short enough to put on the wall: focus on the problem, not the technology.

Those are companies with IT departments and budgets. A smaller business that buys a tool on a hunch has less room for the same mistakes, and the mistakes are the same ones: no defined problem, no baseline, no owner, and nobody tracking what the tool costs against what it returns.

How to implement AI in your business: a 30-60-90 day rollout

If one process passed all five checks, here is a rollout sized for an owner-led company. It is a plan, not a promise — each phase ends in a decision, and stopping is one of the allowed decisions.

• Days 1–30: define and measure. Write the process down as it runs today. Measure it for two to four weeks: volume, time spent, wait time, errors. Write the review rules in three levels — what runs on its own, what runs with a person checking, what waits for approval. Agree on the number that would count as success, in hours or dollars, before you see any tool.
• Days 31–60: pilot in shadow mode. Run the AI alongside the existing process on that one step. It drafts, sorts or reads; a person checks every output and sends it. Log every correction. Track the true cost: subscription, setup time and the hours spent checking.
• Days 61–90: decide. Compare the pilot against the baseline. If the error rate is low and the time saved is real after checking time, widen the AI’s authority one level for the outputs it got right — and keep the rest under review. If the numbers did not move, stop, write down why, and keep the process improvements you made in the first month.

A rollout this narrow can look slow. It is faster than the alternative, which is buying a platform for the whole business, discovering six months later that nobody owns it, and starting over.

What to fix before buying any AI tool

The cheapest fixes in an AI project happen before the purchase. Before signing anything, fix these:

• The hand-off with no owner. Name the person responsible for the step, with a time by which they act.
• Two records for one thing. If the same customer lives in the CRM, a spreadsheet and the accounting file with different details, pick the one that is true and retire the others — an AI reading three versions has no way to know which one is true.
• The missing written rule. If staff have to ask the owner before every exception, write the rule first; an AI cannot follow a rule that only exists in your head.
• The missing baseline. Without a before number, nobody can tell whether the tool worked.
• The data terms. Find out where the vendor stores your customers’ information, whether it is used to train their models, and how you get it back if you leave.

If you are not sure which process to start with, the companion guide on what to automate first maps the chain from inquiry to payment and finds the step where work waits — the same starting point, whether the answer turns out to be AI, a plain rule, or a person with clearer instructions.

When not to spend money on AI — and the free first step

Do not buy an AI tool if the process failed two or more of the five checks; fix the process first and revisit in a quarter. Do not buy one because a competitor did, or because a vendor’s demo used a business that looks nothing like yours. And do not pay anyone, including us, for an “AI strategy” that never looks at your records.

If you want a first read for free, the Revenue Self-Audit scores the six hand-offs of your chain in about four minutes and shows where work is most likely waiting. If you would rather have the answer from your own records, The Revenue Review reads them with read-only access over two weeks and ends in a written report that says which steps are worth automating, whether AI is the right tool for any of them, and which ones need a person and a rule instead. Implementation is scoped and priced separately, and only if the numbers justify it. The 20-minute call is free, and it is a fair place to hear that the Review is not worth it for your situation yet.