An AI readiness checklist for a small company

A vendor has shown you a demo, your nephew says everyone is using AI, and you are not sure whether your company is ready or just curious. Ready, for a first small use, means five questions can be answered out loud, in a sentence each. No consultant is needed. If you stumble on one, that is where to spend your next few days.

Question one: where does your customer information live?

You pass if you can point to one place, a CRM or a single sheet, that holds every enquiry with its source, date, product and outcome. Partly in phones is fine, as long as you know which phones.

If not: spend a day moving the last three months of enquiries into one sheet. This is the first fix because every use of AI reads from your records. A tool pointed at three phones and a notebook sees none of it.

You also need to name the fields you do not trust. Everyone has some, and knowing them lets you keep the first task away from them.

Question two: who agreed to hear from you, and how?

You pass if you know which customers gave their number for orders only, which for offers, and which for neither. Details of health, money or identity papers are kept apart from marketing lists.

If not: stop here before anything else. This is the area where a mistake reaches customers. A broadcast list built from old visiting cards is not a list of people who agreed. Rules differ by channel and change, so ask your own adviser what applies, and do not copy wording from another firm.

Question three: who owns the tool?

You pass if one named person holds each AI account, knows who pays, who can see its history, and what happens to your content if you cancel. One named person approves anything that goes to a customer or the public.

If not: assign the names this week. The common fault is a salesperson who opened a free account on a personal email and now holds your chat history. Move it to a company address and write down who may use it.

Question four: what may staff paste in?

You pass if every person who touches the tool has been told, in writing, what may go in and what may not, and knows whom to ask. Phone numbers, tax and bank details, and a price list marked for one customer stay out. Your own brochure text, a product description, a message you have already written, may go in. Our post on an AI use policy for a sales team gives a full page you can copy.

If not: this is the easiest fix, and it gets skipped because nobody has told the team anything. Sit down with the team and one sheet of paper.

Question five: what is the one task?

You pass if you can name a single job that repeats often and that you can judge by eye in a minute. A mistake in it would be embarrassing but not harmful, and a person reads each output before it goes out. Drafting a first reply or tidying meeting notes fits.

If not: pick the smallest one. "Use AI for marketing" is not a task. "Draft the first reply to portal enquiries, which I read before sending" is.

In what order do I fix the failures?

Take them in this order: permission first, then ownership, then records, then staff rules, then the task. Permission comes first because its mistakes reach customers. Ownership is next because it costs nothing. Records take longest, so start them while you do the rest. A "partly" on records is normal and need not hold you back, provided the first task does not depend on the weak fields.

What does a first pass look like?

Imagine, as an invented example, a dhurrie and rug maker in Warangal with a showroom, a dozen weavers and online orders. The owner wanted AI to answer repeated questions from dealers: sizes, colours, sample requests and delivery.

Going through the five, the answers came out uneven. Records: quotations were in a sheet, but dealer outcomes lived in two salespeople's phones. Permission: the dealer broadcast list had been built from visiting cards collected at exhibitions. Ownership: a salesperson had opened a free account with a personal email. Staff: nobody had been told anything. Task: dealer replies, the most tempting job, which would draw on the weakest records and the least certain permission.

The mistake would have been to begin there. Instead the owner moved the account to a company email, wrote a half-page on what could be pasted, and chose a smaller job. The tool drafted first replies to website enquiries from retail buyers, using no dealer data at all, and the owner read each before it went. The team learnt how to spot a wrong draft, though a few drafts used a colour name the showroom no longer stocked, and the owner kept a note of each. The dealer list was cleaned in the meantime and got its own question later.

When is AI not the next step?

If enquiries are not recorded at all, if replies take days because nobody owns them, or if buyers cannot tell what you offer. An unanswered enquiry is a process fault, and a tool will not close that gap.

Avoid tasks where a wrong answer costs money and nobody will check it: prices in a quotation, anything touching compliance, claims about performance.

Write the trial on one card

Put four things on a card: the task, its owner, what a good result looks like, and the date you will look at it. Keep a log of corrections; if you correct most outputs, the task is the wrong one.

Write that card today, with one named owner and a date to read the log. When the first task holds, AI for sales and marketing shows what to try next, and AI readiness and governance is the service for this stage. For a second opinion before you begin, tell us what you need.