A business is ready for its first AI agent when five things are true: the task is done the same way every time and written down, the answers it needs live in a document rather than in people’s heads, the tools it touches can connect to other software, there is enough volume or value to pay for it, and a named person will review it. The 20-point checklist below scores those five areas out of 100. For the same result in three minutes, with what to fix first, take our free AI readiness score.
What does an AI readiness checklist measure?
An AI agent is software that follows your rules, reads your information and acts in your tools: it answers a call, books an appointment, updates a record or drafts a reply. Whether it works depends far less on the AI model than on the business around it. This checklist looks at five areas, each with a fixed weight out of 100:
| Area | Weight | What it asks |
|---|---|---|
| Clear processes | 25 points | Is the task done the same way every time, with known exceptions? |
| Knowledge and records | 20 points | Are the answers and records written down where an agent can read them? |
| Connected tools | 20 points | Can the agent act in your calendar, CRM, inbox or store? |
| Volume and value | 20 points | Is there enough volume, or enough value per customer, for an agent to pay off? |
| Ownership and comfort | 15 points | Does someone own the agent, and are the limits on what it may do alone clear? |
Clear processes carry the most weight because an agent cannot follow rules nobody has written. These are the same five areas and weights our AI readiness score uses, so the checklist and the tool agree.
The 20-point AI readiness checklist
Each item is worth 5 points. Tick it only if it is true today, not if it is planned. Answer for the one task you most want to hand to AI, not for the whole business.
Clear processes (25 points)
- The task is done the same way by everyone. Two people handling the same enquiry, booking or order follow the same steps.
- The steps are written down. A checklist, a script or a short document exists, even if it is rough.
- You can write the rules for the hard cases. When to book, when to refund, when to say no, when to pass it to a person.
- The exceptions are known, and so is who approves them. “Discounts over a set amount go to the manager” is a rule an agent can follow; “use your judgment” is not.
- You measure the task. Response time, missed calls, hours per week or errors: some number tells you how it goes today.
Knowledge and records (20 points)
- The answers to customer questions are in one place. Prices, opening hours, services, policies, delivery and returns, in a help center or a single document.
- That information is current. The prices and policies on your website match what your team actually says on the phone.
- Customer, order or job records live in one system. A CRM, booking system, practice system or store admin that everyone uses, rather than paper and personal notes.
- The records are reasonably clean. No large backlog of duplicates, missing phone numbers or jobs nobody closed.
Connected tools (20 points)
- You can list the tools the task touches. Calendar, CRM, inbox, phone line, store, spreadsheet or practice system.
- Each one has an API or a ready-made integration. Check for an API, or a Zapier or Make connector, on the vendor’s integrations page.
- You control the accounts. You, not a former employee or an agency that has since moved on, hold the admin logins.
- You already automate something between them. Even one working Zapier rule or webhook shows the tools can talk.
Volume and value (20 points)
- The task happens often. Dozens of calls, messages, leads or documents a week, not a handful a month.
- Some of it is lost today. Calls go to voicemail, leads wait hours for a reply, or staff retype the same data.
- A customer is worth enough to matter. Losing one new customer to a missed call costs you real money over their first year.
- You can name the payoff. Hours saved, enquiries answered after hours, faster replies or fewer errors: one number you will check after a month.
Ownership and comfort (15 points)
- One person is named as the owner. They will review the agent’s conversations or runs for about an hour a week in the first month.
- You have decided what the agent may do alone. Routine answers and bookings, yes; refunds, discounts and anything sensitive go to a person.
- You are comfortable with AI in this role. If it talks to customers, you are fine with that for routine questions, as long as there is a hand-off to a person.
Add up your points. To be more precise, or to get the agents that fit the problems you pick, answer the 13 questions of the AI readiness score: it scores each area from your answers instead of from yes or no.
What does your score mean?
The checklist and the tool use the same three bands:
- 0 to 39: groundwork first. AI can help, but a little preparation will make the first agent work far better. Start with one process, not a platform. Write the task as a checklist and write the top customer questions down; both take days, not months.
- 40 to 69: ready for a first agent. You have what a first agent needs. Put it on the task where you lose the most, and measure it for a month.
- 70 to 100: ready to scale. Your processes, records and tools are in good shape. You can run several agents together, or connect AI to your own systems.
Look at your two weakest areas, not just the total. A score of 60 with no owner is riskier than the same 60 where most of the gaps are in connected tools, because a missing integration can often be bridged with email or a shared sheet, while an agent nobody reviews drifts.
How do you pick the first use case?
A good first use case passes four tests:
- High volume. It happens many times a week, so the agent has enough work to pay for itself and you get enough examples to tune it quickly.
- Clear rules. You could write the steps and exceptions on one page. If every case is a judgment call, start somewhere else.
- Low risk. If the agent gets one wrong, a person can fix it the same day. Keep money, medical and legal decisions behind approval at first.
- Measurable. You know today’s number (missed calls, reply time, hours of data entry) and can compare it after a month.
