Bringing data science to the construction industry | Blake Abbenante | Data Science Hangout

Transcript#

This transcript was generated automatically and may contain errors.

Welcome back to the Data Science Hangout, everybody. If we have not met yet, I'm Libby. I'm a data science community manager here at Posit. I'm also a data science educator with experience teaching both Python and R, and I am in San Antonio, Texas. If you don't happen to be familiar with Posit, Posit builds enterprise solutions and open source tools for people who do data science with R and Python. We are also the company formerly known as RStudio , so if you know RStudio, you know Posit.

I am joined by the creator of the Hangout and my co-host today, who is behind the scenes in the chat. Rachel, would you like to introduce yourself? Sure. Hi, everybody. Nice to see you today. I'm Rachel. I lead customer marketing at Posit, and as Libby said, I'll be behind the scenes here in the chat, but helping to gather the questions as well.

Yeah, the Hangout is our open space to hear what's going on in the world of data across different industries. We get to chat about data science leadership and connect with other people who are facing similar things to us. We get together every Thursday, same time, same place, so please, please, please come and join us live if you are watching on YouTube, and if you are, there's information in the description for you to join and put it on your calendar.

So, I want to thank everybody. You have all made this the wonderful, welcoming, friendly space that it is today, and we're very committed to keeping it that way, so if you have any feedback about your experience, there's going to be a Zoom survey. It's going to pop up in your browser after you leave this meeting.

At the Hangout, we love hearing from you. It doesn't matter what your years of experience are, what your title is, what your industry is, what languages you do or do not use. We really encourage you to hang out in the chat together. Introduce yourself. Say hello to everybody. What's your role? Where are you based? What do you like to do for fun?

Also, if you have a role that you're hiring for, please share that. If there's a role on your team or that you're aware of, or if you are looking for a role, please be very vocal about that. Let's see if we can connect people in our community with roles.

We are going to ask questions in Slido only, and let's just try this, right? So, there's going to be a Slido link in the chat. So, in Slido, if you are not familiar with it, it is anonymous, but you can put your name there. So, put your name next to your question, and then there's also the ability to upvote questions.

I am so excited to be joined by our featured leader today, Blake Abbenante, Director of Analytics and Data Science at Suffolk Construction. Blake, would you like to introduce yourself? Tell us a little bit about you, what you do, and what you do outside of work for fun.

So I always prefer to, like, do more kind of bespoke analysis, and then when you get to the point of understanding what is really important to the business, like, that is what you codify and put into a dashboard, so you have, like, fewer dashboards, but they're more meaningful, because they kind of measure the things that are more important to the business.

I mean, I feel like you need to meet people, you need to meet your business stakeholders, like, where they are, like, just all of a sudden standing up a bunch of dashboards, especially if they have, like, no context, you know, data, I mean, construction, Suffolk in particular being a really good example that, you know, people went from, like, no dashboards to all of a sudden, you know, three dozen Tableau dashboards, and then just trying to navigate their way through that is not super helpful.

So, like, we're trying to move that kind of stuff to Posit, right, like, does it need to be a dashboard, do you need to interact with it, or is it just I want to get a notification when this thing changes below a certain threshold, and, like, so we look to, like, move and automate that stuff through, you know, Quarto and emails and scheduling through Connect. And then kind of, like, more broadly, our guidelines are, like, if it's just tabular static data that people want to pivot and filter, like, that would be a BI dashboard, but if it's something that, like, you need more interactive calculations or, you know, any kind of interacting with a model or other API, we usually look to build something, you know, custom through and deploy through Connect.

Historical data challenges in construction

Does the construction industry have issues with a lack of historical data for new projects that are really ambitious like that? Yeah, so, I mean, even, like, you know, the Empire State Building itself, like, even given that, you know, the, there's the construction project itself, like, how complex is the building, but then as more buildings get built, like, get built, like, even just where you're working becomes, you know, the challenges of that are compounded as well.

