A small business is ready for an AI project when it passes six checks: one repeated task, 100 to 200 real examples of it, data you can export, a person who knows the right answer, a number to measure success, and permission to use the data. Fail two or more and fix those first. This AI readiness checklist for small business owners costs nothing to run an takes one afternoon.

It is written for clinic owners, CA and law firm partners, school administrators and shop owners who keep hearing "use AI" but have no tech team. By the end you will know whether to start, wait, or skip AI.

Key takeaways

  • Pick one repeated task, not "AI for the business".

  • 100 to 200 correct examples are a realistic starting pile for document and message tasks.

  • Data stuck in paper, photos and WhatsApp chats is the most common blocker.

  • If the data names a person, the DPDP Act, 2023 applies to how you use it.

  • Some tasks are better left to a spreadsheet or a simple rule.

What is an AI readiness checklist, and who needs one?

An AI readiness checklist is a short list of things to check inside your own business before you pay anyone to build AI. It saves you from buying a pilot that fails because the examples were never ready.

Two terms first. A "model" is software that learns a pattern from examples instead of following rules you type in. "Training data" is the pile of examples it learns from: old invoices, past enquiries, patient forms, case files.

Anyone who has a repeated task and a few months of records can use this list. You do not need to understand how a model works.

Which task should you hand to AI first?

Start with a task that happens often, looks similar each time, and has a clear right answer. A vague goal such as "make the business smarter" gives a developer nothing to build or measure.

Task

Good first AI

project?

Why

Reading supplier invoices and

typing them into Tally or Excel

Yes

Same fields every time, and the

right answer is easy to check

Sorting incoming enquiries into

sales, support and spam

Yes

Plenty of past examples, low cost

of a mistake

Answering the same 30 patient or

parent questions on WhatsApp

Yes

Questions repeat, and answers

can be approved in advance

Predicting which students will

drop out

Later

Needs several years of clean

records with the outcome noted

Deciding who gets a loan or a job

Not first

High stakes, legal risk, and a

human must stay in charge

"Using AI to grow sales"

No

No task, so nothing to measure

What is in an AI readiness checklist for small business owners?

Six checks decide most outcomes. Each one is something you can verify yourself in an afternoon, with a pen and your own files.

Check

Ready looks like

Not ready looks like

Quick fix

1. One task

You can describe it in

one sentence

"We want AI for

everything"

Pick the task that eats

the most staff hours

2. Examples

100 to 200 past cases,

with the correct

outcome noted

A handful of cases, or

outcomes never

recorded

Start recording the

outcome from today

3. Format

Excel, CSV, PDF, or an

export from your

software

Paper files, photos,

WhatsApp chats

Scan paper in batches;

export chats

4. Right

answer

One named person

who can say "this is

correct"

Three staff who

answer differently

Write down the rule and

settle disagreements

5. Measure

A number such as

"minutes per invoice"

or "enquiries answered

per day"

"It should feel better"

Time the task manually

for one week

6.

Permission

You know whether the

data names people,

and you can use it

Customer data copied

between personal

phones

List where the data sits

and who can see it

How much data does a small business need for AI?

For document and message tasks, 100 to 200 correct examples is a practical place to start. A pilot on your own data then shows whether that was enough, and what accuracy you get.

Prediction tasks need more history. To predict late fee payment or customer churn, you need months or years of records where the outcome was written down, not only the starting facts.

Quality beats quantity. Two hundred clean, correctly labelled examples beat two thousand where half the outcomes are blank or wrong. Nobody can promise an accuracy figure before testing, so be wary of any vendor who does.

Is your data stuck in places AI cannot read?

Most small businesses fail check 3, not check 2. The records exist, but they sit in WhatsApp chats, phone galleries, paper registers and one Excel file per staff member.

Fixing this is dull work, and it is usually the longest part of any AI project. A few habits help:

Keep one shared folder per task, with a fixed file naming style.