Common first agents that pass all four: answering calls after hours or when the front desk is busy with a Receptionist Agent; answering the same customer questions every day with a Support Agent; replying to new leads and booking them with a Booking Agent; and copying data between email, sheets and the CRM with an Ops Agent, which holds anything irreversible for a person’s approval. If your team keeps asking where things are, a Knowledge Agent answers from your own documents with citations.
Starting narrow is also what most businesses that use AI actually do. A U.S. Census Bureau working paper based on the 2026 AI supplement to its Business Trends and Outlook Survey found that 57% of firms using AI integrate it in three or fewer business functions (U.S. Census Bureau).
What should you prepare before your first AI agent?
Four documents and one decision cover most of the setup work:
- An FAQ document. The 30 questions customers ask most, with the answers you want given. Include prices, hours, services, what you do not do, and where to send people for everything else. This becomes what the agent answers from.
- Booking rules. Which services can be booked, how long each takes, who can do them, buffers between appointments, how far ahead people can book, and what needs a deposit or a call back.
- Approval rules. A short list of actions that always need a person: refunds, discounts, cancellations inside a set window, complaints, anything about health or money. Write who approves each one and how fast.
- A hand-off path. How the agent passes a conversation to a person (a transfer, a text to the on-call phone, a ticket) and what it tells the customer while they wait.
- A named owner. One person who reads the agent’s conversations each week, fixes wrong answers in the FAQ document and decides when it may do more.
If you are short on any of these, that is normal. For our Receptionist Agent, for example, setup includes writing up your services, prices and booking rules with you; bring your price list and calendar access.
What governance does a small business need for AI?
You do not need a compliance department, but you do need a few written rules before an agent talks to customers or touches records:
- Approval gates. List the actions the agent may take alone and the ones it can only propose. Start strict and loosen the rules once the owner trusts its work.
- Sensitive data. Decide whether the task involves health, financial or identity data. If it does, keep the agent’s access to the minimum the task needs, check where the vendor stores and processes data and whether it is used for training, and check which privacy rules apply to you before launch. Staff should not paste that data into personal AI accounts either.
- Human hand-off. Customers must be able to reach a person, and the agent should hand over on its own when it is unsure, when the customer is upset, or when the topic is on your sensitive list.
- A log and a weekly review. Keep a record of what the agent said and did, and have the owner read a sample each week.
- A register. A one-page list of every AI tool in use, who owns it, what data it sees and what it may do.
If you want a framework behind these rules, the U.S. National Institute of Standards and Technology publishes the AI Risk Management Framework, released on January 26, 2023 and intended for voluntary use, along with a Generative AI Profile (NIST). Its core is organized into four functions, Govern, Map, Measure and Manage, and the Govern function calls for documented roles and responsibilities and for policies that define roles for human oversight of AI systems (NIST AI Resource Center). For a small business, the practical version is the list above: a named owner, written limits and a record you review.
What are the most common AI readiness mistakes?
- Starting with a platform instead of a process. Buying an AI tool and then looking for a job for it. Pick the task first, then the tool.
- Automating a broken process. If the team does the task five different ways, an agent will copy the confusion faster. Write one way down first.
- Picking a judgment-heavy task first. Complex complaints, pricing exceptions and clinical questions are poor first jobs. Save them for later, with approval gates.
- Stale facts. An agent that answers from an out-of-date website will quote old prices with complete confidence. Check the facts before launch.
- No baseline. Without today’s number for missed calls or reply time, nobody can tell after a month whether the agent helped.
- No owner. An agent nobody reviews keeps repeating its mistakes. One hour a week in the first month makes the biggest difference.
- Letting it act on money too early. Refunds, discounts and payments should wait for a person until the agent has a record you trust.
- Waiting for perfect data. You need clean information for the first task, not a company-wide data project before you start.
Is your business behind if you are not using AI yet?
No. The U.S. Census Bureau’s Business Trends and Outlook Survey put the national rate of businesses using AI at 19.8% as of May 3, 2026, with 37% of firms with at least 250 employees using it; between December 2025 and May 2026, AI use grew among firms with at least 20 employees but did not change significantly among smaller firms (U.S. Census Bureau). Most small businesses are still at the start, which makes a careful first agent a real advantage rather than a catch-up exercise.
What should you do next?
- Take the free AI readiness score: 13 questions, a score out of 100, your two weakest areas and the agents that fit the problems you pick.
- Read what agents cost in our AI agent cost guide, and compare every packaged agent and bundle on our pricing page.
- Deciding between an off-the-shelf tool, a packaged agent and a custom build? Read build vs buy AI agents, or price a custom build’s drivers with the project estimator.
- Want someone to do the assessment with you? Our AI Opportunity Audit costs $490 and takes one week. It ranks your use cases by value, effort and risk, checks data readiness, gives build-vs-buy advice and governance steps, and the fee is credited in full if you go ahead with a build within 60 days. Or book a free 30-minute call and get a fixed quote within 24 hours.