So, I think, I forget what someone, I was talking to someone here, one of our estimators, and I forget, I want, I don't want to say what the number is exactly, but, you know, building in Boston, about, like, 40 percent of the budget is, you know, just moving materials around or getting in or off site because the access is so limited.

And there's no data for that, like, to account for that, but, like, even to your basic question, yeah, there's definitely a lot of challenge with historical data. You know, we generally think we're pretty good about data relative to the rest of the construction industry, and we do have a lot of it, but, you know, some of the tools, especially if you think about scheduled data, they use, it's a tool that's now owned by Oracle. Primavera P6 is, like, the, like, go-to, like, and that is, you know, it's owned by Oracle, like, most Oracle stuff. It doesn't seem like it's been touched in 25 or 30 years.

So, there's not a lot of great ways to get data out. It's kind of very messy, and then how people set up their projects are always different. So, like, looking for, like, a through line between a project that was stood up 25 years ago and how, you know, a project is built today, there's not a lot of continuity in that data.

Getting buy-in from site managers

Do you get pushback from site managers who place more trust in their subjective experience than what the data is telling you? And if so, how do you get buy-in from those people?

Yes, 100%. And, you know, I wouldn't even say, like, I don't know if I frame it as pushback. A lot of it is just, like, they know better, right? Like, you know, there's nuance in the actual management of a project that is not captured in the data that we have available. So, generally, what we want to do before we roll anything out is kind of, you know, identify site managers, superintendents, PXs, people who have more intimate knowledge of projects themselves to get their insight and feedback on what we're doing before we go roll something out.

And I will say, I hope we've done a better job since I've started, but that has, I think, been, again, probably not specific to Suffolk. But one of our big challenges here is that, you know, we've done a lot of good technical work, like the 80% of getting the technical work done, and then we've really dropped the ball on the last 20% work, which is, like, you know, what's the productization of this? Like, how do you get end users in a way that it makes sense to them?

Some of it is as simple as, like, how you label it. Like, when we rolled out, like, you know, the example I gave before where we called it, like, the safety model, it somehow morphed into, like, and the way the data was presented, like, it was kind of like you have a higher likelihood of incident. So, it turned in from, it turned, the initial rollout turned from something being, you know, helpful or useful to, like, a list you wanted to avoid, and that actually would impact how people collected data and, you know, tuned out the results just by the way that we framed up with what the solution was.

So, you know, a lot of lessons that we've learned there, which is the technology itself, whether, you know, the model building itself is, you know, maybe at most, like, 50% of the problem. The rest of it is, like, how do you roll it out and make it something that, you know, communicate with your end users so they understand what you're doing and how it can best make them more efficient at their job. It's not something that we're trying to use data to be big brother and flag people for things that are beyond their control.

AI applications in construction

Yeah, sure. Great question. So, you know, I would say, so Suffolk, one of the maybe benefits of Suffolk is we are still private. So John Fish, you know, founded Suffolk and still leads it, and he is a big proponent of data, and this past year has made kind of a corporate-wide edict. I mean, I look at the sign across the street, it says we're aiming to be the future leader of AI in the built world. So it's kind of like a company-wide initiative to figure out how we can best leverage AI.

So the way that, you know, over the past year it's kind of taken shape, there are sort of two flavors of it. One is kind of what you would imagine what everyone's doing, which is, all right, we have a bunch of SOPs or PDFs somewhere, so like let's throw like those in one that we can build a bunch of chat agents that are, you know, does rag to retrieve specific information about particular domains. That one's less exciting.

But the other areas that we're looking to do it, I mean, there's kind of two large ones, and it's actually in not necessarily either of the ones that you identified, but we're doing in the design, like how can we use AI to augment and speed along design of buildings themselves, and then in the planning of projects. So can we, as I mentioned before, like the kind of planning tool that is the industry-wide standard, P6 Primavera, is very antiquated, and it takes, you know, so just from a user experience perspective, it's not very intuitive or friendly, and where we think we can like gain a lot of efficiencies is automated using AI to kind of, if not create an entire schedule, you know, take something that it may take a plan or two weeks now and get them 80% of the way there in a manner of minutes or hours even.

