AI in a small business with an online store: where to start, minus the hype

Every week a new AI tool comes out and every vendor promises that theirs changes everything. Meanwhile, the same thing happens in most small businesses with an online store: someone on the team uses a chat tool on their own to write emails, someone else tried it for descriptions and gave up, and nobody really knows what data is being pasted into those windows.

This article is not about specific tools, which change every month. It is about how to decide where to use AI in a small company with a store, how to measure whether it works and what rules to set so as not to create a new problem.

The process first, the tool second

The most common mistake is to start with the tool (“we have to use AI”) and then look for a use for it. It works the other way round: look for tasks that meet three conditions.

• They are repeated a lot. Ten minutes done forty times a week is worth more than a two-hour task once a month.
• They follow a pattern. Answering similar questions, reading documents with the same structure, writing texts in a set format.
• They can be checked quickly. If checking the result takes as long as doing it, you gain nothing.

Make a list with your team of what takes up most of their time. Almost every company comes up with the same things: emails answered the same way every time, data copied from one place to another, documents someone reads just to pull out three figures.

Where it usually delivers in a store

Customer service

Draft replies to the usual questions (where is my order, how do I return it, which size should I choose), with the real order and product data. A person reviews and sends. The move to replying without review comes later, and only for questions where the draft never needs correcting any more.

Catalogue

Descriptions, translations, attributes extracted from the supplier's product sheet, image alt text. AI saves a lot here, but quality needs watching: generic text across thousands of product pages helps neither sales nor rankings. Better fewer well-reviewed pages than the whole catalogue filled with padding.

Administration

Reading supplier invoices and delivery notes, suggesting reconciliations, sorting incoming emails. These are structured documents, and the saving shows every quarter.

Operations and management

Sales summaries, stock alerts, quick answers to questions about your own data (“which products have dropped most this month?”). For this, the data has to be organised and accessible, which is usually the real work.

Three examples of a first use case

What they have in common: a person still has the final say, the result is checked in seconds and the saving shows every week. None of them requires changing software or redesigning entire processes.

Where not yet

• Decisions with legal or tax consequences without review: prices, terms, taxes.
• Automated support with no way through to a person. An assistant that cannot say “let me pass you to someone” frustrates more than it helps.
• Processes that are not even defined. AI does not tidy up chaos: it speeds it up.
• Figures nobody checks. Language models write well and calculate with errors. Numbers should come from data and rules, not from generated text.

Data and usage rules

Before rolling AI out across the company, decide in writing:

1. Which tools are used and with which accounts: company accounts, not personal ones.
2. What data can be entered and what cannot (customer data, contracts, financial information), according to each service's terms.
3. Who reviews what goes out: an email to a customer, a published text, a document.
4. Where what you learn is kept: the instructions that work are company knowledge and should not leave with the person who wrote them.

How to measure the first case

Choose one and measure it before and after: time per task, number of tasks per week, errors caught in review. If after a few weeks the team uses it without anyone reminding them and the time has gone down, move on to the next one. If not, adjust it or drop it without regret. A pilot that never gets beyond the pilot stage is not a failure of AI; it is a sign that the case was not the right one.

Count the cost too: pay-per-use AI tools look cheap per query and get expensive with volume. Note how much each case costs per month and compare it with the hours it saves.

And do not forget the technical foundation: an outdated store, with abandoned modules and no accessible API, greatly limits what any integration can do with its data.

How to handle it in PrestaShop

IA en tu empresa (AI in your company) is Rekire's service for bringing artificial intelligence into a company's real processes: consultancy, training and development for a monthly fee with no minimum term. It starts with a process review and a prioritised roadmap, and continues with automations, assistants that work with your company's data and integration with your ERP, your store and your tools, with clear rules on usage and privacy. The Inicia plan focuses on one area of the company, with the first automations and training for that team; the Acelera plan takes AI to several areas at once, with custom developments, integration with your systems and ongoing consultancy.