“What should we play?” is rarely one question.
First the group has to find a night. Then everyone has to remember what they own, what supports the right player count and whether somebody is willing to buy something new. A random picker solves almost none of that.
What Should We Play? grew from treating the whole evening as one product.
One game night, two jobs
The app separates the workflow into Plan and Pick. Both sit inside a shared Game Night, so a group has one link and one place to return to.
Planning remains usable without an account. Guests can mark each day as available, maybe or unavailable using a mobile flow or a desktop heatmap. The ranking considers duration, time zones, weekend hours and the minimum player count. A host can lock the result and download a calendar invite.
Picking becomes more useful with persistent accounts because recommendations depend on personal libraries. Steam and Google sign-in support cross-device profiles, while Steam and IGDB feed one internal game model.
A recommendation should explain itself
The app scores games out of 100 using ownership, player count, genre, playtime, freshness, price, interest and several multiplayer signals. Modes such as Co-op Night, Backlog, Cheap, Familiar and Fresh change the emphasis.
I chose to show the factor breakdown instead of hiding it behind one score. The group can see why a game ranked highly and where the weak fit came from. Alignment sits separately from the average because one person strongly disliking a choice still matters when everyone else is enthusiastic.
This makes the recommendation a starting point for a decision, not a verdict from an invisible algorithm.
Permanent preferences and tonight’s mood are different
Ownership, ratings and favourites belong to the person. “Not tonight” belongs to the session.
Keeping those concepts separate avoids polluting a long-term profile with a temporary preference. The same applies to friend groups, saved libraries and a particular Game Night’s shortlist. Reusing stable information should remove work without making every session identical.
The edges became the product
Private Steam profiles need graceful fallback. Deal APIs can be slow, so prices are cached. Shared links cannot be allowed to lock a time or remove a game, so host actions require a signed session cookie. Discord reminders have to deduplicate even when a free hosting plan only runs scheduled jobs once per day.
None of these are the headline feature, but they determine whether the app remains helpful outside a demo.
The project began as a picker. It became a lesson in looking at the joins between tasks. The useful product was not the cleverest recommendation. It was reducing the amount of coordination needed before friends could actually play together.