Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
E5: Is generative AI good for scheduling?
68 Plays3 years ago
In this episode Carlos & Jason chat all things AI in construction, and where the potential is to really add value and save time. They also revisit how software is purchased across large projects, and how this is reminiscent of franchises.
Recommended
Transcript
Introduction to Offsite Podcast
00:00:00
Speaker
Hi everyone and welcome to the Offsite podcast where we chat all things construction, technology and everything in between. My name is Carlos and I spend most of my days talking to construction teams about how they deliver projects. And I'm Jason and I help build products that construction teams use to deliver their projects. So today we're going to be talking about OpenAI, a new release from NPLAN and we'll finish with a focus on how organizations purchase software.
00:00:28
Speaker
Hopefully, that last one's not as dry as it sounds. Super unique topic. We'll talk about AI. No one's really doing that at the moment. Yeah, it's not exactly on every news story, on every platform at the moment.
What makes OpenAI a leader in AI?
00:00:42
Speaker
Yeah, straight into open AI. Obviously, it's everywhere at the moment. More specifically, chat.
00:00:47
Speaker
GPT, which is the sort of leader in this space, which has been funded by Microsoft and a bunch of other organizations. So we've seen all the news with the sort of public interface where you can sort of ask questions and it gives you results. And it's a competitor for traditional search market like Google, but more specifically, we're seeing apps integrate with it.
00:01:10
Speaker
So we saw Slack have an integration now where it will do things like summarize a thread into actions or takeaway points. For the typical construction audience, you might need to say Slack is similar to like Microsoft Teams. Yeah, imagine like Teams and WhatsApp had a baby and it wasn't shit.
00:01:31
Speaker
That's probably how I get to that. So pretty cool because non-IIR companies can sort of tick the AI box and utilise what is crazy expensive sort of software and models to build. So everyone's not trying to build the sort of basics. But quite interesting when we start to think about how those sorts of applications could be built into construction or utilised in construction teams. We know construction's naturally cautious.
00:01:56
Speaker
with adopting tech, particularly things like this. But where could it go? I'd be interested to see your thoughts. I've got a few thoughts in the space.
OpenAI's GPT 3.5 Turbo and its implications
00:02:08
Speaker
And obviously, I think just recently, in the last handful of days or weeks, OpenAI announced a new endpoint for an API, which is GPT 3.5 Turbo or whatever it's called. And basically, the request cost has dropped like tenfold to use the service.
00:02:26
Speaker
With that cost being totally reasonable, the proliferation of applications kind of grows. I think in construction, obviously with any super-hypy thing, there is a proliferation of
00:02:43
Speaker
people trying to get on a bandwagon or try something out. And I see a lot of opportunity in the space, obviously lots of people trying in the space of applications, existing SaaS applications, software as a service applications, integrating some form of AI enabled features.
00:03:03
Speaker
And I think with so much hype around it, there'll be a lot of stuff that's not useful. But I really see, especially in construction, that any application or platform that is used by construction teams today to do data entry or data manipulation, if they, at some point in the future, a couple of years, 10 years, who knows how long,
00:03:33
Speaker
at some point, a number of those workflows will be AI enabled. And so if you're not experimenting with them now, at some point, it's going to become the norm. You know, if I think of a specific example cast,
Can AI revolutionize construction site diaries?
00:03:46
Speaker
like
00:03:46
Speaker
Let's say you're doing a site diary. So you have a platform where people capture site diaries at the moment, notoriously like takes a while to do. You've got to record every person that was on site, how many hours that they were there for, what was worked on. If you fast forward and you can easily imagine a situation where you just say, everyone was here except Bill and your hours are recorded. That might be buggy for now. That might not be a hundred percent right. But if you could get that right,
00:04:15
Speaker
90% of the time or some percentage of time, the amount of effort that that would save would be substantive for someone to be able to understand the context and then translate that into the application that you're using. So I guess to rewind, I see enormous opportunity in the space of existing applications and platforms
00:04:34
Speaker
utilizing AI to leverage or make the people using them more productive, less so the like pure AI play because that's not how my brain works. But yeah, entering data and being able to say or type something that understands contextually what you're trying to do, understands how the application that you're using works and translates that into actions that would have taken you 15 minutes or something to enter would be super powerful.
00:05:03
Speaker
The benefit of it being an existing application is also, if I think through like the user experience, if you were to make an application or a bot or something where you would talk into your phone and say, everyone was here except Bill, if it was wrong, correcting that records kind of tedious. Like if you've, you know, it's entered in a system somewhere, maybe you don't have the interface on your phone. Yeah.
