The J-curve is a familiar concept in private equity and economics. Performance declines before it improves, and the shape of that trajectory looks like the letter J. Anyone who has managed a significant technology deployment has experienced a version of it firsthand. New systems disrupt established workflows. People who were productive on the old process become temporarily less productive on the new one. Support tickets spike, workarounds proliferate, and for a period of weeks or months the organization is measurably worse off than it was before the investment. AI deployments in commercial real estate are producing exactly this pattern, and the depth and duration of that dip is where most of the difference between successful and failed initiatives actually lives.

The first and most consequential decision that determines the shape of that curve is made before any technology is deployed. It is the expectation that gets set with leadership about what the initiative will accomplish and when. “A typical mistake we see is treating the early productivity dip as evidence that AI isn’t working.” said Chris Campbell, Managing Director of Technology Advisory at JLL. “The J-curve is a normal part of transformation. The objective isn’t to eliminate the dip, but to manage it deliberately and shorten the time it takes to get to measurable results.” That fear is understandable. A team requesting budget for an AI initiative is competing against other priorities, and the temptation to promise transformative results in the first quarter is real. But an initiative that promises too much creates a much steeper J-curve, because the gap between what was promised and what materializes in the early months reads as failure rather than as the expected transition period. 

AI has the ability to transform even the hardest tasks needed by a real estate organization. But it can’t do so right away. Realistic expectations are what makes the transition from pilot to production possible. A pilot that was promised to deliver a 40% productivity improvement and delivers 12% gets shut down. A pilot that was promised to deliver 10% and delivers 12% gets expanded. The technology performed identically in both cases. What differed was the framing, and the framing determines whether leadership sees a promising foundation or a failed experiment. Getting that framing right at the outset is the cheapest and most effective way to protect an AI program through its most vulnerable phase.

The people involved in an AI deployment are the other variable that determines how quickly an organization climbs out of the dip. The employees who will be building these systems and the ones whose daily work will change as a result are the ones who can either accelerate adoption or quietly resist it. Keeping them engaged through the transition requires visible progress, which means starting with projects small enough to produce results quickly. “Productivity can be difficult to measure at an enterprise level, particularly in an AI program. We encourage clients to start by identifying specific, repeatable work that can be done faster, better or differently, and then measure the outcome,” Campbell said. A general claim that AI has improved efficiency across an organization is easy to dismiss. A concrete example of a specific team that stopped spending six hours a week on a particular task is not. “You can start small in order to get these really good examples of how it is making people’s jobs better, not just more efficient,” Campbell said. That distinction matters more than it might appear. Efficiency is an organizational benefit. A better job is a personal one, and personal benefits are what drive the enthusiasm that carries an initiative through its difficult phase.

Identifying champions early, and doing so across functions rather than concentrating them in a technology group, is one of the strongest predictors of whether an AI program scales successfully. “We have seen good success with clients that have embedded champions in multiple different functions,” Campbell said. “Champions need to understand the current workflow at a granular level, including the workarounds and friction points that rarely show up in a process map, in order to redesign work, rather than simply automating it.” That granular business knowledge is what separates a useful champion from a well-intentioned one. Understanding how work actually gets done, including the informal steps and workarounds that never appear in a process document, is what allows a champion to identify where AI will genuinely help and where it will create friction. Someone who understands the technology but not the workflow will consistently propose solutions to problems that do not exist.

Those champions typically get formalized into a committee with its own governance structure and mandate, and the effectiveness of that committee depends heavily on whether it has clear priorities to work against. Strategic alignment to business goals is the starting point, but the specific criteria that guide decisions matter just as much. Resilience, cost optimization, and user experience are all legitimate priorities, and they sometimes conflict. A committee that has explicitly ranked them can make consistent decisions about which projects to pursue and which to defer. A committee that has not, tends to relitigate the same tradeoffs with every proposal.

One responsibility that these committees increasingly own is the lifecycle management of AI agents, which is a genuinely new organizational function that most companies have not built infrastructure for. Agents get deployed for specific purposes, and those purposes change or disappear over time. Without a clear process for retiring them, organizations accumulate agents that run without oversight, consume budget, and occasionally produce outputs that nobody is reviewing. “An AI agent performing a business task needs many of the same disciplines we apply to people: a defined role, appropriate training, clear boundaries, performance management, and ultimately a decision about whether it should continue doing the work,” Campbell said. That framing is useful because it brings a familiar management discipline to an unfamiliar problem. Organizations already know how to onboard, train, evaluate, and offboard employees. Applying that same structure to AI agents gives the committee a workable model for governance rather than requiring them to invent one.

The J-curve in AI deployment is not entirely avoidable. Any meaningful change to how work gets done will produce a period of adjustment, and organizations that expect to skip that phase are setting themselves up for the exact disappointment they are trying to avoid. What is controllable is the depth of the dip and the speed of the recovery. Realistic expectations set with leadership from the start, small early projects that produce concrete and personally meaningful wins, champions embedded across functions who understand the business at a granular level, and governance structures that treat AI agents with the same rigor applied to any other organizational resource all shorten that curve considerably. The organizations that build those foundations are not the ones that avoid the difficult phase of AI adoption. They are the ones that get through it fast enough to still have leadership support on the other side.

The post How Real Estate Organizations Can Avoid the AI Adoption J-Curve appeared first on Propmodo.

​