What AI can actually do for a small business, without the hype

What is AI good at right now?

Strip away the marketing and today's AI is very good at language. Give it a messy WhatsApp message in Hinglish and it understands what the person wants. Give it a photo of a handwritten invoice and it reads the amounts, dates and names. Give it a long email thread and it tells you what was decided. Give it your price list and a question, and it answers in a polite, complete sentence. It does these things in seconds, at any hour, for a fraction of a rupee each.

That is enough to remove a surprising amount of daily work from a small business, because most of that work is reading, sorting, typing and replying.

What is AI bad at?

• Knowing when it is wrong. AI will write a confident answer whether or not it has the facts. If you ask it your delivery charge to Vashi and it has not been told, it may invent one.
• Arithmetic and totals. It is a language tool, not a calculator. Totals, taxes and stock counts must be computed by ordinary code and checked.
• Judgement calls. Whether to give a refund, whether a customer is lying, whether a deal is worth it. It can summarise the situation for you; it should not decide.
• Anything it was not given. It does not know your business unless you connect it to your business. "Add ChatGPT to the website" without that connection gives customers a chatbot that knows everything except your prices.

None of this makes AI useless. It means the design of the system matters more than the AI inside it.

Which five jobs are worth automating first?

1. Answering repeat questions on WhatsApp. Prices, availability, timings, delivery areas, directions. A bot that answers from your price list and hands over when unsure. This is the most common first project and usually the one with the clearest payback. How a WhatsApp chatbot takes orders walks through it.
2. Reading paperwork into a spreadsheet. Supplier invoices, purchase orders, forms, delivery challans: photographed or emailed, read by AI, checked, and entered as rows. No typing, fewer mistakes.
3. Drafting replies and quotes. Every enquiry gets a draft answer in your tone, with the right details pulled in, for you to read and send. You still press send.
4. Summarising long threads. A 40-message group chat or a month of emails about one project, turned into five lines of what was agreed and what is pending.
5. Sorting and routing. New leads, complaints and orders tagged by urgency and sent to the right person with the key details pulled out, so nothing sits unread.

Notice that none of these replace a person. They remove the part of the person's day that did not need a person.

How do you stop AI from saying wrong things?

This is the part most "AI solutions" skip, and it is the part that decides whether the thing is safe to put in front of customers.

• Answer only from your information. The AI is given your price list, policies and FAQ, and is instructed to answer from those only. If the answer is not there, it says so and hands over.
• Hard limits in code, not in the prompt. Discounts, refunds, delivery promises and prices are checked by ordinary code against your rules. The AI cannot approve a 50% discount because the system does not allow it to approve anything.
• Numbers come from the database. Totals, stock and dates are calculated, not generated. In a report system I built, every number the AI writes is checked against the real data before the report is sent.
• A human handover that actually works. One tap or one phrase gets a person, with a summary of the chat so far.
• Everything is logged. You can read every conversation and change the rules. If the bot gets something wrong once, it gets fixed once, for everyone.

This is what I mean by guard-rails on the AI and chatbot development page. It is not optional.

What does it cost?

Two parts. The AI itself is paid per use, to the provider, and for a small business the monthly amount is usually small: thousands of messages cost less than one staff lunch. WhatsApp adds its own per-message charges if the bot lives there. The real cost is the build: connecting the AI to your price list, your sheet, your email or your website, writing the rules, testing it with real messages and setting up the handover. That is a one-off project. I quote it fixed, within 24 hours, with an honest estimate of the monthly usage alongside. There is no monthly fee to me unless you want maintenance.

How should a small business start?

1. Pick one job from the list above, the one that eats the most time or causes the most mistakes.
2. Write down the rules a new employee would need for that job. That document becomes the AI's instructions.
3. Start with a version where AI drafts and a person approves. Watch it for a month.
4. Once it is right nearly every time, let it handle the easy cases alone and send the rest to you.
5. Only then look at the next job.

Businesses that start with "let's add AI everywhere" tend to end up with nothing working. Businesses that fix one annoying task tend to come back for the second.

What have I actually shipped?

A WhatsApp ordering bot used by over 250 active customers of a clothing wholesaler. An assistant that reads incoming customer emails, pulls out the details and replies on its own. A paid assessment platform where AI writes personalised reports, with every number checked against the real data so nothing is made up. Each one follows the rules above. If you want to know which of the five jobs makes sense for your business, describe your day in the form below and I will tell you plainly, including if the answer is "none yet". If you are also weighing up a website or an app, do I need an app or is a website enough? covers that, and the app development page shows what I build when the answer is an app.

Common questions

Will AI replace my staff?

In a small business, rarely. It removes the repetitive reading and typing from their day so they can handle customers, sales and problems. The businesses that benefit most keep the same people and get more done.

Is my data safe with AI?

It depends on how it is built. Your data should go only to the AI provider you have chosen, under their business terms, and never be used to train public models. I set this up explicitly and explain where every piece of data goes.

Can AI work with the tools I already use?

Usually yes: Gmail, WhatsApp, Google Sheets, Tally exports, your website or your CRM. Most projects plug into what you have rather than replacing it.

Do I need to be technical to use it?

No. You describe the job in plain words and write the rules as you would for a new employee. The technical side is my job, and you get a dashboard, not a command line.