And then kind of maybe in between area where we've seen a lot of traction is in data transformation. So like we have a lot of data in PDF, so as part of our data pipeline, can we use AI to like extract information out of that and turn it into tabular formats as part of our data warehouse? So we've picked up a lot of stuff like that.

So I honestly, like in the past two or three months, like I feel like, you know, a year ago everyone was really trying to force AI and like put it in, you know, as a square peg in a bunch of round holes, and I think over the past month, like we've had like a bunch of really smart and novel use cases come up. Like literally, I think just this past week one of my colleagues built, he has an AI bot that, or an AI agent that looks at GitHub submissions. So basically like looks for any changes to our data model and then goes and looks through all our GitHub repos and like figures out any references to it and like find out the products that are associated with it. So we can send notifications to say, you know, this change to this data model is going to have a potential impact to these products downstream.

So I honestly, like in the past two or three months, like I feel like, you know, a year ago everyone was really trying to force AI and like put it in, you know, as a square peg in a bunch of round holes, and I think over the past month, like we've had like a bunch of really smart and novel use cases come up.

So yeah, so I think, you know, I think probably still in the short term, aside from like those two outward facing ones, a lot of the efficiencies that we're gaining are more upstream in the data pipeline side at the moment.

And then, oh, in terms of what we use. So as I mentioned, like from an enterprise agreement, we kind of have a partnership with Microsoft. So we started using, you know, the Microsoft Azure suite of open AI integration. Since then we've been, you know, we've been tasked, we've been given the ability to kind of pilot other ones as well. So Databricks has their own native LLM. So we use that somewhat. We're kind of looking or maybe potentially piloting. So we have a bunch of people using individual open AI, like business accounts, but so looking to consolidate that potentially as enterprise as well and try that. And then maybe we're, maybe again, some individuals have used Gemini as well and make, we might look to make that more and more formal as well. So really like trying to run the gamut, whatever's out there, like we're, it's kind of like the wild west still in terms of technology, like trying to figure out which one works best for which scenario.

Getting executive buy-in for data science

How important is it to have an executive like you've got that's championing data science efforts? And is there any advice that you have for people who are sort of in that battle right now where they're trying to get executive champions on board?

Yeah, so I will say, I mean, that is probably the biggest lesson I've learned over my career. I mean, my somewhat snarky, not snarky answer is like, you know, the best way to do it is to, when you're looking for a job, find a place that already has that infrastructure built in. Because I will say that has been like my biggest challenge in my career. Like, you know, you talk to somebody who's really about all about data, like they're the hiring manager, and then you get hired in, and maybe they're like a senior manager or director, depending on what, like, if that's as high as it goes, and then, you know, they could have buy-in, but that doesn't mean that they have buy-in from whomever is above them.

So I, whenever I would interview, I would always ask like, what's the highest role of somebody who has data in their title? And like, that would kind of, to me, be a proxy for like, how bought-in are you on actually using data? But otherwise, you know, you kind of go down the same path that you normally always do, which is like, all right, you find your strategic business partner, you like get your small wins, and then you fight for a budget.

But I wish I had a better answer. But like, to me, like, there is no shortcut to it if you don't have people who, I mean, you know, you can, I've been places where you can take individual data wins to, like, people at the C-level or even right below the C-level, but unless it's like a company-wide, like, edict, like, I find it really challenging for everyone to just, for it always to be, like, top of mind and not something that will get cut as soon as, you know, those are the type of decisions that need to be made.

Are there established practices for sharing data across businesses in construction, or do you only have to rely on your own company's data only?

Yeah. So there are no established practices. I mean, I would say even, not even across businesses, even within businesses, there's very little established practice for sharing data. I mean, there are tools that, you know, construction management tools that allow, you know, multiple different people to kind of access from a operational perspective the same data, like Procore is a big one. And, you know, you can have, like, a GC like us and subcontractors, like, having, or even owners having access to that same instance.

