Provost and Ipeirotis Publish eROI: Why Compass's Likely-to-Sell AI Made Nine Figures and Its Pricing Tool Was Rightly Shelved

arXiv 2607.23733·medium signalXBlueskyLinkedInCopy link

This paper (arXiv 2607.23733) tackles the catch-22 that teams cannot estimate ROI until they know whether an AI project will work, but cannot know without building it. The eROI framework decomposes each bet into three separately rated components — Value if Successful, Likelihood of Success, and Investment Required — so a team can argue a product would be valuable if it worked while independently weighing how likely that is. The worked case is real-estate brokerage Compass, where a Likely-to-Sell recommendation product went on to account for nine figures in annual gross commission revenue while a championed Time-on-Market pricing tool was correctly killed; a simple ROI estimate could not distinguish them. The framework also asks whether enough good ideas are on the table before ranking, and pushes toward funding a portfolio rather than only the top-ranked bet.