Real estate organizations considering AI deployment should first assess their operations, data structure and strategic direction, according to Propmodo’s guidance. The preparation stage helps teams decide what business outcome they seek before choosing technology. Thomas Mackean, a digital strategy expert at JLL, says AI features age while outcomes endure; clients who gain value start by trying to fix an outcome, then use AI if it proves the best way to do so.

Starting with the desired outcome prompts organizations to define success, identify users and simplify the process. They should then audit the data used in decisions and set governance rules. In large firms, teams handle information about different clients, assets and tenants, so access may need to be restricted by function. Mackean says segregating agents can limit what information each can access and prevent AI from learning across clients. Early safeguards can restrict access to sensitive data and keep a record of who accessed what.

Integration needs ongoing attention because property management systems, lease platforms, CRMs and market data providers are often separate systems connected imperfectly. A successful pilot does not establish readiness for month six, Mackean says, when the model, data and portfolio may have changed. Updates can silently alter data structure or interpretation, degrading outputs. In one example, an office utilisation agent showed an unexplained 15% increase across all buildings over a few weeks, even though a hybrid-work policy change should have lowered utilisation. An investigation linked the rise to an upgraded security badging system with more advanced tailgating logic and additional Wi-Fi access points, deployed by separate workstreams.

Mackean advises building integrations that can use more than one major AI provider to avoid vendor lock-in as models improve. Organizations should also match model tiers to tasks: lightweight models cost less, while sophisticated ones cost more. He warns that 90 percent of staff could end up using the highest level on tasks that do not need it, driving significant costs. Clear guidance on which workflows warrant each tier can help manage spending as use grows.

Before tools go live, organizations need to define outcomes, audit data, map integrations requiring continued maintenance and build flexibility to adapt as models improve. The work is difficult and time-consuming, especially for firms unused to examining operations in detail, prompting some to bring in outside advisors before choosing a deployment path. The guidance says pilots are more likely to produce results worth scaling when companies take the preparation stage seriously; skipping it can leave them discovering readiness gaps after spending the budget.

Real estate organizations considering AI deployment should first assess their operations, data structure and strategic direction, according to Propmodo’s guidance. The preparation stage helps teams decide what business outcome they seek before choosing technology. Thomas Mackean, a digital strategy expert at JLL, says AI features age while outcomes endure; clients who gain value start by trying to fix an outcome, then use AI if it proves the best way to do so.

Starting with the desired outcome prompts organizations to define success, identify users and simplify the process. They should then audit the data used in decisions and set governance rules. In large firms, teams handle information about different clients, assets and tenants, so access may need to be restricted by function. Mackean says segregating agents can limit what information each can access and prevent AI from learning across clients. Early safeguards can restrict access to sensitive data and keep a record of who accessed what.

Integration needs ongoing attention because property management systems, lease platforms, CRMs and market data providers are often separate systems connected imperfectly. A successful pilot does not establish readiness for month six, Mackean says, when the model, data and portfolio may have changed. Updates can silently alter data structure or interpretation, degrading outputs. In one example, an office utilisation agent showed an unexplained 15% increase across all buildings over a few weeks, even though a hybrid-work policy change should have lowered utilisation. An investigation linked the rise to an upgraded security badging system with more advanced tailgating logic and additional Wi-Fi access points, deployed by separate workstreams.

Mackean advises building integrations that can use more than one major AI provider to avoid vendor lock-in as models improve. Organizations should also match model tiers to tasks: lightweight models cost less, while sophisticated ones cost more. He warns that 90 percent of staff could end up using the highest level on tasks that do not need it, driving significant costs. Clear guidance on which workflows warrant each tier can help manage spending as use grows.

Before tools go live, organizations need to define outcomes, audit data, map integrations requiring continued maintenance and build flexibility to adapt as models improve. The work is difficult and time-consuming, especially for firms unused to examining operations in detail, prompting some to bring in outside advisors before choosing a deployment path. The guidance says pilots are more likely to produce results worth scaling when companies take the preparation stage seriously; skipping it can leave them discovering readiness gaps after spending the budget.