But in terms of just, like, raw data, there's not. And, like, that, like I said, even, like, there's a lot of places, especially when you get down to, like, the subcontractor level where it's all data in Excel, and then, you know, there's not even going to be a standard format as to what that Excel is.

As people think about implementing AI and LLM, like, as I mentioned, like, if you're a large scale construction company and you're only doing, you know, say 30 to 40 projects a year, there is, like, a real understanding that there was, like, a lack of enough data, but there is not, like, an aggregate data broker that, like, has, you know, there's maybe a handful that have enough data construction-wide to, like, make this kind of inferences. So there's, it is a real challenge to think about, you know, do you have enough data even from your own business to, like, to do anything, like, at scale?

Getting started with AI for coders

For coders who are currently working outside of AI, what would you suggest as their first step into AI?

So I'll say, and this is, like, a non-paid endorsement. You know, my first foray into, like, using AI was using LangChain, and, like, I, that was just kind of a disaster for me. Like, I could, like, it doesn't, didn't really click. And then recently, like, we've done some, you know, kind of other integrations, and then over the past couple of months, using the ellmer package through Posit and also, like, the chatlas one as well, we've started both using. I think, like, makes the, like, coding to AI, like, much more approachable.

So I would, like, if you have not used, like, the Posit suite of AI integrations, I think, like, that makes it all very approachable. That would be how I would recommend getting started, and it's really, it is really easy, then, to integrate it to, like, any other kind of work that you've done or if you do happen to be, you know, Posit or any of the Posit suite of tools. So building it into a Shiny app or whatever else. I think that is probably the easiest way to get started or get your toes wet.

Supporting the Posit tool suite alongside enterprise tools

I'm really curious from a kind of operational perspective how support works for the set of Posit tools you use alongside that big enterprise-y stack of Azure Databricks, Power BI. How do you carve out support there? Is that dedicated support or is it kind of shared?

Yeah, that's a great, great question. You like kind of hit on a, I don't call it a sore spot, but like we kind of almost do run two parallel environments. We have like the Microsoft environment, which is fully supported by IT. And then we have a separate AWS environment, which is where Databricks and all our Posit tools run. It's split that way prior to me being part of Suffolk. You know, we've been kind of had some inroads over the past six months or so about like, can we bring them back a little together? It doesn't necessarily make sense for us to be so separate, especially now in the world of AI, where we want to integrate some of the AI agents that we're building on this AWS side as a co-pilot.

So for the Azure and all the Microsoft side, we do have support from IT. On the AWS side, kind of be careful what you wish for. We're a little on our own, but we have to support our own. We do have some outside contracted support, like MLOps or excuse me, DevOps help. But generally like, you know, I kind of administer all of our Posit tool suite at the moment, but it is my hope that over the next year that we can have more support from our IT team.

The value of data science for the construction industry

If you were sharing with other construction companies who aren't really using data as much today, how has the data science made an impact? So what impact has data science made at Suffolk?

Yeah, so, you know, I think if you think about, like, if you think of the inverse, which is, you know, just generally in the construction industry, like most, a lot of the people are, it's an older demographic in general, right? Like a lot, like, so there's a lack of resources and it becomes, but a lot of, a lot of kind of like what you would use data for right now is done just by kind of like tribal knowledge. Like, you know, I have a superintendent who's been on site for 35 years and he knows, like, every kind of project. Those people, like, in the next five or ten years are all going to age out of the construction workforce and then you're going to be left with, like, this dearth of not having that, like, institutional knowledge.

So to me, like, that is how you, that's biggest value of data, which is, like, all right, as you kind of, you know, as these people leave your workforce and, like, they can't, you can't go to them to be, like, what happened on this project 10 years ago and, like, why, you know, what was the cost and why is it two times more than all the others? I think that's where you see the real value is, like, it creates that transparency across the org so you don't, you are not bottlenecked by, like, individuals who just kind of have the entirety of your firm's knowledge in their head.

I think that's where you see the real value is, like, it creates that transparency across the org so you don't, you are not bottlenecked by, like, individuals who just kind of have the entirety of your firm's knowledge in their head.