00:05:27
Speaker
But if you, if you plug that sort of power into an existing application where you used to type that in and it shows you that, and then you can make some adjustments, you know, as it's, as it's being tested and getting better, it might pre-fill most of it might be 90% correct. You can make some adjustments where it's not, but you're still saving time. Yeah. I think the, um, the adjustments part is interesting because.
AI suggestions and liability concerns
00:05:54
Speaker
my mind immediately leapt to suggestive or advice type use of it. So we had a project here in the UK, it's probably about a year ago now, who effectively asked us to build something that would allow engineers breaking down P6 activities
00:06:13
Speaker
that it could suggest what the breakdown activity should be based on lots and lots of documents of how you should build concrete or whatever activity you're looking at. And that would be really tough because it was just their own data sets. It's not like these massive models that are accurate because they have a wealth of data. But if a Putma Q has QS hat on, if the advice is wrong and then something goes wrong on the project or there's a safety incident or there's something,
00:06:42
Speaker
you're in obviously a horrible space then of like who's liable. So I can imagine that being one of the key things that are going to be discussed over the next five to 10 years when these sort of things are looked at. But I think going back to your point around checking, you're going to need the checkpoint where it's a human going. Yet I'm actually confident that this is an answer I'm proposing that was suggested to be my system, not just system suggested answer. Answer is sort of executed immediately.
00:07:10
Speaker
Yeah, 100%. I think that's the power of the existing interface and having something like
What is the co-pilot concept in AI?
00:07:16
Speaker
this. We spoke last week about the idea of a co-pilot relationship and leveraging. You know, if we think about how our team works with regards to building, like developing a product and coding in a product like GitHub, the GitHub co-pilot feature from every one of our team that uses it is an enormous time saver. And it's wrong, like quite a lot.
00:07:38
Speaker
So the team starts typing. Google mail is wrong sometimes when it guesses what you want to say. Right.
00:07:45
Speaker
Yeah, correct. Or like, um, yeah, or you could imagine the guessing the next task that you want to build in a sequence of tasks, like the scenario that you were talking about, you might be wrong. But if you, if you, if we look at what developers would say that would use co-part, like, okay, it might, it might auto complete a function that I'm trying to write. It might be, it might not be totally right, but it might be like 80%. And it saves me naming everything.
00:08:13
Speaker
perfectly, like, gets rid of a bunch of typos, might save me a couple of minutes and cost me 20 seconds to correct it, something like that, you know? Yeah, and the grand scheme of things, you're saving time, right?
00:08:26
Speaker
which is the most important thing. The last point on this topic, I haven't read the API documentation. You'll be shocked today. Does OpenAI keep slash store all of the data that's presented to it? Because if we look at the context of sort of cyber chats with our customers,
00:08:44
Speaker
offshoring of data, the usual bits and pieces that we have to go through. Is that an immediate no with a tool like this? Because it's obviously, yeah, over in the US. They do store, I've not read it line by line, but I've listened to discussions on this topic. They obviously store some amount of it because that's training the model. That's helping them. Exactly. Yeah, it works. Because it grows, right?
00:09:07
Speaker
I don't think they're directly feeding that back in currently to the corpus of data that's trained on, but I think it is definitely a body of it. Yeah, there was an initial period of time it was like the training data, right? Yeah. So it didn't have anything like the last six months or something like this.
00:09:24
Speaker
A year or so, something like that from memory. But it's not just the language. There's examples of products that will, let's say you took a, even if you put your QS hat on and play with spreadsheets again, there are models that will understand the rules of a spreadsheet and you could say,
00:09:45
Speaker
summarize this data showing month by month of this subcontractor costs broken down by WBS over the next 12 months and it will output that. So data entry, anything that's data entry or data manipulation, I think both of those areas in going back to your initial question about construction, I see massive potential
00:10:08
Speaker
to, I guess, supercharge existing platforms with the AI. That's like the first wave that I can envisage.
MPlan's new tool versus market alternatives
00:10:17
Speaker
Yeah, yeah. Right. Moving on then, straight into an AI platform that we know called mPlan. So they have just released a tool which is, and we're on a topic there, which is probably the least experience I have on any topic we've spoken about so far, which is
00:10:32
Speaker
integrity or schedule integrity checks so it's a tool that will check out your program and go right this is sort of a an actionable area of your program where you could have bad logic or missing logic or it tries to look at how the makeup of your schedule is sort of yeah working so
00:10:53
Speaker
Don't pass to me, keep going. I'm trying to think of the right words to do this as a complete amateur. The quality of your schedule. At this point now, all planners have turned off anyway.
