TITAN-REC 3.0

Customer record

Customer 48213 · consumer

9 orders · €281.00 · €31.22 average basket · 47-day cadence · last order 67 days ago

Recommendations served on 18/06/2026 — unfold a line for its calculation

Selection and business rules

104 active references → 71 shortlisted candidates → 10 products served. 12 higher-scoring candidates were dropped after the model: 8 by the aisle quota (3 products maximum per aisle — without it, the selection would hold ten hygiene products, which make up 78% of their history), 4 by the price floor.

Demonstration: these settings are not saved and train no model. They do recompose the block preview at the bottom of this section, though — you read there the exact shape your setup would take.

The model maximises a single quantity, and everything else follows from it: the ranking, the bonuses, the trade-offs. The objective is chosen once, with your Titane team, because it changes what the model is trained on — not merely how its output is sorted. The three settings below apply to the ranking produced by that objective.

Push a reference, a whole aisle, your most profitable products, or your latest releases towards the top of the ranking. The model keeps control of the order: a bonus moves a product, it does not glue it to the top.

Remove a reference, a whole aisle, or your site’s oldest products from the block. An exclusion is firm: the product no longer comes out, whatever its score.

Keep one position in the block for a choice of your own. The model ranks the other 9 as usual: a reservation takes one slot away, it does not disorganise the rest.

A setting applies to all 45,800 customers; the preview takes one as an example — customer 48213, whose first five lines are the ones on their record above. The movement shown on the right compares the position to the model’s own ranking. The preview applies fixed bonuses — named product +6, aisle +3.5, attribute +2.5 — to the ranking’s starting weight: it is a mock-up of the setting, not the 789-signal re-ranker.