Most businesses that have never automated anything are behind for a dull reason. Every week is already full, and setting up anything new means stopping to think about how the work is done today. The inbox gets answered when someone has a minute. Quotes go out when the person who writes them is back at a desk. Invoices are sent on Friday, or the Friday after.
So when someone says "you should be using AI", the honest reply is often "where would I even start?" An AI readiness assessment is a short way to answer that. It tells you whether your business can get value from AI this month, which job to start with, and what to fix first if the answer is "not yet".
Most of the readiness assessments you'll find online are written for large companies with a data team. They score your cloud setup, your data warehouse and your model governance. That is useful if you run IT at a bank. It is close to useless if your systems are an email inbox, a calendar, an accounting app and a spreadsheet. This post is for the second kind of business. We make Sanaf, an AI employee, so we have a stake in where this ends up. The assessment and the plan stand on their own, and the product shows up near the end.
The short answer
An AI readiness assessment checks whether a business has what it needs to put AI to work: a repeated task worth handing over, the information that task depends on, access to the apps where the work happens, a person who will check the results, and a clear rule for what the AI may do without asking. For a business with no automation yet, it can be ten questions and twenty minutes. If you can answer most of them with something specific, you are ready to try one job.
Why a business with no automation is often more ready than it thinks
There is a common belief that you need to "get your data in order" or "digitise first" before AI is worth trying. For the old kind of automation that was true. Rule-based tools needed clean, structured inputs and a person to map every step, and a business without that groundwork had nothing to build on.
Language models changed the starting point. A modern AI can read an email written in plain English, find the customer in your accounting app, and draft a reply that sounds like your team. It can work from the messy, half-written information most businesses actually run on. What it still needs is a clear job, access to the right apps, and someone who decides what "done" looks like.
That is why a business that has never automated anything can sometimes start faster than one with five years of brittle workflows. There is nothing to untangle. You pick one job, connect the apps it touches, and watch what happens for a few weeks.
The 10-question AI readiness assessment
Answer each question with a score from 0 to 2. Zero means "no" or "I don't know". One means "partly" or "sometimes". Two means "yes, and I could show you".
Section A: the work
1. Can you name one task your team does every week the same way?
Answering new enquiries, sending appointment reminders, chasing overdue invoices, writing up a weekly sales summary. If you can name it in one sentence and say who does it today, score 2.
2. Do you know roughly how long that task takes?
You don't need a time study. "About three hours a week, mostly on Mondays" is enough. If you have no idea, score 0 and spend a week noticing.
3. Is there a number that would tell you it got better?
Replies sent within an hour, invoices overdue more than 30 days, quotes sent the same day. A first try without a number tends to end in a vague feeling either way.
Section B: the information
4. Does the information the task needs exist somewhere you can reach?
Your price list, your opening hours, your customer records, your past emails. It can be a spreadsheet, a shared drive or a page on your website. If it lives only in one person's head, score 0.
5. Could a new hire do the task from what is written down?
This is the most revealing question on the list. If a capable temp would need you to explain the job every morning, an AI will too. Writing the explanation down once is most of the work of getting ready.
Section C: the apps
6. Does the task happen inside apps that connect to other tools?
Gmail, Outlook, Google Calendar, QuickBooks, Xero, HubSpot, Slack, Microsoft Teams and most other common business apps can be connected. Paper forms, a desktop program with no internet connection, or a supplier portal that only works through a browser login score lower.
7. Can you give access to one app without giving access to everything?
Most apps let you grant a connection to one account or one part of the account. If the only option is sharing your personal password, score 0 and fix that before anything else, whatever tool you pick.
Section D: the people
8. Is there one named person who will check the first few weeks of results?
One person, with their name written down. A first AI project with no owner drifts until people stop trusting it.
9. Will the team actually talk to it?
If your team lives in Slack or Microsoft Teams, an AI that works there gets used. If it sits in a separate dashboard, it gets forgotten. Score 2 if there is one place your team already reads every day.
