Why AI Projects Fail in Real Estate (and How to Actually Ship)
MIT found 95% of enterprise generative AI pilots deliver no measurable P&L impact, and the cause is implementation, not model quality. JLL found 88% of real estate players piloting but only 5% hitting all their goals. The five concrete reasons projects stall in real estate, and the discipline that ships a system that lasts. Full version in French.
Why AI Projects Fail in Real Estate (and How to Actually Ship)
English brief
Most AI projects fail on implementation, not models. MIT's 2025 "The GenAI Divide" study measured that 95% of enterprise generative AI pilots deliver no measurable P&L impact, with only about 5% driving real revenue, and named enterprise integration, not model quality, as the root cause (MIT NANDA, via Fortune). Real estate mirrors it: JLL found 88% of investors and owners piloting AI but only 5% hitting all their program goals, and 60%+ unprepared to scale (JLL).
The five concrete reasons projects stall: no audit or prioritization, data that isn't ready, no human in the loop on costly decisions, no capability transfer (dependency), and automating the wrong end, MIT found the biggest ROI in back-office work, not the visible front-office. The discipline that ships is audit, then roadmap, build, deploy on your own infrastructure, and transfer, starting small and proving before extending. Our proof is on the record, from a deal-screening build that cut underwriting from fifteen hours to three minutes to the memo automation a firm now runs itself. The full version is in the French edition; to scope your own, book a call.
Questions and answers
Why do most AI projects fail?
On implementation, not models. MIT's 2025 'The GenAI Divide' study measured that about 95% of enterprise generative AI pilots deliver no measurable P&L impact, with integration, not model quality, as the root cause. Real estate mirrors it: JLL found 88% of investors and owners piloting AI but only 5% hitting all their program goals, and 60%+ unprepared to scale. The bottleneck is getting to production, not curiosity.
How do you ship an AI project that lasts past the pilot?
By discipline, not a better model. Audit before building and decide what not to build, prepare the data first, keep a human in the loop where errors are costly, deploy on your own infrastructure and transfer capability, and start small, prove, then extend. A system that earns trust on a real bottleneck funds the next one; trying to do everything at once is the surest way to join the 95%.
Related Articles
Claude Google Drive Connector for CRE: Drive, Sheets or API?
Choose Claude’s Drive connector, Sheets sidebar or a scheduled API workflow for CRE deal files. Compare document access, edits, limits and review checks.
Claude Skills for Real Estate: The Complete Guide (2026)
Install and test Claude Skills for real estate: seven downloadable skills, current setup steps, and sourcing and underwriting packs for CRE teams.
AI for Commercial Real Estate by Asset Class
The AI playbook for commercial real estate is not one playbook. It changes by asset class because the data sources, the buy signals, and the underwriting math are different for multifamily than they are for self-storage or hotels. This is the map: what is specific to each asset class, and what stays constant no matter what you invest in.