AI x QS = Opportunities

This article rounds off our generative AI mini-series, in which we have covered how generative AI works, the risks, the limitations, and some thoughts on some of the key tools QS’s have, or will, use.

This final article takes a bigger picture view of some of the day-to-day problems we experience as QS’s and how generative AI could present a solution to those challenges.

I wrote a similar article about twelve months ago. I was minded to write this as a quick refresh of my previous article but, given the extent and speed of advancements in the period since, I thought it would be more effective to provide my thoughts on a completely blank canvas.

The approach I have taken to this article is to consider how generative AI might be used for day-to-day quantity surveying tasks and either improve our productivity or performance. I have decided to focus on tasks very much in the QS’s domain, and picked the following: procurement / tendering; applications for payment / payment notices; and change control.

Procurement / tendering

For me, this is where the biggest opportunity exists. The volume of documents which are created for a tender process between contractor and employer and then used again for contractor to subcontractor tenders is well beyond manageable.

Whether you are preparing or receiving tender documents, you are expected to handle a huge volume of information which will more than likely include: lengthy bespoke amendments to a standard form; extensive design documentation (and for key subcontractors you receive it for every design discipline, “just in case”); technical standards specific to the building and client type; and, for subcontractors, the main contractors many policies covering everything that might be relevant.

We could, and should, be very clear about what we want the tenderers to price. This used to be communicated with a bill of quantities. However, given the way information complexities have strained time demands, documents are pushed out of the door and the next person in the chain is asked to take responsibility for making sense of the information; measuring quantities; identifying errors, ambiguities, or discrepancies; and, more often than not, defining the precise scope of the package.

The opportunity for generative AI to make a difference here is immense. Can you imagine asking a generative AI tool “what is the scope of this contract / package?”, “what is the extent of the design responsibility obligation?” or “are there any discrepancies between the design documents and the technical standards”? And what if the generative AI could lead you towards where the answers are, even if it gets you 60-70% of the way you have done most of the heavy lifting in an often time constrained period.

Could we even see a situation where generative AI produces pricing documents? Could generative AI be the key to unlocking NRM compliant bills of quantities and seeing a return to these documents as standard in a tender enquiry pack?

In my view, it is only a matter of time.

Applications for payment / payment notices

I expect this would be slightly trickier, but not impossible.

When we prepare applications for payment and payment notices, we consult various data sources. This would start with the contract where you will find the agreed sum, the price basis of that sum, how the sum is adjusted and how the sum is progressively paid. You would prepare a valuation against this primarily using your own site inspections, progress information (programme), and design information (drawings and specifications) up to a certain date in the month (often termed the valuation date).

I don’t think it is inconceivable that generative AI could be used to prepare an interim application for payment or payment notice. The AI would need access to the right information but it is possible. Some might say it would only be as good as the information the AI accesses, but that issue would be prevalent regardless of whether it was completed by a human or a machine. Maybe using AI would give us time to fix those issues with information flow and presentation, rather than spending the bulk of our time processing the data?

The advantages would be an improvement in the substantiation of applications for payment and payment notices. In theory, every progress percentage should be backed up by information. In reality though, given the speed and volume involved, it doesn’t happen all the time to the required level. This impacts cash flow, and leaves the door open to a dispute.

The outputs would require skilled checking and amendment but, again, to get a head start by the AI asking questions of other data sets would make a huge difference.

Thinking on the challenges in this area, I do wonder if generative AI could help to solve the legal issues associated with the timing and form of payment notices and pay less notices. Given case law is constantly developing and there are those looking to exploit notice provisions to gain a commercial advantage, generative AI could assist those working at the coal face to remain compliant.

Change control

This gets trickier again.

In my mind, I wonder if generative AI could be used to identify variations to a subcontract? This is certainly feasible when the changes are based on design information, however I think it would be very difficult, if not impossible, for generative AI to detect site issues which constitute a variation.

There is definitely mileage here though. What if you could ask the generative AI whether there are variations within a set of design documents, and the generative AI could prepare compliant notices to be issued to your client. What if the AI could indicate whether you are in danger of being time barred because a design change occurred three days ago which you need to notify within five days?

It’s not inconceivable. In my mind I wonder if this is too far away but there is a part of me that feels this is closer to being a reality than I think.

Final reflections

This is a big area and the subject is developing every day. However, I hope that my thoughts and reflections spark some ideas about the possibilities for the future of quantity surveying and how the profession might embrace developments in technology.

Whilst there are plenty of possibilities, you should be very careful about uploading commercially sensitive information to an AI tool. It is likely that you will lose confidentiality over that information and the AI will use it to independently train and learn. There could be all sorts of ramifications here so tread carefully.

This concern probably leads us back to square one because, to realise these possibilities, you may need to either give away commercially sensitive information or develop and train your own AI.

Neither of these options is desirable in an industry which is time and profit poor, so it does lead me to wonder how we unlock the benefits of generative AI. It all feels a little bit "chicken and egg".

Despite this, the opportunities are exciting for QS’s and I am certain that those who embrace this evolution and find solutions to the challenges will thrive.

Back to insights