Section E: the rules
10. Have you decided what the AI may do on its own, and what it must ask about first?
A sensible starting rule: it can read, draft and organise freely, and it asks before anything it can't undo, such as sending money, deleting records, or emailing a customer for the first time. If you have never thought about this, score 0. It takes ten minutes to decide.
How to read your score
Add up the ten scores. The maximum is 20.
| Score | What it means | What to do next |
|---|---|---|
| 15 to 20 | Ready. You have a job, the information and an owner. | Hand over one job this week and measure it for 30 days. |
| 9 to 14 | Nearly ready. One or two gaps, usually written instructions or access. | Fix the lowest-scoring section first, then start. |
| 0 to 8 | Not yet. The work itself isn't pinned down. | Spend two weeks writing down how one task is done and who owns it. |
A low score usually means the knowledge of how things are done lives in people's heads, which is normal for a business that has grown by doing the work. The fix is the same with or without AI: write it down.
If you want a longer version of this check, aimed at a team planning a bigger build, our AI readiness checklist goes through use cases, data, integrations, guardrails, ownership, budget and success criteria in more depth.
Which job to hand over first
How can AI help my business? That's the question most people start with, and it's too broad to answer. A more useful question is "which of my weekly jobs would I most like to stop doing, and which would hurt least if it went slightly wrong in week one?"
Good first jobs share four traits. They repeat at least weekly. They follow a pattern you can describe. A mistake is cheap and easy to catch. The result is easy to check.
Here are the first jobs that tend to pass that test:
| First job | Why it works as a first job | What to measure |
|---|---|---|
| Inbox triage | Every business has one, and sorting and drafting is low risk when a person approves the send. | Time to first reply |
| Lead follow-up | New enquiries go cold within hours, and follow-up is pure repetition. | Share of enquiries answered the same day |
| Appointment booking and reminders | The rules are clear: free slots, buffers, reminder times. | No-shows and back-and-forth emails |
| Overdue invoice reminders | The amounts and dates already sit in the accounting app. | Invoices more than 30 days overdue |
| Weekly numbers summary | Reading-only work, so nothing can be broken. | Whether anyone reads it on Monday |
Jobs that make a poor first try: anything where a mistake reaches a customer before a person sees it, anything that touches payroll or bank transfers, anything where the rules change every time, and anything nobody on the team can explain. Those can come later, once you trust what you have.
Each of these first jobs has its own page on the AI employee roster, with what the job covers and what stays with a person. The AI invoicing and AI scheduling assistant pages are good examples of how narrow a first job can be.
A 30-day AI implementation plan
An AI implementation plan for a first job fits in four weeks, with one owner and no project manager.
Week 1: write the job down
Write a one-page brief, the same one you would give a temp on their first morning. What the job is, which apps it touches, what a good result looks like, two or three examples of past work done well, and the rule for what needs your approval. Measure the number you picked in question 3, so you have a baseline.
Week 2: connect and watch
Connect the apps the job needs, and only those. Let the AI do the job with every output going to the owner for approval. Read everything. This week is for learning where it gets things wrong, and those mistakes tell you what the brief is missing.
Week 3: correct and loosen
Fix the brief where the AI got things wrong. Most errors in week 2 are missing instructions: a price that changed, a customer who needs special handling, a tone the team prefers. Once a type of output has been right for a week, let it through without approval. Keep approvals on anything that can't be undone.
Week 4: measure and decide
Compare your number against the week 1 baseline. Ask the owner three questions. Did it save time? Did it make mistakes we would not accept from a person? Do we want it to take a second job? If the honest answer to the first question is no, stop or change the job. Deciding in advance what failure looks like is what keeps a first try from dragging on.
What an AI adoption strategy looks like after the first job
Once one job works, the temptation is to hand over ten more. An AI adoption strategy for a business new to this is simpler than that. Add one job at a time, each with its own brief, owner and number. Keep the approval rule the same across all of them, so nobody has to remember which job is allowed to do what. Move to the next job when the current one has run for a month without surprises.
