Recommendation systems on gambling platforms work on the same principles as those on any content service, and the objective they optimise is worth knowing. Play history, session length, stake patterns and responses to previous promotions all feed the model. The output is a selection presented as suggestion.
The optimisation target is engagement rather than player outcome. Recommendations reflect what increases play, which is not the same as what suits the player. This is a straightforward commercial arrangement and continue reading worth stating plainly.
Placement of recommendations affects behaviour measurably. The ordering is a commercial decision. Commercial arrangements with studios sometimes influence placement alongside behavioural data.
The consequence for game selection is direct. Return figures are not an input to the model. Checking the published return of a suggested game frequently produces a surprise.
Promotional targeting works similarly. Segmentation explains why two players describe the same site differently. Accounts showing reduced activity frequently receive the most aggressive offers.
Data protection rules provide some control over this in several jurisdictions. Rights to object to automated processing and to profiling exist under some frameworks. Exercising them reduces targeted promotion without affecting the service.
The practical response is to use a different selection method. Sorting on published figures replaces an engagement-optimised order with a value-optimised one. Browsing the full catalogue with filters, rather than the curated front page, takes slightly longer and produces materially different choices.
One practical consequence is worth drawing out. The full catalogue is available and ordered differently. Using filters changes the selection process entirely. Sorting by provider or filtering by return percentage produces a list assembled on different criteria. The difference in the resulting choices is usually substantial and continue reading takes no longer to arrive at.
(Image: YouDATA)