00:11:05
Speaker
No, you can say whatever you want. Yeah. So it does these checks and they've released it as a free tool. So there's lots of tools out there. It seems from a quick Google that you pay for this service. Oracle do have a tool that does this. I'm going to guess it's expensive and maybe it's not as good. It's quite hard to see Oracle documentation without being a user that you have to sign up for the latest yesterday.
00:11:25
Speaker
So, yeah, keen to understand, Jason, I guess maybe you could give a more accurate intro as to what it is. But from what I was seeing, I was thinking, obviously it's great that people can do this for free and not pay for it. If it's a quick check that you just pull your XER out, pile out a P6, for example, upload it and do that. But obviously we know Enplan is built on, I think they have 500,000 schedules. So is it them trying to increase that pool of schedules that then improves their actual core offering? So it's a bit of a tech ticket play.
00:11:59
Speaker
The, um, I guess for background and first preference by saying I'm not a planner, but for context, there's a, there is a thing that is called a schedule and integrity check. There is a, there was like a standard check, which is called a DCMA 14 point assessment. And basically it will, the idea is that you're checking whether you, you're checking the robustness of the schedule. So you're checking whether you have, you know, lags that are holding two tasks apart tasks without
00:12:20
Speaker
Yeah. Jason, how would you describe it?
00:12:29
Speaker
No logic, right? So yeah. The idea being that like the general premises premise is, okay, your schedule looks right today, but how robust is it? If things change, you know, that you don't end up with weird outcomes, you know, if you stick a big lag in somewhere to keep things, or you don't have a predecessor and the plan changes, then yeah. My meltdown then was missing the word robust. I think that summarizes it quite well. Okay.
00:12:56
Speaker
So there's an enormous rabbit hole to go down about how effective this check is in general, but they're often required under construction contracts to perform them, whether or not they have some value and oftentimes the schedules will be required to check them.
00:13:16
Speaker
If you think about, there are, like you said, other tools on the market that do this schedule assessment. And like if the most common that I'm aware of is a, is a product called Acumen Fuse, I think, uh, from memory, it's like if we're talking about Australian dollars, something between five to 10 K a license per annum. Wow.
00:13:34
Speaker
So it's an expensive piece of software from people that I know that have had a go at the, or use the free tool that mPlan have provided. You know, it's, it's definitely not as mature as what you're getting when you purchase that, but if their goal is to keep evolving it to the point that it does, you know, meet the requirement that teams have to do this check, it's.
00:13:57
Speaker
I think it's great for construction projects to not have to purchase another piece of software.
Software as a loss leader strategy
00:14:02
Speaker
You know, from an MPaN perspective, it's probably costing them nothing to provide the service relatively. Yeah. Massive tangent on this, but whenever I go to Project Contrast Expo, every slightly older guy seems to have been an ex-employee of Acumen Pews. Sure, it's not the same person wearing a different disguise every time you run into him.
00:14:28
Speaker
What's really interesting to me, sorry to cut you off, is if you think about like as a software company, ballpark, I think something like the service from M plan and we'll probably get corrected if I'm wrong. You can probably expect to spend probably 150 K or something or something in that range for a project.
00:14:46
Speaker
So if you're trying to build a software company, one of the things you're trying to do is generate leads of the people that would purchase your product. And when you try and do that, you'll suddenly realize that, say you're selling scheduling software, you'll get Sally, the hairdresser from Wisconsin, sign up because she was trying to get a piece of software to schedule hairdressing appointments. You'll get rubbish leads, basically. The brilliant thing about
00:15:13
Speaker
This tool is obviously anyone that's looking to do a schedule integrity check is pretty much the target customer for mplan. So it's a great way to, you know, kind of do like a loss leader and get people familiar with the platform, get them enrolled in the platform, and then start to show them the, the, the additional benefits of the AI offering that is the core product from mplan. So I think it's a great idea for them.
00:15:39
Speaker
Yeah, it must be it must be a good product because you really are in a bad spot if you give away a free tool, it's shit. And then everyone judges the rest of the company on the thing that they used for free. So they must be backing it and actually being a really good free tool too. Because you're sort of that's your first impression right of a whole bunch of leads. So yeah,
00:16:04
Speaker
Yeah, it should be decent. Cool. So, onto the third topic, really hard to get a segue here, but software, I guess is the one.
Challenges in construction software purchasing
00:16:14
Speaker
I'm super interested to ask you some questions on this because one, this is like part of what you do every day. Two, you've also got your X quantity surveyor hat on here. The way that construction teams purchase software, do you think that it's fundamentally like, what would be, what's different about construction?
