Strategy
AI for e-commerce businesses: where to start
AI for e-commerce businesses: how to choose the first process and what to prepare before you buy another tool. A ShopGuru guide. Book a 30-minute call.
Many online stores start with the wrong question: which AI tool should we buy? With artificial intelligence for e-commerce businesses, the first decision is a different one: which of the store's processes moves to AI first, under which rules, and who approves. Here are three questions to choose that process, and what to prepare before talking to anyone.
Article summary
- An online store gets off to a good start with artificial intelligence when it chooses a specific process first and the tool only afterwards.
- In 2025, according to Eurostat, 11.54% of Portuguese enterprises with 10 or more people used AI, against 19.95% on average across the European Union.
- Among EU enterprises that considered using AI and didn't go ahead, the most common reason given was a lack of relevant expertise.
- The best first process repeats every week, has rules that can be written down, and lets a mistake be caught before it goes far.
- Answering "where's my order?" is, in many stores, a good first process; deciding prices or refunds outside the return window rarely is.
- Before talking to a supplier, the store should have a month of customer conversations, its written policies, its list of tools and who does what.
- On day one, the ShopGuru agent asks for approval for every action, and the store's team grants autonomy one action at a time.
What the numbers say about AI in Portuguese businesses
Eurostat's most recent data are for 2025. In Portugal, 11.54% of enterprises with 10 or more people used at least one AI technology. The European Union average was 19.95%, according to the article Use of artificial intelligence in enterprises.
Among EU enterprises that had considered using AI and didn't, 70.9% pointed to a lack of relevant expertise. Next came unclear legal consequences (52.5%) and concerns about data protection (48.8%).
Our reading: what holds businesses back isn't a shortage of tools, since there are dozens on sale. It's not knowing where to start, who sets things up and who answers when something goes wrong. The legal questions have answers of their own, which we cover in AI Act and GDPR for an AI agent.
A note for e-commerce in Portugal: many online stores have fewer than 10 people and fall outside this statistic. We don't know how many of them use AI, and we won't make that number up.
"There are dozens of AI tools on sale. What's missing is knowing which process to hand over first, and to whom." ShopGuru
Why starting with a tool tends to go wrong
A new tool assumes the team knows what to ask it and has time to do so. In a store with 3 people, time is what's missing, and the tool risks going unused.
The AI that comes with the e-commerce platform has a different limit: in most cases, it doesn't know the store's rules, and the exceptions are where the team loses hours. The comparison is in platform AI or a custom agent.
That's why we argue for the reverse order: the process first, then that process's rules, and the technology last.
How to choose your business's first AI process
Three questions are enough to rank a store's processes. They take a sheet of paper and no supplier in the room.
Does it repeat every week?
A process that happens twice a year doesn't repay the work of teaching it to anyone, person or agent. The first candidate is what the team repeats every week.
Can the rule be written down?
If the team can explain the rule to a new hire on one page, the agent can follow it too. "Exchanges up to 30 days, refunds up to 14" can be written down. "It depends on the customer" can be written down as well, as long as it says what it depends on.
Can a mistake be fixed in time?
An answer about delivery times can be reviewed. A refund already issued or a price already published is harder to undo. The first process should be one where a mistake is caught before it leaves the store, or where the action asks for approval.
In the example from the Diagnostic page, a fictional shoe store, the three questions sort the processes like this:
| Process (example) | Every week | Written rule | Mistake can be fixed | First? |
|---|---|---|---|---|
| Answering "where's my order?" | Yes, 6 h | Yes | Yes | Yes, the agent does it alone |
| Product pages for the new collection | Yes, 5 h | Partly | Yes, with approval | Next, as a proposal |
| Refunds outside the return window | Yes, 2 h | No, they're exceptions | Hardly | Asks for approval |
| Orders to suppliers | Yes, 3 h | No | Hardly | Stays with the team |
In many stores, order status is the most repeated after-sales question. We gave it an article of its own: "Where is my order?".
"The first process for AI is the one the team repeats every week and can explain on one page." ShopGuru
What to prepare before talking to an AI supplier
With this ready, the first call gets to a decision sooner:
- A month of customer conversations, from every channel: WhatsApp, email, Instagram and the website form.
- The written policies: exchanges, returns, shipping, delivery times and warranties.
- The exceptions that aren't written down, such as shipping to the islands or a late exchange accepted for a long-standing customer.
- The store's list of tools: platform, invoicing, carriers, email marketing and spreadsheets.
- Who does each process today and how many hours a week it takes, even if it's an estimate.
The last point is the one most often missing: without knowing who does what, nobody can say where AI saves hours and where it only adds review work.
What stays with the store's team
The team still decides what the agent does. For each type of action, the store owner chooses one of three levels: does it alone, asks for approval, or never does it. On day one, every action asks for approval, and autonomy is granted one action at a time, as the team comes to trust the answers. Above a limit in euros or as a percentage, an action the agent would take alone asks for approval instead.
When the agent doesn't know, it says "I don't know. I'll ask." and hands the conversation to the team with a summary. How this differs from a chatbot is covered in chatbot or AI agent.
From the first process to the plan: how the Diagnostic works
The Diagnostic is this work, done with us. It starts with a 45-minute kick-off meeting, followed by a video session with each person who does a process.
The written plan arrives 10 working days after the kick-off meeting and is presented at a final meeting. It includes the process map, what moves to AI by priority, the integrations one by one, and an Implementation proposal, as a full project or a pilot. If the store doesn't go ahead, the plan is its to keep and use on its own.
"A written plan says more than a demo: what changes in the store, who approves and when." ShopGuru
If it goes ahead, the Implementation takes 3 to 4 weeks, with the first process up and running by the end of the first week. These timescales, like the plan's 10 working days, are indicative. After that, ShopGuru's engineers keep looking after the agent every month: they watch it, test its answers and propose fixes.
Frequently asked questions
Does a small store need artificial intelligence?
It needs a process that repeats and takes up hours. In a 2-person store, if one person spends 6 hours a week answering questions about orders, there's somewhere to start. If there's no process like that, AI can wait.
Do I have to change e-commerce platform?
No. ShopGuru connects to the store you already have, whether it's on Shopify, WooCommerce, PrestaShop or another platform. The Diagnostic plan sets out which tools we connect, which we don't, and which are still to be assessed.
How much does it cost to start with AI in an online store?
It depends on the process and the tools to connect, so we don't publish a fixed figure for the Implementation. The Diagnostic plan ends with a proposal covering the weeks, the dates and the price, so the store can decide with the numbers in front of it.
Can I start with operations instead of customer service?
Yes. On the Operations front, the agent prepares work for the team to approve, such as product spec sheets, suggested prices and images. If the store's most repeated process sits there, that's where to start.
Where to start
The first step is to take the list of the store's processes and run each one through the three questions. With that list done, a 30-minute call with a ShopGuru engineer helps you find out whether the Diagnostic makes sense for your store. Book a call.
About this article
Written by ShopGuru's engineers based on the design of the service. Published and reviewed on 28 October 2026. The shoe store and the numbers in the example are fictional. Source of the adoption data: Eurostat, Use of artificial intelligence in enterprises and dataset isoc_eb_ai (2025 data), enterprises with 10 or more persons employed. The Diagnostic and Implementation timelines are indicative.