FenixL

Guide

AI for small businesses: what actually works, what it costs and where to start

Updated November 3, 2026 · 6 sections

The short answer

For a small business, AI pays off when it removes a task you repeat every day — answering the same questions, writing the same replies, copying data between systems. It does not pay off as a replacement for judgement, for sales relationships or for having something worth selling. Below: the six uses we see work, what each one costs, and how to start in 30 days without betting the business on it.

01 What AI actually does in a small business

In a small business AI does not replace anyone: it eats the repetitive tasks that are costing you hours today. That is all it does well, and it turns out to be a lot.

It is worth separating three things that get sold together but are not the same:

  • Automating: moving information from one place to another without anyone copying and pasting. An order comes in and shows up by itself in your spreadsheet, your invoicing and your WhatsApp. It is the most boring one and the one that gives back the most time.
  • Answering: an assistant that handles the questions you get asked twenty times a day — opening hours, prices, availability, order status — using your real business information, not invented answers.
  • Producing: drafting text, product descriptions, review replies, spec sheets. Useful, but always with someone reviewing before anything is published.

The practical difference: automating and answering run on their own once they are built. Producing always needs a person checking. If someone sells you the third as if it were the first, that is a problem.

02 Six uses that pay for themselves

These are the six we see working again and again in businesses under twenty people, ordered by how fast the money comes back.

1. WhatsApp and website support, around the clock. The assistant answers what you already answer: prices, hours, directions, whether something is in stock, where an order is. What it does not know, it hands to a person instead of making it up. This is the one you notice first, because most of the enquiries that arrive at night or at the weekend are lost today.

2. Capturing and following up leads. Every enquiry is logged with what they asked, when and where it came from, and gets an automatic follow-up two days later if nobody replied. Most small businesses do not lose customers on price: they lose them by never writing back.

3. Quotes and standard replies. The ten questions that arrive by email every week have the same ten answers. A draft prepared with the customer's details and your real terms, ready for you to check and send in a minute instead of twenty.

4. Orders and invoicing without copying and pasting. The order comes in once and appears by itself in the store, the tracking sheet, the invoice and the warehouse notice. AI barely features here: this is plain automation, and it is where the most hours come back.

5. Product descriptions and spec sheets. With a catalogue of hundreds of references, writing each one by hand is not viable, and leaving them empty costs you Google rankings. A first draft per product, reviewed by someone who knows the product, clears the whole catalogue.

6. Reviews and reputation. An alert when a review lands, a draft reply depending on whether it is good or bad, and a monthly summary of what people keep saying. Repeated complaints are business information, not noise.

What they have in common: every one of them replaces a specific task you do by hand today and could time with a stopwatch. If you cannot say which task it removes and how many hours a month, it is not an AI project, it is a fashion purchase.

03 Three uses where AI does not pay off (yet)

Saying this costs us work, and we say it anyway: there are three places where putting AI into a small business is expensive and disappointing.

  • Closing sales for you. AI can bring you the enquiry and prepare the quote. Closing a sale of any real size is still a conversation between people, and a buyer can tell within two sentences that they are talking to a bot trying to sell them something.
  • Making decisions with consequences. What to charge, which supplier to switch to, who to hire, which complaint to accept. It can give you the analysis; the decision and the responsibility are yours.
  • Publishing content nobody reads first. Search engines have spent years getting better at spotting mass-produced content with no judgement behind it, and one wrong detail about your own product does more damage than having no description at all.

And a fourth, the most expensive of all: automating a process that is broken. If the way you handle orders does not work, automating it only makes it fail faster and in more places at once. Fix the process first, then automate it.

04 What it really costs

An AI assistant built on your real business information sits in our Scale plan, 500 USD (typical agency price for the same thing: around 5,000). Simple order or lead automations fit in Business, 250 USD. And if what you need first is a page that actually captures enquiries so there is something to automate, that is Launch, 100 USD.