00:16:33
Speaker
Do you think this would be similar to how non-construction businesses purchase? That's a good question. So selling software to projects, which is like if you compare it to a non-construction company, I can imagine, I haven't got direct experience of it, it would be unusual or uncommon to sell software to teams within a company.
00:16:58
Speaker
I'd imagine if you go finance, tech, industries that are pretty far from construction, it's always centralised and it's sort of a business need that drives a procurement process which then deploys software and rolls out across the company.
00:17:11
Speaker
So I think the first option is probably quite unique to construction and that's I guess partially because construction projects, especially large ones, are almost set up like their own companies. It's not just a team delivering a bit of scope that's reporting back to head office. Like you really are in your own little world for
00:17:30
Speaker
a number of years. When I was on a construction project, one of the things that we always used to get told off for was using purchase orders for everything when the terms of service are really not applicable to purchasing a lot of things. We used to try and do a purchase order for everything. You need to hire a consultant to a purchase order. We used to do a purchase order to hire our five-a-side football pitch every Wednesday. Yeah, exactly.
00:17:58
Speaker
So there's a whole infrastructure in construction companies around purchasing that is really tied to this thing that is maybe not super fit for purpose. I don't know if it is totally unique to construction. But a lot of the rest of the economy has credit cards. A lot of the rest of the economy has direct debits.
00:18:24
Speaker
Obviously, when sums get higher, the credit card becomes quite an expensive option to use. And obviously, there's limits and all sorts of things. You mentioned before that projects sort of become like their own company. We used to talk about it when we were on a project that we were kind of like the franchise of construction company A in this town or working on this project.
00:18:49
Speaker
The head office was there and you sort of wore their colors, but at the end of the day, you were responsible for the profit and loss of your franchise. The project being like its own company or franchise, does that normally translate to its having its own bank account? Yeah. So projects will have their own bank account. They'll use sort of enterprise systems to do their sort of accounting, which is within the rules set or governed by one of the chosen JV partners or the contractor on a project.
00:19:17
Speaker
they'll have a project bank account, they will have a project card, albeit the rules are pretty strict on the card should only be for bike.
00:19:25
Speaker
We hit a milestone and they do something social for the team. Things that shouldn't really be on patch orders. Just to interrupt, do you think that's because historically the contractors haven't had a way to capture the approval flow for charges on those cards? I'm not sure on that. I know one of the fundamental reasons for having a project bank account is each month the contractor asks the client for a pot of money and they go, right, this is what the money's for. And they give them a breakdown.
00:19:55
Speaker
they then pay their supply chain from that money. So to stop the contractor trying to get more money than they're actually going to pay out, they monitor the bank account to make sure it's emptying each month too. So they're not sitting there trying to build up a bank of cash or interest and things like that. So the bank account is monitored by the client organisation.
00:20:17
Speaker
So it's literally cash in, cash out each month and there shouldn't be this lot left. But I mean like the credit card thing versus a purchase order. Say you are going to purchase a, I don't know, what's the scenario? Say you're going to put a barbecue on or something.
00:20:34
Speaker
Yeah. You might put that on a credit card, but you were going to spend two grand on a piece of software or something for a couple of people on a project and that will go through a purchase order. Is that purchase order the preferred way historically do you think because the card transactions didn't capture an approval flow?
00:20:57
Speaker
Because a lot of projects, you know, a lot of projects will have a matrix. It's like this person has an authority up to this and up to this and up to it. On one of my last projects, it was petty cash or minor purchasing under 500 quid.
00:21:14
Speaker
So your P card could be anything from a medkit that maybe was used, they need to replace to, as you say, something social. And that will just get done. The accountant would then produce a report at the end of the line saying, this is our petty purchases. As soon as it goes to a grand or two grand purchase order system, it understands the full, it's like an automated set of approvals, who it is on the job.
00:21:39
Speaker
the order and everything else. So there is the control measure rather than like had some other card and then deal with it later like we would do when we do our accounting at the end of month and zero. Because like obviously I wonder whether that does change over time with you know with expense management type because if you go to other yeah like if you think about what we do we get we you know someone can try and purchase something and then there's an approval process for that transaction.
00:22:07
Speaker
Yeah, 100%. They're using legacy systems.
The scrutiny over disallowed costs in projects
00:22:10
Speaker
A lot of contractors have their own accounting software that is originally SAP or something like that, but it's pretty old school. It's not like they're using zero and they're hooking in wise and everything else that we do. I think another... Oh, yeah. Go ahead. No, no, go ahead.
