# 60,000 AI Agents in 18 Months: The Power Law Behind Real ROI > Author: Chris Jon Graf (AI Strategist & CEO) > Updated: 2026-10-07 > URL: https://ai-outsourcing.ch/insights/60-000-ai-agents-in-18-months-the-power-law-behind-real-roi ## Summary The true ROI of AI agents is highly uneven: at Prosus, 2% of 60,000 agents drive nearly all business impact. Success comes from redesigning work around the agent, not just bolting it onto existing processes—and testing every agent with the delete test. ## The power law: 2% of agents deliver nearly everything In 18 months, Prosus built more than 60,000 AI agents with 40,000 employees on its internal platform Toqan. The report "The Coming Age of AI Colleagues" reveals just how uneven the real ROI is. The key finding: a power law determines success—2% of the most active agents drive almost the entire business impact. **2%** — of the most active AI agents at Prosus deliver the bulk of business impact The numbers are stark: 82% of productivity agents save less than 20 hours per month, 17% save between 20 and 173 hours—the upper end roughly equals one full-time equivalent—and under 1% deliver thousands of hours. For value agents, the majority generate less than $1 million per year, some between $1 million and $10 million, and one outlier in the affiliate marketplace reaches $83 million. This distribution is classic power law: a few top performers dominate the outcome. > **What this means for you** > > You cannot manage agent ROI through the total number of agents built. The decisive factor is whether you systematically find and scale the outliers. ## Five traits of the top performers - They solve a clear, recurring problem. - They are shared team-wide rather than used personally. - They are connected to internal systems. - They make previously unprofitable work profitable—in the long-tail restaurant example, this led to 119% more orders and 73% better customer retention. - They are measured and scaled, often with the delete test. ### The delete test: what happens if the agent disappears overnight? Instead of collecting endless metrics, Prosus asks a simple question: What would happen to revenue or costs if this agent were deleted? That lens separates nice-to-have experiments from real value creation. Knowing which processes are truly worth automating is a strategic discipline—not a tooling decision. ## The electric motor lesson: redesign work, don't just bolt on Economist Paul David showed in 1990 that the biggest productivity gains from electrification came only when factories redesigned production around the electric motor—not from merely replacing the steam engine. Prosus sees the same pattern with AI agents. Attaching them to existing processes yields early gains. The big leaps come when you rethink the work itself. ## What this means for DACH leaders The Swiss baseline is encouraging but not yet productive: according to an EY study from July 2026, 89% of employees use AI, but only 55% use it deliberately in business areas, and 31% are still in the pilot phase. This gap shows why the Prosus lesson matters: it is not about adding more tools, but about finding and scaling a few agents with real leverage. A deeper discussion on why AI is a management issue—not an IT project—can be found in the [Swiss AI podcast](https://www.ki-podcast.ch/ki-im-mittelstand-management-thema-nicht-it-projekt). ## Your next step You do not need a 40,000-person company to apply these patterns. Start with a recurring process you can actually measure, and define your own delete test. Full implementation is a separate project—but the first step is clear. For common strategic mistakes to avoid, listen to this [Swiss AI podcast on the three biggest errors in SME AI strategy](https://www.ki-podcast.ch/ki-standort-schweiz-kmu-strategie-und-globales-rennen). ## FAQ ### How many AI agents has Prosus actually built? Prosus built over 60,000 AI agents in 18 months with 40,000 employees on its internal platform Toqan. ### What is the power law in AI agent ROI? 2% of the most active agents drive the bulk of business impact. 82% of productivity agents save under 20 hours per month, 17% save 20 to 173 hours, and under 1% deliver thousands of hours. ### What is the delete test? The delete test asks what would happen to revenue or costs if an agent disappeared overnight. Prosus uses it to separate real value creation from experiments. ### Why does bolting agents onto existing processes only bring limited gains? Similar to electrification, the biggest gains come when work is redesigned around the agent. Early wins come from attaching; large leaps come from rethinking the work itself. ### What can DACH leaders learn from the Prosus report? Focus on a few recurring processes with clear metrics, share them team-wide, and scale the outliers. The EY study shows 89% use AI but only 55% do so deliberately—the opportunity lies in focus. ## Sources - [What 60,000 AI agents teach us about returns – ai.nl](https://www.ai.nl/en/articles/prosus-report-60000-ai-agents-coming-age) - [Deploying agentic AI at scale – Prosus press release](https://www.prosus.com/news-insights/2026/deploying-agentic-ai-at-scale-from-the-company-that-built-60000-agents) - [The Coming Age of AI Colleagues – Prosus report (PDF)](https://www.prosus.com/~/media/Files/P/prosus-corp-v2/documents/the-coming-age-of-ai-colleagues.pdf) - [KI setzt sich in Schweizer Unternehmen zunehmend durch – kmu.admin.ch / EY](https://www.kmu.admin.ch/de/ki-setzt-sich-in-schweizer-unternehmen-zunehmend-durch) - [Paul David (1990): The Dynamo and the Computer – AER](https://www.aeaweb.org/articles?id=10.1257/aer.80.2.355)