Prosus (iFood, OLX)

Objective

Get ahead of the moment when the customer relationship moves from apps to agents. The CEO sets out the thesis: the world is changing, customers will use services through agents and assistants, and the companies that lead on this ground become the leaders. Prosus's answer is a proprietary commerce model, the Large Commerce Model, trained on the group's data and connected to conversion, plus a family of consumer assistants meant to hold the relationship at the agent level. iFood serves as the foundation: the model already runs there, ahead of expansion to India and Europe.

The deployment

Prosus, the holding company behind iFood (meal delivery), OLX (classifieds), and a fintech portfolio, has built a proprietary AI model it calls the Large Commerce Model. Rather than buying a general-purpose LLM, the group trains its own model on its transactional data (orders, searches, clicks, repurchases) to drive commerce. At its Prosus Forward event and in its FY2026 annual results (year ended March 31, 2026), Prosus describes this model as already in production on iFood: live across more than 40 use cases, it feeds the homepage, advertising, and notifications. The company puts the effect at +19% homepage conversion, -51% ad conversion cost, and +75% notification conversion. CEO Fabricio Bloisi states that more than 100 million customers are modeled inside the model and that expansion to India and Europe is under way. On top of this, Prosus claims a family of consumer conversational assistants: Ailo, iFood's assistant, is approaching one million users; Zapia, the ecosystem agent, claims six million users across Latin America; Sofia, Luzia, and Toqi round out the set. The stated thesis is that the customer relationship is shifting from apps to agents, and that Prosus wants to lead on this ground in Latin America, Europe, and India. At the group level, the FY2026 results show $9.7 billion in revenue, $1.3 billion in adjusted EBITDA up 44%, and 12% ecosystem revenue growth. These figures measure the group, not AI alone: Prosus publishes no accounting isolation of the model's contribution.

Results Proof B

Two primary Prosus sources agree on the deployment: the official Prosus Forward event release quantifies the Large Commerce Model's effect on iFood (+19% homepage conversion, -51% ad conversion cost, +75% on notifications, across more than 40 use cases), and the FY2026 results transcript has the CEO say that the model powers iFood, that more than 100 million customers are modeled there, and that the India and Europe expansion is under way. An independent press source (ITWeb, October 2025) corroborates the existence of the LCM, born first with iFood. The group publishes earnings-level figures ($9.7 billion in revenue, adjusted EBITDA +44%), but these amounts are group P&L and are not attributed to AI. The conversion uplift is quantified but remains self-declared by Prosus, which promotes its own model, and is not isolated in the accounts or audited: this caps the case at level B (a quantified in-house case study) rather than A.

The internal detail is not public. Here is a proven approach that leads to the same result, to adapt to your stack.

1. 1
   Assembling the data data team

   The group aggregates the transactional history of its platforms (iFood first) to train an in-house model rather than buying a general-purpose LLM.

2. 2
   Training the proprietary model AI

   The Large Commerce Model learns from this data; Prosus claims more than 100 million customers modeled.

3. 3
   Connection to conversion surfaces marketing

   The model's outputs feed iFood's homepage, advertising, and notifications, across more than 40 use cases.

4. 4
   Consumer assistant layer AI

   Conversational assistants (Ailo, Zapia, Sofia, Luzia, Toqi) carry the relationship at the agent level, beyond the apps.

5. 5
   Multi-market expansion data team

   The model, proven in Latin America on iFood, is being extended to India and Europe.