From behavioural events to ranked products.
Theklen turns customer interactions and catalogue data into a ranked list of products for a specific visitor, context and placement.
Signals the engine uses.
The model learns from commerce data already generated by the shop. Inputs can be adapted to the data available in each implementation.
Customer events
Product impressions, views, clicks, cart additions and purchases create the behavioural signal.
Request context
Customer or session identifier, current product, page placement, channel and requested result count.
Product data
Product identifiers, categories, attributes, price and availability connect behaviour to the sellable catalogue.
Business conditions
Stock, eligibility and placement-specific exclusions can limit what is allowed to appear.
A continuous ranking pipeline.
The neural model scores eligible products from the current context. Customer response becomes new feedback for later rankings.
Collect
Receive behavioural events and catalogue changes.
Represent
Turn customer, context and product signals into model inputs.
Score
Estimate which eligible products best fit the current commercial context.
Rank
Order products for the requested placement and return the top results.
Learn
Feed subsequent customer responses back into future ranking.
A response the storefront can render.
The engine does not generate product prose or invent catalogue items. It returns an ordered selection from eligible products.
Ranked product IDs
An ordered list, limited to the number requested by the placement.
Request context
The response stays connected to the relevant visitor, placement and source context.
Storefront-ready result
The shop retrieves its own product content, price and availability for display.
Illustrative request
{
"customer_id": "c_1842",
"placement": "product_page",
"context_product_id": "sku_482",
"limit": 6
}
Illustrative response
{
"recommendations": [
{"product_id": "sku_917", "rank": 1},
{"product_id": "sku_233", "rank": 2}
]
}
Illustrative integration pattern, not a published API specification.
Similarity is an input, not the objective.
The recommendation problem is commercial. The ranking can be aligned with the outcome the retailer values, while customer behaviour supplies the feedback signal.
Purchase likelihood
Help shoppers reach a product worth buying.
Average order value
Support relevant cross-sell and upsell decisions.
Revenue per visitor
Evaluate the value created across the shopping session.
Events in. Ranked IDs out.
Storefront or app
Customer interactions and recommendation request.
Theklen engine
Signal processing, neural scoring and ranking.
Recommendation placement
Product IDs rendered using current catalogue content.
What this system is not.
Not a rules engine
Manual merchandising rules do not create the underlying customer model.
Not similarity-only
Visual or textual resemblance does not equal the best commercial next product.
Not an LLM chatbot
The core output is a ranked product set, not generated conversation.
Largest tax free retailer in Northern Europe.
The recommendation technology has been used in real-world retail operations at significant catalogue and customer scale.
Evaluate the fit with your commerce data.
We can map the available events, catalogue fields, placements and commercial objective before discussing implementation.
rando.parna@theklen.ai