With AI, ‘walk before you run’: Skanska USA exec

Anita Nelson wants you to start small with artificial intelligence.

The chief strategy officer for Skanska USA Building, the U.S. arm of the Swedish builder and developer, Nelson discussed the suite of AI products that the contractor has rolled out to its employees, which it calls Sidekicks. 

The Safety Sidekick, Skanska’s newest internal offering, helps employees access Skanska data on safety and jobsite conditions. It helps employees plan toolbox talks and huddles, and gives them a resource at their disposal.

Here, Nelson talks with Construction Dive about the Safety Sidekick’s origins, how it fits into the company’s current AI suite and the advice she has for builders looking to incorporate AI into their own workflows.

Editor’s Note: This interview has been edited for brevity and clarity.

CONSTRUCTION DIVE: How did the idea of the Skanska Safety Sidekick come about?

ANITA NELSON: We saw a need to make these insights more immediate and more actionable on jobsites. This data is not often leveraged in real time. Sidekick allows us to look at that data in a specific way. 

Using generative AI trained on Skanska-specific safety data and jobsite conditions, it allows us to deliver custom toolbox talks, safety observations and proactive suggestions to workers.

On a day-to-day basis, that means workers can ask the AI tool safety related questions via mobile or desktop and get instant support for planning a morning huddle or addressing a new hazard.

Beyond text, if you’ve got a picture of site conditions that day, you can ask questions related to that image. The tool is also embedded into our workflow, so it’s not an added step in the day, it’s very user friendly.

Safety Sidekick is one of a suite of AI offerings that Skanska has. How does it fit into that?

Our data solutions team, in collaboration with information technology, they’ve developed four secure internal AI tools. 

There’s Skanska Sidekick, which is based on OpenAI’s GPT-4o model. That’s a general use chat bot that protects user business data so you can ask it anything within our organization or outside our organization. 

Then we have three “expert” sidekicks. 

One’s called My Skanska Sidekick, and that searches for key documents on our internal intranet site. So, if you’re looking for a risk management protocol or our sustainability report, or who’s the leader of something, you can ask it for that.

Then, we have an Operational Risk Sidekick, which is one of my other favorites, for our folks in the field. It’s comprised of thousands of case studies and experiences across different project types and building systems. It gives you strategies and mitigation plans for existing risks on jobs.

Safety Sidekick is the newest member of our suite.

What hurdles did you have to clear to make this AI available to your team?

Because we had developed others, it was really about making sure that we got the documents in place.

Our early users constantly give us feedback in real time on what’s working and what’s not. We don’t launch things, you know, to all 3000-plus building employees right away. We take time to pilot, see what’s working, see what’s not. 

I don’t think we ran into any hurdles on this particular launch, because the technology is there.

I don’t want to say it’s as easy as plug-and-play, because my data scientists would tell me it’s more complex than that. But really, a lot of our challenges are going to be about the quality of the data and whether our data warehouse has the right information.

What have been the benefits?

Our early users are definitely reporting time saved when preparing safety briefings and improved jobsite engagement. I think that is critical. 

When you’re preparing for a morning huddle, you might be preparing for that the night before. You might be planning that day before on your commute. 

This makes it really easy for people. They’re not getting to the trailer and figuring it out. They’re not staying late the night before to figure it out.

What advice would you give a smaller contractor who wants to implement this kind of tech?

Start with the data that you know. You start with the low hanging fruit. When you’re looking at generative AI or language models, if you have documents and you want to have them scanned and create a chat bot on top of it, that’s all well and good.

But, if there’s not an infrastructure that you invest in to do that, that’s where you’re going to run into trouble. 

Additionally, we were able to build this ourselves. Don’t think you have to build it yourself.