Recommendation systems for e-commerce
Short answer
SwankyForge builds product recommendation systems for online stores and marketplaces: similar items, frequently bought together, personalised home pages and ranked search results. We start from your order and click history, ship a simple baseline first, and prove the gain in an A/B test before adding anything heavier. A typical project takes 2–4 months and costs €14k–€42k.
Ranked for a returning visitor who viewed 3 kettles
Glass kettle 1.7 l
$490.94
Tea infuser set
$180.88
Descaling tablets
$90.81
Ceramic mug pair
$240.77
Stock and margin rules applied after ranking
Sounds familiar?
- Shoppers leave without finding what they came for.
- Everyone sees the same bestsellers.
- Cross-sell is a hand-picked list nobody maintains.
- Search returns relevant items that do not sell.
- Nobody can say whether the current widget earns its place on the page.
What we build
Similar products
Alternatives on the product page, by behaviour and by attributes.
Frequently bought together
Bundles and add-ons from real baskets, not a manual list.
Personalised pages
Home and category pages ordered for each shopper.
Search ranking
Results ordered by what this shopper is likely to buy.
E-mail and push
The same model behind your campaigns and reminders.
Business rules on top
Stock, margin and brand policies applied after the model, where you control them.
How a project runs
- 1
Agree the metric
One number the system must move, such as clicks on recommendations, add-to-cart or revenue per session, and how it will be measured.
- 2
Baseline from your data
A simple model from co-views and co-purchases. It often beats the current widget on its own and sets the bar for everything after.
- 3
Find candidates, then rank
A first stage picks a few hundred plausible products; a second orders them for this shopper. Stock, margin and policy rules apply last.
- 4
A/B test
Part of the traffic sees the new system, the rest the old one. The decision is made on the agreed metric, not on a demo.
- 5
Production and handover
An API your storefront calls, monitoring, a rollback switch, and the repository in your organisation.
What you get
- Recommendation API with a response-time budget agreed up front
- Offline evaluation report and A/B test design
- Monitoring dashboard on the agreed metric
- Retraining pipeline and runbook
- Code, models and documentation in your repository
Typical price
€14k–€42k
Typical timeline
2–4 months
From the published rate card. The final price is fixed once the scope is agreed.
Estimate your projectQuestions
How much data do we need for recommendations?
Enough interaction history to see repeated behaviour, usually a few months of orders and product views. With little history, recommendations start from product attributes and popularity, and personalisation is added as data accumulates.
Can it run next to our current recommendation widget?
Yes. The new system is tested against the existing one on split traffic, and the old one stays on until the agreed metric says otherwise.
What about new products and first-time visitors?
New items are placed by their attributes (category, price, brand, text and images) until enough interactions arrive. First-time visitors get popularity and context, such as the page they came from, until they have history.
Do we need a data team on our side?
No. We need read access to orders, the product catalogue and site events, and one person who can approve changes to the storefront.
When is a recommendation system not worth building?
With a small catalogue or low traffic, a curated list and a popularity sort often perform as well and cost far less. We say so on the first call.
Tell us about the task.
The first call is free. If this system is not the right answer for your numbers, we say so.