Prosus Revolutionizes Ecommerce with AI Model Trained on 10 Trillion Data Tokens

Prosus  unveiled the next evolution in ecommerce, introducing Large Commerce Models (LCMs) – agentic artificial intelligence systems that are set to replace traditional search and recommendation engines.

The LCM has been trained on Prosus’s unique global database of commercial transactions, continually learning from over 500 Million users and over 10 Trillion tokens of data. This is a new operating system for ecommerce – a single integrated model which can power diverse tasks from improving basic search to personalised recommendations.

This LCM is already driving substantial impact both in immediate cost savings, in areas such as marketing spend, as well as spending on different models – it is 60 times cheaper to run than the best AI models with higher performance on ecommerce tasks. It is also transforming the consumer experience from a manual, list-based experience to an intelligent, hyper-personalised system which understands consumer intent, such as “I want something special for dinner, to celebrate with friends’.”

Unlike traditional recommendation systems, LCM has long-term memory, allowing it to learn from outcomes and refine strategies over time. This helps agents deliver hyperpersonal experiences across multiple interfaces – from voice assistants and AR glasses to social video and in-car commerce.

Breaking From the Old Playbook

For nearly 30 years, ecommerce has relied on keyword search and recommendation engines – the online equivalent of walking down aisles in a supermarket trying to find the right product. LCM is a step-change in the consumer experience, a hyper-personalised system with long-term memory that learns from outcomes. It can reason about consumer intent and improves every time you interact with it. It is equivalent to having a personal shopper who understands what you want, why you want it, and how to match you to the best product available on the market.

“LCM is to ecommerce what a personal shopper is to a department store – only it costs consumers nothing extra, stays with you over time, and improves every time you use it. This is a whole new operating system for ecommerce. This will be available to our global network of companies we have invested in – adding immediate value to their tech stack, way beyond what they could develop on their own.”

– Fabricio Bloisi, CEO, Prosus

“We have built an intelligent AI system which learns constantly by more than 200 billion data tokens every day. Traditional ecommerce has been good at cataloguing products and enabling transactions. We are now making it good at understanding consumer intent and reasoning.”

– Euro Beinat, Global Head of AI, Prosus

Built for Scale, Trained on Billions

Our Large Commerce Model (LCM) was first developed with iFood, Latin America’s largest food delivery platform, and is now trained together with the portfolio of Prosus Group companies, such as OLX, eMag and Despegar. That gives it an unmatched ability to learn, adapt, and improve, creating a flywheel of insights that powers all our businesses. At its core, LCM brings together five powerful capabilities:

• Ecommerce specialist – Unlike generic AI, it’s trained on billions of real shopping and transaction moments, giving it a deep understanding of how consumers buy and what drives their choices.
• One brain, multiple capabilities – A single system that handles everything from search to recommendations, replacing the patchwork of machine learning models companies used to rely on.
• Smarter understanding – It doesn’t just guess what people might do, it understands intent – why they’re shopping, what they’re comparing, and how those choices connect over time. For example, it can help us make better holiday ad recommendations based on previous clothing purchases and food preferences.
• Shared advantage – By learning from activity across all our companies, each business benefits from insights gained elsewhere – making local services stronger and more personalised.
• AI agents that act – Beyond recommendations, it can actually take action – like adjusting a marketing campaign on the spot.

Measurable Impact – at Scale

Already in production, LCM has achieved significant results.

• Push notifications – Click-through rates consistently beating all iFood benchmarks, with a 4x uplift in orders.
• Personalisation and search – conversion rate for “you may also like this” suggestions increased 66%, mapping personal preferences to order opportunities.
• This also applies to specific categories: overall search for pharma increased orders 4.6%.
• Scale – serving 20 million users and processing more than 200 billion tokens daily, placing it among the top 10 deployments1.
• Efficiency – operating at costs 60 times lower than proprietary models in a similar class.

While the metrics tell an impressive story, a compelling validation of our LCM comes from the genuine enthusiasm and gratitude expressed by iFood restaurant partners.

“Restaurant owners actually called to thank iFood because the notifications felt so personal and human – some notifications even went viral. People who had turned off push notifications started turning them back on. That’s when we knew LCM wasn’t just another recommendation engine – it’s creating real connections that drive value at every touchpoint.”

– Zülküf Genç, Director of AI, Prosus, leading the LCM team