The Apocalypse Is Not Nigh! (Nor AI)
AI in CRE Inspections
First, let’s cover the obvious: AI is both a terrific tool and, as I wrote in letters recently to brokers, dumber than a bag of hammers. It is also my experience that if I ask dumb things, I get dumb responses. GIGO is a universal constant in the programming world.
One of my earliest questions, once I began to figure out how to apply AI to my inspection work, was this: “What percentage of my job can be replaced with current and foreseeable (0-1 years) AI systems?”
The answer was encouraging. AI will not be fully replacing me anytime soon. Per Claude, which is my go-to for most projects, and Grok, which is just handy for fast questions, about 15-25 percent of my job can be automated with AI.
The reasons are pretty easy to understand. A large portion of my work is in the field, making observations, performing hands-on testing, and developing judgments from those observations. This is 40 percent of my workload that cannot be delegated to an AI agent. It is hard enough to delegate to a competent human agent. The skill of observation is not well-distributed through the population, nor is an insatiable curiosity that creates a very broad range of knowledge prevalent.
But, wait! I see someone has a question!
“Can’t we upload pictures and train an AI to perform the observations?”
Nope, not yet. There are three factors that come into the picture (yes, a Dad joke!) when we try to use photos in place of eyeballs. The most obvious one is that photography is a 2D representation of a 3D system. Depth perception is critical.
When I inspect a foundation and see a crack, it triggers immediate responses. Is that hairline crack superficial or does it run deep into a structural member? Is the wall bulging outward or inward? By how much, ¼ inch or 2 inches?
I can stand in one spot, shift my head just a bit, and instantly judge distances, contours, and significance. An automated system, lacking true stereoscopic vision, can not replicate that process without very specialized and expensive tools such as LiDAR.
Which brings us to the second objection on using photos uploaded to an AI. The amount of fine detail that the human eye processes is (currently) unmatched by a photograph. Every photograph has limitations, whether it is the sample rate (megapixels), the compression rate which leads to a small degree of data loss, and it represents a single instant.
The last limitation is the training of the AI agent to ‘read’ the photograph and provide an accurate comment regarding the condition of the element under inspection. It takes thousands upon thousands of pictures to begin the process, and those pictures need to cover every scenario, every material, every lighting condition, all the variables, to properly recognize a condition and report on it. This assumes that every condition, material, etc. can be trained. The reality is that there are always exceptions and, for AIs that have a tendency to ‘hallucinate’, this is a near fatal flaw.
I think we have answered the ‘use AI with photos to replace the inspector onsite’ question. Let’s move on to my second function.
Writing the Report with AI – But Not All of It
I joke that 90 percent of our job is making accurate observations with great judgment, but 100 percent of our pay comes for delivering accurate reports.
Report writing is not the fun part of the inspector’s job, but is a top-two critical one. AI can actually be helpful here. Just for starters, AI can be used to generate a series of strong professional commentaries for inclusion into the reports.
As an example, let’s consider parking lots. These are frequently neglected and suffer substantial material defects. They also share close similarities in the nature of those defects. I used AI to create comments for material defects by directing it to ASTM 6433, Standard Practice for Roads and Parking Lots Pavement Condition Index Surveys, and asking it to generate ten comments about parking lot defects. Once I had the list, I edited it down to the ones commonly observed in Idaho where I work. Now, instead of manually creating the commentary for the report or using very vague language (all to common in inspection reports, unfortunately), I have concise two-sentence comments. I built another comment structure to evaluate severity, With two taps in the software, I or my assistants have a competent comment describing the problem and level of concern.
Nerdy, but effective.
The other report section that I use AI for extensively is the executive summary. I like doing things efficiently and executive summaries always annoyed me since I get to repeat the information I already wrote into the body of the report. Now, I feed the nearly complete report to the AI with specific instructions on how to generate the type of executive summary I want, and it does the consolidation for me. I still have to proofread it, but that single function reduces my time on that task to about five minutes, instead of 15 minutes to two hours, depending on the size of the project.
The various AIs I queried indicated a 30-50 percent time advantage with the AI. I would estimate that we are at about 10 percent improvement right now.
Again, that guy with the questions!
“Why can’t the AI do all the report-writing?”
The simple answer is that AI does not have the professional discernment to know how and when to use even the pre-crafted comments and cannot reliably generate anything competent that is ‘unusual’ in nature.
AI is great at mechanical functions, and we use it there. For those situations where judgment and expertise is required, we default to humans. We take no chances on the AI ‘hallucinating’ a condition or rating that does not match the reality of the structure.
Lastly, using AI for the back-office functions.
I replaced myself with an office assistant, and the first job I off-loaded was the accounting. Our accounting is not complicated but qualifies as tedious. We have since cut the time on this from about an hour a month to five minutes a month. Definitely a win!
We also use it to run a first pass on property research prior to issuing a bid. Rather than having staff traipse across the internet gathering data, we send the AI out to collect it and provide us with a summary. With that in hand, we can start estimating our time on site, report-writing time, recommended ancillary services, and such. That same research gets included into the proposal, along with our bid numbers. The AI drafts the proposal, proofs it for errors, and sends it back to us for review. A couple of hours of work for larger projects is reduced to 30 minutes or less.
We also generate much of our marketing with AI. Once we got AI to stop being wordy and constantly trying to sell, it worked well for us at generating informational pieces. We still review and edit because we have trust issues, but that’s just us.
We don’t use it for all of our articles. This one, for example, was written by a human. We tried to get an AI to draft it.
The result could be best described as, “Blech!”
Lots of buzz words and boring text. Here is an example: “At Calibre Commercial Inspections, we treat AI as a workflow accelerator.”
We are an inspection company specializing in commercial property, not a trendy tech firm. We deliver clear and concise reports in a timely fashion. I suppose if we wanted to be fancier, we could have a workflow accelerator. Instead, it is a tool, nothing more. Like all tools, it serves a purpose or it stays in the bag.
Speaking of keeping this particular tool in the bag, there is one place we will never use AI.
I have a personal animus to non-personal communication. Phone trees with 87 different options, none of which allow me to talk to a human, are terrible. We might send you to voice mail if everyone is busy, but we’ll call you back personally. If you email us, you can expect a human to email you back. There is some automation, for scheduling notices, for example, but only for the routine elements.
If it is a client-facing function, we will have a human in the process.
The Pragmatic Reality
AI is excellent at synthesis, structure, and first-pass drafting, and we use it for that, but It is not a substitute for skilled field inspection and professional judgment. AI is in our tool bag, helping us improve our quality and turnaround times, helping with the informational aspects of our marketing, and handling mundane office tasks.
It will not replace the dedicated and skilled inspector and the inspection team any time soon.