AI for e-commerce
We run an online store ourselves. These are the automations that survived contact with real customers, real stock and real returns.
- Platforms
- Woo, Shopify, Presta +3
- Usual first win
- Catalog content
- Pilot
- 2-4 weeks
- We run a store
- Every day
What AI for e-commerce actually means
AI for e-commerce means automating the parts of running a store that involve a lot of repetitive work with text and data: answering product questions, writing catalog content, spotting stock and pricing problems, and preventing returns. It works best on processes you already do by hand and understand well. Applied to a chaotic process, it only speeds up the chaos.
We sell auto parts online as our day job, which is a category with punishing compatibility rules and an unforgiving return rate. Everything below is something we either run ourselves or have deployed for a client, together with the risk that vendors tend to leave out of the pitch.
The seven processes worth automating
| Process | What you gain | The risk nobody mentions |
|---|---|---|
| Storefront assistant | Instant answers evenings and weekends, fewer lost buyers | Without hard limits it invents discounts and compatibility |
| Product descriptions | A complete catalog instead of empty pages, better search visibility | Plastic identical text and factual errors from messy source data |
| Stock and reorder signals | Fewer sold-out bestsellers, less cash frozen in dead stock | Forecasts trail sudden demand jumps, so they inform rather than decide |
| Price monitoring | Pricing decisions from data instead of instinct | Automatic repricing can zero your margin or start a price war |
| Order anomaly detection | Fewer fraud losses and misdirected shipments | Rules set too tight block honest buyers |
| Fitment checks | Fewer returns, lower logistics cost, more trust | A wrong "yes, it fits" is worse than no answer at all |
| Multilingual catalog | Export markets without a per-update translation budget | Terminology: a clip and a clamp are different products to a buyer |
The long version of this list goes through each one with the implementation detail.
The returns problem, which is where the money usually is
A large share of returns happen because the buyer ordered the wrong item, not because anything was faulty. In categories with compatibility rules, meaning parts, components, consumables and equipment, that share is high enough to dominate the logistics bill.
The fix is a fitment check before purchase rather than a returns policy after it. The buyer states their situation, such as the car model, the dimensions or the equipment they already own, and the system either confirms the fit or warns them. Two rules make it work: the fitment data has to be actively maintained, and where the data is missing the system must say "we do not know" instead of guessing. One compatibility question before the purchase is cheaper than one return, every single time.
Platform notes
- WooCommerce. The most flexible target. REST API plus direct database access, so almost anything is reachable. Watch for plugin conflicts on stores that have accumulated forty of them.
- Shopify. Clean APIs and a strong app model. Constraints are Shopify's, not ours: rate limits and checkout customisation rules shape what is possible.
- PrestaShop. Module-based integration. Common in the EU and well suited to multilingual catalog work.
- BigCommerce. Solid API coverage, straightforward for catalog and order automation.
- Adobe Commerce (Magento). The heaviest of the group and the most capable. Integration takes more hours, and enterprise catalogs are where generation pays back most.
- Custom stacks. More integration work, identical principles. If it can expose orders and products over HTTP, it can be automated.
The question is almost never "is it possible on our platform". It is "how many hours does the integration take on ours", and that is answerable in a call.
When not to bother
Under a few hundred orders a month, most of these automations will not pay back the integration hours. Fix the catalog data first: AI generating descriptions from wrong attributes produces wrong descriptions faster. And never automate a process nobody currently does properly by hand, because there is no correct output to compare against.
How to start
- Count the hours. Where does the week actually go? Usually customer questions or product content.
- Automate one process. One only. Measure over four to six weeks against the baseline you took before starting.
- Then move to the next. Stores that take on all seven at once finish none of them.
If you are not sure which of the seven applies to your store, the AI readiness audit follows a faster route for e-commerce, because the process inventory is largely known in advance.
Questions we get asked every time.
What can AI actually do for an online store?
Seven things reliably: answer product questions in the storefront, generate product descriptions from attributes, signal stock and reorder points, monitor competitor prices, flag suspicious orders, check fitment before purchase to prevent returns, and maintain the catalog in several languages. Everything else on the market is currently a variation of those.
Which e-commerce platforms do you work with?
WooCommerce, Shopify, PrestaShop, BigCommerce, Adobe Commerce and custom stacks. The platform is rarely the obstacle: all of them expose orders, products and customer enquiries through an API, which is exactly what these automations need. The real variable is how many integration hours the specific store needs.
Will AI-generated product descriptions hurt our SEO?
They will if you generate them from thin air. Generate from your own structured attributes such as dimensions, material and compatibility, keep each description factually specific to the product, and review a sample by hand. A complete catalog with accurate generated text outranks a catalog of empty product pages every time.
Can AI reduce our return rate?
Yes, when returns are caused by the buyer ordering the wrong item, which is a large share in categories with compatibility. A fitment check before purchase asks the buyer for their situation and either confirms or warns. Selling car parts ourselves, we learned that one compatibility question before purchase is far cheaper than one return.
Should we let AI change our prices automatically?
No, at least not at the start. Blind automatic repricing drives margin to zero or starts a price war with an identical robot on the other side. Automate the monitoring and the recommendations, and leave the final change to a person inside clear limits until the data says otherwise.
Keep reading.
Where is your store leaking the most hours?
Unanswered product questions, half the catalog with no description, bestsellers running out, preventable returns. Tell us which one hurts most.
Book a free 30-minute call