AI and software for online retailers
We build AI for online retailers, marketplace sellers and omnichannel chains: product search that understands how people shop, service agents that answer order questions from live data and catalogue work that keeps pace with your range. Merchandisers and service staff stay in control of what customers see.
Where the work gets stuck.
Searches that return nothing
Shoppers type what they need, like a gift for a keen cyclist or a rug that survives a dog, and keyword search shows an empty page. Those shoppers leave for a marketplace.
Queues full of the same questions
Where is my order, how do I return this, will it fit: the same questions reach Zendesk or Gorgias every day, and each needs a lookup in the shop, the warehouse and the carrier.
A catalogue that outgrows the team
Every supplier sends data in its own format, and titles, descriptions, translations and channel feeds are written by hand. New products wait for content before they can sell.
How we improve online retail.
Each one is scoped around your systems and rules, and each one keeps a person in charge of the decisions that matter.
Semantic product search
Embeddings over your catalogue let search understand intent, synonyms and descriptions, combined with keyword matching for brands, sizes and SKUs. Merchandisers keep their boost and pin rules for campaigns and margin.
Order and returns service agent
The agent reads a ticket, looks up the order in Shopify or Adobe Commerce and the shipment in Picqer or the carrier's system, then drafts a complete reply or starts a return. Refunds above a limit you set wait for a service employee.
Product content from supplier data
Supplier spreadsheets and spec sheets become titles, descriptions, attributes and translations in Akeneo or your shop, written to your style guide. A merchandiser reviews each batch and approves it for publication.
Marketplace and feed integrations
Stock, prices and orders sync between your shop, the Bol.com Retailer API, Amazon and Channable feeds, with alerts when a listing is rejected. Channel managers see every sync in one log.
Shopping assistant with product knowledge
A chat assistant answers product questions from your catalogue, manuals and reviews, and says so when it does not know. It is clearly labelled as AI and passes the conversation to your team on request.
Custom storefront features
Configurators, subscription flows, bundle builders or a Shopify app for a process your platform does not handle out of the box. We build them on your platform's APIs and hand over the code.
Built around the rules.
What we design for from the first week. Your legal and compliance people keep the final word.
European Accessibility Act
Since 28 June 2025, e-commerce services for consumers must meet accessibility requirements, with microenterprises exempt. Storefront features, search interfaces and chat assistants we build follow WCAG criteria and are tested with keyboard and screen reader.
Omnibus rules on prices and reviews
A price reduction must refer to the lowest price of the previous 30 days, and shops must explain whether and how reviews are verified. Pricing automation and review summaries we build follow those rules and keep the underlying data for checks.
EU AI Act transparency for chatbots
Customers must be told when they are interacting with an AI system. Our assistants say so at the start of the conversation and offer a route to a person.
GDPR and ePrivacy for personalisation
Personalised search and recommendations rely on behavioural data, which depends on valid consent for tracking. We design personalisation to work with consented data and to fall back to general results for everyone else.
Works with what you run.
If a system has an API, a database, an export or an inbox, we can build on it. These are the ones we meet most.
- Shopify
- Adobe Commerce (Magento)
- Akeneo PIM
- Channable
- Bol.com Retailer API
- Picqer
- Zendesk
- Gorgias
- Klaviyo
- Algolia
Where to start.
Semantic search for one category
We add semantic search next to your current search for one category or country and route part of the traffic to it. You compare zero-result searches and search-to-cart behaviour on your own shop before deciding on a wider rollout.
Talk it throughWhat it includes
- Embeddings over catalogue and product attributes
- Hybrid ranking with keyword matching for SKUs and brands
- Merchandising rules for boosts and pins
- A/B comparison with current search
Related work.
A guide to start with.
Further reading.
Embeddings and semantic search: a practical guide for product teams
How embeddings power semantic search, how to choose a model and a vector store, and how to measure whether your results actually improved.
5 min read
How to evaluate RAG: retrieval metrics, faithfulness and golden sets
How to measure a RAG system properly: separate retrieval from answers, check faithfulness claim by claim and build a golden set you can trust.
4 min read
Human-in-the-loop AI agents: design patterns for production
The patterns we use to run AI agents safely in production: approval checkpoints, confidence thresholds, tool permissions, audit trails and fallbacks.
4 min read
Common questions.
Algolia's AI search features cover many shops well. A custom build is worth it when your catalogue needs domain knowledge a general model lacks, when you want full control over ranking or when search has to combine data from several systems. We test both on your own queries before recommending one.
It can prepare them, within limits you set. A common setup lets the agent start returns and small refunds within policy, while anything above a threshold or outside policy goes to a service employee for approval.
Yes. We integrate with the Bol.com Retailer API, the Amazon Selling Partner API and feed tools like Channable, so orders, stock and listing errors land in the systems your team already uses.
Yes. Dutch, German, French and English work well when content starts from structured attributes and a style guide. We check translations for units, sizes and legally required wording, and a native speaker reviews samples per language.
We load test at peak volumes before your busiest weeks, cache what can be cached and set cost limits on model calls. Search falls back to keyword results if a model provider slows down.