We took XLA Metasites to the buyers who already had one

Corporate employers, onboarding and training across sites

Here the buyer is an employer with sites in several countries, and a training cost that scales with every one of them.

We built the experience around replicas of the company's own locations. A new hire walks the office, runs process simulations, takes safety drills, follows the production line and works through quality control, with AI characters guiding and taking part. Quizzes and content earn rewards, and a manager can see the progress.

The saving is the business case. Weigh what it costs to put a new hire in front of a trainer against a platform cost that does not rise with headcount or geography.

Residential developers, from the sale through to living there

This buyer is a developer building master-planned communities. It already owns the audience twice over: the buyers in its pipeline, and the residents of everything it has finished.

We defined three journeys, because the developer serves three populations. Residents get a digital twin of the property they bought, a verified identity and neighbours. Businesses rent virtual premises and staff them with employees or AI characters. Institutions such as banks, clinics and schools replicate a branch or run a purely digital one.

One mechanism sits between purchase and handover. Somebody buys an apartment that will be finished in two years. In the meantime they walk through it, configure the layout and try interiors. Their choices pass straight to suppliers, who can meet them inside the same space to discuss it. Two years of silence become two years of contact, and the fit-out spend lands with the developer instead of leaking away.

After handover the environment carries property management: service requests, maintenance, documents, bills and community meetings. The developer's revenue does not stop at the sale, which is the point of building it this way.

Football clubs, for the supporters who never get to the ground

A club's audience is enormous and mostly distant. Most supporters will never attend a match, and the commercial programmes have to reach them anyway.

We designed the club's history as something you walk through, with the decisive matches and the player profiles built as exhibits. Quizzes and leaderboards sit over the top, and an AI guide gives tours. The commercial idea is where the merchandise sits. Put the shirt next to the moment that made it worth owning, with the full store a room away. Add a fan zone, sponsor spaces, and concerts on the days there is no match.

Football alone ran past a hundred pages of game design documentation, and it is the concept Xsolla's team took furthest. Watch the demo.

Universities, where the clinical case is the product

Here the buyer is a university teaching something practical. Dentistry was the developed example.

Clinical teaching is limited by scarce things: staff time, clinic hours, patients. A student cannot repeat a difficult case on demand.

So we made the case itself the unit of learning, and had the system generate each one. A student examines the patient, questions them, selects a treatment from real instruments and medications, and gets feedback on the reasoning. An adaptivity layer tracks where each student is weakest and generates cases aimed at those gaps.

One design decision carries the whole thing. The AI patient does not always tell the truth. Real patients forget, minimise and misdescribe, so a student who takes the answers at face value reaches the wrong diagnosis, which is the lesson.

The AI characters had to do work in the room

Every one of the four products depends on AI characters, and they are the reason the environments behave like services rather than showrooms. Xsolla had integrated GPT into the platform, so a character could hold a conversation instead of playing an animation.

What we defined was what each character is for. A guide and simulation participant in onboarding. A sales adviser and concierge in real estate. A museum host in football. In dentistry, the patient, the one place where the character is the obstacle. We have since taken an assistant of this kind into a working product, on a loyalty platform.

We also wrote the constraints. A general model placed inside a branded environment will answer questions the brand never asked it to answer, and it will do so on somebody else's servers. So we set the rules first, and they are product rules before they are engineering ones. Classify the content, define which role each answer serves, prepare the questions that role actually asks, then decide what the model may reach and what it may say.