The jobs that come second are usually the neighbours of the first. A business that started with inbox triage often moves to lead follow-up next, because the enquiries are already flowing through the same inbox. One that started with invoice reminders often adds a weekly cash summary, because the AI is already reading the accounting app.
Five mistakes that sink a first try
Starting with the hardest job
The task that causes the most pain is often the one with the most exceptions. Start with a boring one and build trust.
Giving it access to everything
Connect what the job needs. An AI with access to your whole business on day one is harder to reason about and harder to check.
No written brief
"Handle the inbox" is not a brief. The AI will guess, and its guesses will be generic.
Nobody checking
Week 2 is the most important week, and the one most often skipped because things look fine. Read the outputs.
No exit rule
Decide before you start what result would make you stop. Without it, a project that isn't working keeps going because stopping feels like admitting a mistake.
Where to start with AI if you only have an hour
If you take one thing from this post, take this. Pick one job from the table above. Write the one-page brief. Note the number. That hour is most of the readiness work, and it is useful even if you never buy any AI at all, because a written brief also makes it easier to hand the job to a person.
Where Sanaf fits
Sanaf is an AI employee. You hire it into your team's Slack or Microsoft Teams, give it jobs from the roster, and it works inside the apps you already run. The integrations directory lists what it can connect to.
It is built around the same rules as the assessment above. You give it a job in plain words, the way you would brief a person. It works only in the apps you connect. It asks before anything it can't undo, and every job leaves a record the team can read. The security page explains how access and approvals work, and hiring an AI employee walks through the first day.
On price, you start with $120 of free credit and no card, and that free credit lasts 30 days, which covers the 30-day plan above. After that you buy credits from $20 when you need them, and credit you buy never expires. You pay for the work, not per seat. The pricing page has the details.
A good way to run the plan with Sanaf: score yourself on the ten questions, pick the first job, paste your one-page brief into Slack or Teams, and keep approvals on for the first two weeks.
Questions people ask about AI readiness
What is an AI readiness assessment?
An AI readiness assessment is a check of whether a business can put AI to work. It looks at whether there is a repeated task worth handing over, whether the information that task needs is written down and reachable, whether the apps involved can be connected, whether a person will own and check the results, and whether there are rules for what the AI may do without asking.
How do I know if my business is ready for AI?
Your business is ready for AI if you can name one weekly task done the same way each time, say roughly how long it takes, point to a number that would show improvement, give access to the apps it uses, and name one person who will check the results for the first month. You don't need a data team or clean data to start.
Where should a business start with AI?
Start with one repeated, low-risk job whose results are easy to check, such as inbox triage, lead follow-up, appointment reminders, overdue invoice reminders or a weekly numbers summary. Write a one-page brief for it, measure a baseline, and keep a person approving outputs for the first two weeks.
What should an AI implementation plan include?
A first AI implementation plan should include the job, the apps it may touch, a one-page brief with examples, the number you will measure, a named owner, a rule for what needs approval, a timeline of about 30 days, and a result that would make you stop. Week 1 is the brief and baseline, week 2 is supervised work, week 3 is corrections, week 4 is the decision.
Do I need clean data before using AI?
No. Modern AI can work with the emails, spreadsheets and documents most businesses already have. What it needs is a clear job, access to the right apps and written instructions for anything that lives only in someone's head. Cleaning data helps, but it is rarely the first step for a single job.
How long does it take to implement AI in a business?
For one job in a business new to AI, about 30 days is a realistic first cycle: a week to write the brief and measure a baseline, a week of supervised work, a week of corrections, and a week to measure and decide. Larger builds that change several systems take longer.
What are the biggest risks for a first AI project?
The biggest risks are starting with a job full of exceptions, giving the AI more access than the job needs, skipping the written brief, not checking outputs in the first weeks, and having no agreed result that would end the project. Each is avoidable with a narrow job, limited access, a named owner and approvals on anything that can't be undone.