Export WhatsApp group chats monthly instead of scrolling for them later.

Scan paper at 200 DPI or better, one file per document, not one file per bundle.

Ask your accounting or billing software vendor for a full data export, and confirm you are allowed to have it.

That last point matters. If a vendor holds your data and charges for access, you have found a blocker before spending anything on AI.

Do you need consent before using customer data for AI?

If the data names a living person, such as a patient, student, client or borrower, the Digital Personal Data Protection (DPDP) Act, 2023 treats it as personal data. Using it for a new purpose generally needs a lawful basis, and often a clear notice to the person.

In practice, ask two questions. Does this task really need names and phone numbers, or can they be removed before the data goes anywhere? And does your privacy notice

already say how you use the data?

Our post on DPDP compliance for website owners in India covers the notice side. This is general information, not legal advice, so check your own case with a lawyer.

When should you skip AI altogether?

Skip AI when a simple rule or a spreadsheet formula does the job. AI adds cost and a margin of error, and it earns its place only when rules are too messy to write down.

Three signs you should wait:

1. The task happens fewer than a few dozen times a month, so manual work is cheaper.

2. Nobody can say what a correct result looks like.

3. A mistake would hurt a person, and no human reviews the output.

An honest vendor will tell you this up front. If one never says "you do not need AI for this", treat that as a warning sign.

How do you score the checklist?

Count the checks you pass out of six. Then decide.

  • I can describe the task in one sentence

  • I have 100 to 200 past cases with the correct outcome noted

  • The data can be exported or scanned into files

  • One named person decides what "correct" means

  • I have a number I can measure before and after

  • I know whether the data names people, and I have permission to use it

Six ticks: you are ready to talk to a developer. Four or five: spend two to four weeks closing the gaps, then start. Three or fewer: do not buy anything yet. Fix the data habits first, because they pay off even if you never use AI.

What to do next

If your task involves papers such as invoices or KYC forms, read AI document processing in India for invoices, KYC and GST. To see what a pilot and a full build tend to cost, read AI development cost in India (2026).

At Ureka, every AI project starts with a fixed-price pilot of about 4 weeks on the client's own data, and the first thing we do is run this checklist together. When the answer is "do not use AI for this" , we say so. You can read how that works on our AI pilot page.

Have a question about this? Ask us on WhatsApp, no sales pitch: https://wa.me/919625030157? text=Hi%2C%20I%20read%20your%20post%20on%20the%20AI%20readiness%20ch ecklist.%20My%20question%3A%20

Written by the Ureka Technologies team

FAQ

How do I know if my business is ready for AI?

Check six things: one repeated task, 100 to 200 past examples with the correct outcome, data you can export, one person who decides what is correct, a number to measure, and permission to use the data. If you pass at least five, you are ready to talk to a developer. Fewer than four means fix the data first.

How much data do I need to start an AI project?

For document and message tasks, 100 to 200 correct examples is a practical start. Prediction tasks, such as late payments or dropouts, need months or years of records with the outcome noted. A small pilot on your own data shows whether the amount is enough, so you do not need to guess in advance.

Can I use AI if my records are on paper?

Yes, but scanning comes first. Scan each document at 200 DPI or better as its own file, and then software can read the text and fields. Messy handwriting and stamps lower accuracy, which is why a pilot should measure it on your own papers before you commit to a full build.

Do I need a data scientist to prepare my data?

No. Preparing data is mostly business work: collecting files in one folder, noting the correct outcome for each case, and agreeing on the rules. A developer handles the technical cleaning. What a developer cannot supply is someone from your team who knows what a correct result looks like.

Is AI worth it for a business with fewer than ten staff?

Sometimes. It pays off when one task eats many staff hours every week and follows a pattern, such as typing invoices or answering the same WhatsApp questions. If the task happens only a few dozen times a month, a spreadsheet or a simple rule is cheaper and safer.