the evidence he leads with is hard to dismiss<\/a>. Prosus analyzed more than 60,000 AI agents deployed across 40,000 employees and found roughly 2% of those agents generated a disproportionate share of measurable business value. The rest produced real but diffuse gains, individual time savings and convenience, that show up in adoption dashboards but don’t move financials. Three Prosus portfolio examples anchor the argument: an affiliate marketplace agent projecting $83 million in new annual revenue, a restaurant support agent lifting orders 119%, and a travel platform agent driving 138% higher conversion.<\/p>\nWhat this means for your business<\/h2>\n
Agent count has become the vanity metric of the current AI cycle, and CIOs who report it to the board are quietly setting themselves up for a credibility problem. If your organization has dozens of pilots running and you can’t name which three would crater a revenue line if deleted tonight, you’re in the majority and that’s not a comfortable place to be. The Prosus data doesn’t indict broad deployment; it indicts the failure to distinguish between agents that reshape business economics and agents that replace a sticky note.<\/p>\n
The “delete it tonight” test Maetsa describes is genuinely useful as a governance forcing function, meaning a structured process for deciding which AI investments get more resources and which get cut. Its logic is simple enough to run without a data science team. Ask the business unit owner what happens to revenue or costs if the agent disappears permanently with no recovery option. Vague answers reveal vague value. The uncomfortable reality is that most AI governance conversations inside enterprises are still oriented around risk controls, which are necessary but leave the value-concentration question entirely unasked at the executive level.<\/p>\n
The argument is structurally sound, and the power law pattern, where a small fraction of inputs produces most outputs, holds across enough analogous domains, customer revenue, sales team performance, investment returns, that its appearance in AI deployments is more predictable than surprising. The practical risk for CIOs is that the political economy inside large organizations rewards participation over prioritization. Every business unit wants its AI initiative treated as strategic. Governance that honors that pressure equally across all departments isn’t governance; it’s budget appeasement dressed up as a portfolio. The organizations that will pull ahead are the ones willing to defund the 98% aggressively enough to compound the 2%.<\/p>\n
Watch whether your next AI portfolio review asks any question beyond adoption rates and use-case counts. If the answer is no, the budget allocation conversation you’ll face in 12 to 18 months, when boards start demanding ROI proof rather than deployment proof, will be harder than it needs to be. The leading indicator to track now is whether any of your current agents would meet the deletion test with an immediate, specific, revenue-linked answer. One or two that do is a starting point. Zero is a governance problem hiding behind a metrics problem.<\/p>\n
Concept deep-dive: Power law distribution<\/h2>\n
A power law distribution describes systems where a small number of inputs, customers, products, or in this case AI agents, produce results so much larger than average that the distribution can’t be captured by a simple mean. Think of it as the 80\/20 rule taken to an extreme: the top performers don’t just outperform, they outperform by orders of magnitude. In enterprise AI, this matters because average ROI across an agent portfolio masks whether any single agent is doing the real work or none of them are.<\/p>\n
Based on reporting from The AI power law: Why 2% of AI agents create most enterprise value<\/a>, originally published 2026-07-10 03:40:00.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"Share with your CIO Sasol’s Bramley Maetsa makes a pointed case that most enterprise AI programs are measuring the wrong things, and the evidence he leads with is hard to dismiss. Prosus analyzed more than 60,000 AI agents deployed across 40,000 employees and found roughly 2% of those agents generated a disproportionate share of measurable […]<\/p>\n","protected":false},"author":1,"featured_media":5003,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[142],"tags":[185],"tmauthors":[],"class_list":["post-5002","post","type-post","status-publish","format-standard","has-post-thumbnail","category-ai-agents","tag-cio"],"_links":{"self":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/posts\/5002","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/comments?post=5002"}],"version-history":[{"count":0,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/posts\/5002\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/media\/5003"}],"wp:attachment":[{"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/media?parent=5002"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/categories?post=5002"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/tags?post=5002"},{"taxonomy":"tmauthors","embeddable":true,"href":"https:\/\/workai.tv\/news\/wp-json\/wp\/v2\/tmauthors?post=5002"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}