What almost nobody mentions until the second invoice:

  • Monthly tool costs. A conversational assistant consumes service from an AI provider and usually a connection tool as well. Budget 20 to 60 USD a month for a small business, depending on conversation volume. It is not optional and it does not go away.
  • Preparing your information. The slow part is not building the AI: it is gathering your prices, your terms, your lead times and your exceptions somewhere they can be read. If that lives in two people's heads, there is work to do before starting.
  • Maintenance. You change a price, add a service, close for holidays. If nobody updates the assistant, within three months it is giving your customers false information with your name on it.

Our practical advice: start with one single thing, the one you repeat most often each day. A 250 USD project that gives you back four hours a week is a better deal than a 3,000 one that tries to automate everything and ends up unused. Full numbers for each plan are on pricing, and the hidden costs are in the costs FAQ.

05 A 30-day plan to start

This is how we do it when a small client wants to start without taking a risk. Four weeks, one thing per week.

Week 1 — Measure one normal week. Write down every task you repeat more than three times and how long it takes. Change nothing yet. By the end of the week you will have a list of five or six things, and one of them will stand out on its own.

Week 2 — Pick one and write the rules. The one at the top of the list. Write down in a document how it is done today: what gets asked, what gets answered, what happens when there is an exception. This document is what makes the AI work or not; it is 80% of the real work.

Week 3 — Build the smallest thing that works. It handles that one task and nothing else. It escalates to a person the moment it leaves the script. No integrations yet, no five channels at once.

Week 4 — Use it yourself before anyone else does. Run it for a week with real customers but watch every reply. You will find five to ten cases you had not anticipated; that is exactly what this week is for.

From there, either it visibly gives you hours back and you move to the next task on the list, or it does not and you switch it off having spent one month and a small budget instead of a quarter and a large one. Both outcomes are fine; the bad outcome is never finding out.

06 How to tell a serious AI proposal from hot air

Ask for these four things in writing before signing anything. If one is missing, it is not a project, it is a demo.

  • Which specific task it removes and how many hours a month. With a name and a number. "Improving efficiency" is not an answer.
  • Where it gets its information from. It has to be your real data: your price list, your catalogue, your terms. If the assistant is not tied to a source of yours, it will invent answers, and it will deliver them with complete confidence.
  • What happens when it does not know. The correct answer is "it hands over to a person". Any other answer means it will improvise in front of one of your customers.
  • Total monthly cost, tools included. Development is one payment; running it is every month. Both should appear in the quote.

Two warning signs we see often: nobody asking about your processes before proposing a solution, and a price with no relationship to the work it saves. If you want to check all of this further, the AI and automation FAQ has eighteen more questions with straight answers.

Common questions

Which AI should a small business implement first?

The one that handles the question you repeat most often each day, which is almost always WhatsApp or website support. It is the use you notice first because it recovers the enquiries that are lost at night and at weekends today, and it requires no change to how you work. What you should not do is start with several things at once: if it fails, you will not know which part failed.

Link to this answer

How long does it take to be up and running?

Two to four weeks for a first use case in a small business, and most of that time is not programming: it is gathering and writing down your information — prices, terms, lead times, exceptions. If that is already documented, it gets much shorter. If it lives in two people's heads, that is the task that sets the timeline.

Link to this answer

Do I need someone technical in the company?

Not to use it, yes to keep it current. Nobody needs to write code, but someone does need to be responsible for flagging when a price, an opening time or a condition changes. Without that, the assistant keeps answering with information from three months ago, and the mistake costs your reputation, not ours. It is a few minutes a month, not a job.

Link to this answer

Is it safe to give an AI my customers' data?

It depends entirely on how it is built, and it is the right question to ask. The rule we follow: the assistant accesses only the information it needs to answer, never your full customer database, and personal data is not used to train any external model. Ask for this in writing from whoever builds it, with the provider named and where the data is stored. If they cannot put it in writing, that is your answer.

Link to this answer

Is AI any use if I do not have a website yet?

It is useful on WhatsApp, which for many small businesses is where the customer actually is, but the sensible order is the other way round. Without a page that captures enquiries and that Google can show, you are automating a flow of customers that does not exist. First something that brings enquiries in, then something that handles them on its own.

Link to this answer

Want to know what this would cost in your case?

Tell us which task eats most of your day and we will tell you whether it is worth automating — in writing, free, and we will say no if the answer is no.