Scaling 60,000 AI Agents for Maximum ROI at Prosus
In our latest report at Prosus, we analyzed over 60,000 AI agents built on our internal Toqan platform to uncover the real-world dynamics of scaling artificial intelligence. We discovered that a small fraction of these “AI colleagues” drives massive business returns, requiring a strategic shift toward standardized use-cases and decentralized cost control. This data-driven analysis serves as a practical blueprint for leaders globally who are aiming to transition into autonomous, AI-first organizations without getting lost in the technological hype.
Points clés
- Our 40,000 global employees built over 60,000 AI agents utilizing our internal “Toqan” platform
- Only about 2% of these AI agents drive the majority of business impact, revealing a distinct “Power Law” in enterprise AI deployment
- Data Analytics and Market Intelligence are the most common AI agent departments (18%), followed closely by Operations (15%)
- Across our diverse organizational portfolio, companies organically converge on 20 core AI use cases, such as demand forecasting and security scanning
- High-impact “Value Agents” deliver massive ROI, with one of our AI systems managing a third-party marketplace that independently generates $83M in annual revenue
- Another deployed AI assistant specifically designed for vacation rentals successfully increased customer booking rates by 138%
- When measuring personal productivity, 82% of AI agents save less than 20 hours per month, while the elite top 1% save the equivalent of thousands of full-time employee hours
- We observed diminishing returns on newer AI models like Claude Sonnet 4, as employees strongly prioritize platform stability over regularly chasing the newest model upgrades
- To manage volatile operational costs, we implemented a two-tier financial system where individual business units must pay for the “Production Tier” when usage exceeds 200 requests per hour
- Scaling global adoption requires strategic motivation; we successfully drove rapid engagement using internal co-creation contests that awarded up to $25,000 in cash prizes
À retenir
If you want your organization to survive the impending evolutionary leap toward AI, simply build 60,000 bots, cross your fingers that 2% of them actually do something useful, and bribe your employees with $25,000 to use them. Jokes aside, the best strategy for an AI novice is to start small: deploy the 20 proven core use cases, aggressively test for ROI through the “delete it tonight” method, and force your individual departments to foot their own compute bills before they accidentally bankrupt you. Eventually, AI will rebuild your entire company from the ground up, leaving you plenty of time to practice your golf swing while your digital colleagues do the heavy lifting.
Sources
Quiz sur le document: 10 questions
