I, and my company, Datex Property Solutions, have been building, deploying, scaling and supporting workflow-based automation systems for the commercial real estate (CRE) vertical for over 15 years (the company has been in business for 25 years).
This covers recurring tasks across Leasing, Reporting, Site Visits and Lease Compliance Tracking & Enforcement, so I have a pretty clear sense of what has to go right for the efforts to succeed, and how/why such endeavors go sideways.
With AI starting to fold into the mix, what follows are five observations that would-be adopters need to keep in mind.
- Automation is a No-Brainer: Throwing systems at problems that previously were done by humans with spreadsheets can be a no-brainer in terms of productivity (twice the work in half the time), process (systematic, repeatable), governance (track, manage, enforce) and cost (people are expensive).
- Measure Twice, Cut Once: Being rigorous in codifying jobs, outcome goals and constraints in the context of the macro business objectives is critical, as it defines the field of play – often cross-departmental – that has to be orchestrated and navigated for the effort to succeed. My joke here is that people want more workflow and notifications…until they get more workflow and notifications.
- Process and Governance is Political: Because systems are all about efficiency and getting the job done, they can step on hierarchies, parallel or conflicting business objectives and departmental boundaries. Most basically, they can step on human feet, so ensuring stakeholders and sponsors are clear and on-board with the new SOP (standard operating procedure) is essential. The human equation is as likely to be the critical point of failure as any technical guffaw.
- Promise is Not Always Practice: Sales people oversell capabilities, the client’s underlying data isn’t always perfect, stakeholders don’t always follow the SOP, people quit, new people join, previously unrealized needs surface, so having backstops, such as top-shelf human support, project management, and technical workarounds when things go awry NEEDS to be baked into the process, especially until the rollout is stabilized and real adoption is realized.
- FOMO and AI: I see it, hear it and feel it in every pitch. Clients have a real Fear of Missing Out, and to be clear AI, like automation, is real, transcendent and transformational. But go in knowing this: 1) Agents are Dumb – they have no sense of history, nuance or how to weigh competing goals; 2) LLM’s are Sociopaths – No doubt born of the “tech bro” manifesto of “Fake it until you make it,” they routinely lie, and sheepishly apologize when caught, then lie again. LLM’s are devoted practitioners of the “often wrong, never confused” ethos, where ignorance meets arrogance; 3) LLMs will NOT deliver AGI – Think of an LLM as a “filtering and assembly” machine that feeds a massive input of Potentiality on one side (the Model) and outputs its best Probabilistic assessment (Inference) on the other side. It’s why the same question asked multiple times can yield deeply varying answers EACH time. LLM’s are part of the AGI equation, but real AGI will require a different appendage to be developed; and because of this, 4) Really Bad Things Can Happen: A story making the rounds this week was by the maker of software that car rental operators use to run their entire operations, where the AI broke a bunch of rules, “intelligently” sourced passwords that it shouldn’t have had access to, and in one API call, deleted their production database. It took 9 seconds. (“An AI Agent Just Destroyed Our Production Data. It Confessed in Writing.” https://x.com/lifeof_jer/status/2048103471019434248?s=12 )
The lesson, as always, is NOT to avoid going into the ocean, but rather to be smart, have a plan, iterate that plan, focus on manageable bites, and be holistic.
You CAN do this.
— Mark Sigal, CEO, Datex