00:22:29
Speaker
Another thing that's quite important with the way software is purchased, and I try not to go too heavy and secure as fuck because they'll have a few engineers dropping off, but if they're this thing. There's no planners left by now, so. That's already alienate the rest, right? In construction, there's this massive thing which is disallowed cost. So every pound that you spend on a project, the client will go
00:22:53
Speaker
Is that a pound that should have been spent on the project? If you shouldn't have spent it, they don't give you the money. So that's a loss if you think about it. If it's not contributing to the scope, it's outside of scope, they're not paying for it. And software is a massive gray area because the clients always have this clause. It might have changed, but it won't have changed much, which is it has to be project specific costs.
00:23:15
Speaker
So the first thing they do is get these blanket charges from head office which could be for like Microsoft and all the usual suspects and they'll just say we're just allowing all of that because you can't prove that was specifically used on this project. So it kind of backs up the project-based purchase order because it's quite clearly for this project so that's going to be fine. The enterprise customers that backcharge projects that's why they always ask us for detail because
00:23:42
Speaker
the projects then want to recover that cost if it's in an overhead obviously that's kind of dealt with separately so there's this real drive for contractors to gather the detail they need to have the fight effectively because if you can't prove it was a person that's specifically on this job and that's their license for software they get it disallowed and that comes up your bottom line.
Cost management through vendor reduction
00:24:07
Speaker
That makes it kind of unique because I think if you look at how other industries would purchase software, you would have something like a card or a direct debit at the bottom level, you know, if you've got a team testing something or in this case, like a project.
00:24:25
Speaker
And at that point, your subscription might be quite variable based on how much you use it or how many people depending on what the mechanism of like the measure is. But once you get up to a certain level where it's a tool that the company uses, oftentimes what the companies are looking for are guaranteed costs.
00:24:42
Speaker
fixed for a period, you know? So if you've got like a, I don't know, this is a bad example and I, you know, we don't spend enough to get this yet on like Google cloud, but you'd go from pay as you go model to a fixed cost contract. Does that model, you know, that model from a company perspective is probably beneficial from the fact that you can negotiate a fixed, you understand what your cost profile is and you can, and you can, um,
00:25:09
Speaker
And you can make decisions around it and you can probably negotiate quite well with that approach. But does that cause problems or violate the disallowed cost rule? How do companies deal with that?
00:25:21
Speaker
they'll still, from a QS point of view, they'll still take the total cost that month, look at the actual usage and do a quick back calculation to work out approximate cost. I've had that conversation in the past. Some are fine. Some say, oh, would you mind just signing this document to say my calculation was right? Obviously, that's a weird place that you don't really want to be as the vendor. So
00:25:44
Speaker
Yeah, they'll still back calculate it to work it out if their intention is to charge projects, of which most do intend to charge projects. But as long as they have access to a user list and they can see which projects they're on, it's pretty easy to do that retrospectively.
00:25:59
Speaker
How does it work? Obviously, we're talking about how with your talking from experience from our software where we have a seat-based pricing model. If you were using a piece of software that wasn't seat-based, so let's go back to the AI example. AI probably isn't a seat-based model. If you had projects across a business making different calls to an API to get predictions or results or analysis,
00:26:25
Speaker
What would you do as the colony surveyor then on a project to prove the cost? Because you don't want to go through an API call log over a month and try and work out which one belongs to which project. Yeah, I think tools like that will just end up in that project overhead. So every year, the company will reevaluate what their fee percentage should be. And then within that fee, there is a whole list of stuff that they're including from HR to
00:26:54
Speaker
the lease on the head office building all the way down to specific pieces of software. You'd bet there's always a bit of buffer in there anyway to account for bits and pieces that are being spent so they're not in a loss making position. So that's why they'll constantly try and rationalise tools. We know a lot of contractors, they're getting asked to sort of
00:27:14
Speaker
rattle down 60 software vendors to 20 and sort of back their horse rather than allow everyone to purchase what they want because the cost gets out of hand. Interesting. It does feel like observing from the outside that there's a way of purchasing and accounting and managing costs across projects and companies that might be a little bit behind where a lot of the rest of the economy might be.
Conclusion: Improving construction software procurement
00:27:42
Speaker
Big time.
00:27:44
Speaker
Just so many hurdles, right? Whether it's commercial, IT, the project, you've got BIM teams, lean teams, everyone has a view, everyone has a process, everyone has their own checklist of things they need. There's data, there's data rules from the contractor, from the client, from us on how we actually want to function as a business. So yeah, trying to pull all that together is not easy. What a note to finish on.
00:28:08
Speaker
Yeah, well, a positive way to end a podcast. As we just head off into oblivion. Yeah, right. That's all we've got time for today, everyone. As always, thank you very much for listening.