Artificial Intelligence (AI) is here to stay. It's the future, but is it the future of everything? Is ChatGPT ready to replace estimators? Read below to see where AI is right now when it comes to estimating, and what role I see it playing, at least in the near future.
I looked around for the best AI estimating tools available, and I found Togal.ai, Kreo, buildxact to be the most popular or at least these are the ones spending the most on publicity.
I did use Togal.ai briefly while it was still giving a limited time free trial. (It doesn't now by the way). The result was underwhelming. Lack of features and lack of accuracy are the big issues at the moment. Let's look at them one by one.
What can the AI do?
AI software right now is using image recognition to recognize objects on plans and count them, calculate areas or measure length. All the things you need to do for a takeoff. So far so good.
You can also select different items and group them together. See below, a trial estimate where I took off flooring areas and door counts.
While overall, I was actually surprised that it took off the floor areas very well, recognizing all the walls. But I did notice that it did a few weird things. See below.
Nothing too major, but you have to fix these kinds of errors. Plus of course you have edit or delete stairs, elevators etc.
After it does all the flooring you still you have to select each room that has the same flooring finish and group it together. Which I managed to quite easily as well.
You can also easily cut, merge, edit and group items so all the basic features are there and quite intuitive.
It can also export everything to excel so that you can then add waste and convert it to a bill of quantities or just copy paste the data in your excel template.
What can't AI do?
Well, a lot. Firstly, it won't do what specialized trade specific software can do. For example, it doesn't have the features that a flooring estimating software would have. Like calculating tile waste, creating patterns, carpet seaming etc.
Another issue comes, when count is to be done based on text recognition using multiple texts. For example, if you want to takeoff doors using a door schedule and you want doors with same width and height, same material, shape and hardware set to be part of a separate count, then you better do it yourselves.
Even general estimating software like On Screen and Planswift have various features that are in built or customized for you. For example, in On Screen you can select an area and designate it as a separate bid area or typical area. This automatically separates the quantities from the rest of the takeoff and then you can multiply those quantities easily either within the software or in excel. This is actually extremely helpful for multifamily or any large project.
It also can't read and understand notes and detect discrepancies in plans or information. I don't think anyone is expecting that at this stage, but it needs to be kept in mind.
Is it Accurate?
The short answer is No. Why? Because Estimating needs 100% accuracy. One mistake in estimating could mean anything from hundreds to millions of dollars. It's not about time saving or being efficient or even fancy. It's about accurate quantities, full stop.
We have over 100 active clients and typically perform more than 1500 projects per month. And maybe once a year we get an email from a client saying we missed something, and they have to eat the cost. Sorry for the shameless plug, but that is the accuracy we are achieving with humans and non-AI takeoff software.
My tests showed that accuracy is still an aspiration for AI. Sometimes it was quite accurate. But sometimes, I needed to make substantial corrections. It all depended on the task and the quality or type of drawings.
Is it a better alternative to non-AI estimating software?
The only advantage of AI is that it will create the room areas, item counts and wall lengths automatically and instantly. Now, this may seem like a big advantage but it makes the takeoff very complicated, unless the takeoff is very simple and doesn't require a lot of breakdowns.
For example, if you want to takeoff walls for paint but the height varies based on the height of the false ceiling as given in RCP, then the software is useless. But if just want the overall length of walls and there is no variation in height then it can be useful.
The main problem is that quite often the edits required to the takeoff become very complicated in most projects. Creating various conditions and then applying them where they are needed is much simpler than having all the shapes created and then picking one by one and grouping them together.
Also, when humans do the complete takeoff themselves by scratch using traditional software, they understand the job much better and make fewer mistakes.
Just like prevention is better than cure, it is easier to avoid errors than to fix them.
So what should you do?
Right now? Nothing. If you want, buy one license of any of the popular AI software and test it or play with it. See if it is useful for your business. But be very careful and ask your estimating team to check everything thoroughly before bidding.
My Final Verdict on AI in estimating
Frankly, it was a reality check on AI capability. Right now, it is simply an alternative to existing construction estimating software. If you thought you could just speak to the AI and it will magically give you the entire takeoff, then you will be disappointed.
The AI can draw the lines, counts and shapes magically for you in an instant. But after that the estimator will have to get in there and specify everything. I don't think we are going to see anytime soon, not even within 10-15 years, an AI that can read notes, find discrepancies and identify specific materials.
However, I do see within the next 5 years (if funding doesn't dry up) AI software being a genuine alternative to existing software for at least some use cases / contractors.
To do this AI software needs to add functionality including deep material specific functionality similar to existing software counterparts. Because then it will have all the benefits of AI plus everything else that existing software's have.
P.S. Also check out this reddit thread for other people's opinion on AI.