My December 2024 article was a humorous look at what artificial intelligence could do for electrical estimators. I had a lot of fun with that article, but it is time to get serious. This morning, I had a discussion with the president of the estimating software company I use. The following are some ideas he gave me about what A.I. could do for electrical estimators in the future.

Specifications

My digital takeoff program already uses a bit of A.I. for its symbol recognition feature. However, there are a lot of ways its use can be expanded, starting with specifications. Having A.I. read the specs would save estimators a significant amount of time. 

As I write this, I realize just how complex and difficult reading a specification can be. There are so many things to look for, such as the bid form, scopes of work, voltage drop requirements, manufacturers and grades of wiring devices, conduit and wire requirements, materials that need quotations, testing, scope to exclude, and spare part requirements. The A.I. would have to be able to deal with vague and dangerous language, such as the requirement to review the documents of all other trades for electrical work. 

So much of what I do while reading the specifications and plan notes is based on my experience, which changes as I’m exposed to new information. The A.I. would need to be able to learn, just as I do.

The A.I. would also need to understand the National Electrical Code as well as local codes from cities and states. Additionally, many states have very complex energy codes, such as California’s Title 24. All of these codes change, many on a three-year cycle, so it would need to keep up with the what is different this time around.

Takeoff

Let’s look at the takeoff side of estimating. I can imagine some great things A.I. can do to make this go faster. The first is to make symbol recognition 100% accurate, if possible. However, will it be able to handle the same symbol used for five different lighting fixtures? Will it be able to deal with a note that says, “All fixtures in this room are Type A?” Will the A.I. be able to learn the many different ways fixtures and devices are represented? What will it do when it runs across a symbol that is not on the schedule? 

Measuring branch is usually the most time-consuming part of a commercial building estimate. It would be awesome if A.I. could do that for me. Almost 100% of the time, today’s electrical drawings do not have the branch conduit developed on the drawings. The A.I. will have to read and understand the wiring requirements indicated on the drawings, as well as develop the branch conduit and wire needed to supply power and control to everything on the plans. 

The A.I. will also need to understand the NEC to size the conduit and conductors needed. It will need to make voltage drop calculations as required by several different documents, including the specifications, notes on the plans and the NEC. 

Notes

Speaking of notes, they are already a significant difficulty for estimators. Notes can cover large amounts of material and labor costs in a single sentence. A single note can require additions to feeders, equipment, motor connections and switchgear. They can be very complex or simple, yet expensive. 

Consider the following note I found on the single-line drawing on a set of plans: “All 120V circuits over 50 feet will be upsized to No. 10 wire.” Will the A.I. find that note before doing the branch takeoff? If not, will it be able to correct the takeoff? 

Human error

Finally, let’s talk about conflicts and mistakes. Every project I do these days has conflicts between and within the bidding documents. Every project has mistakes and typos. It is often impossible for me to solve these problems without consulting my customer. Usually, we need to make some assumptions before finishing the estimate. The A.I. will need to be interactive with the estimator to solve these problems.

The ideas, observations and questions above were just the first things that popped into my mind. I think the progress of A.I. will have to be made in small steps. I do not believe a fully functioning A.I. estimating system can be delivered without significant advancements at this point.

My biggest problem with thinking A.I. may help electrical estimators more in the future is this: Considering the complexity and poor quality of bid documents, will the A.I. slip into a downward spiral, ending in the A.I. equivalent of the blue screen of death? Is there a future in A.I. psychotherapy?

Rodrigo / stock.adobe.com