Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
E22: nPlan CEO Dev Amratia on AI and construction
72 Plays2 years ago
In this episode, Jason & Carlos are joined by Dev Amratia, CEO of nPlan.
In the first of three episodes focusing on AI and Scheduling, the guys discuss cutting through the noise in the AI space and prioritising action. They also dive into AI schedule analysis, who benefits, and where it can go.
Recommended
Transcript
Introduction to Episode 22 and Guest Dev from mPlan
00:00:00
Speaker
Hello and welcome to Episode 22 of the Offsite Podcast where we chat all things construction and technology. My name's Carlos Cabole. And I'm Jason Lantini. G'day Carlos. I'm on time and prepared as usual. Are you?
00:00:15
Speaker
That's what works for everyone else. I'm usually five minutes late and have done close to no prep and Carlos is raring a ready to go. So I'm on brand. It's fine. If we break routine, it'll feel weird each week. So we might as well continue. Yeah, that's it. We'll jinx it. Absolutely.
Dev's Career Journey: Aerospace to AI Advisor
00:00:33
Speaker
Right, so over the next few episodes, we've got a bit of a theme around AI and scheduling, and we've got a few guests lined up over sort of three episodes. And we thought we'd kick off this sort of trio of episodes to talk to someone that's on the leading edge of AI and scheduling. He's a well known face for the industry and a huge welcome to Dev CEO of mPlan. How are you doing, mate? I'm good, guys. Thanks for having me here.
00:01:00
Speaker
No problem at all. I really like the air neon sign. I'm going to get an AFIX one made ASAP for these podcasts. So we all offered to start with a bit of background around our guests, but I think I might just point the question towards you. How did you start at aerospace engineering transition through Shell and end up founding and plan?
00:01:21
Speaker
You know, as any good graduate does, just go and take the job with the highest salary you can possibly get, even if you can't do it. So that's what I did. Jordan Shell worked there for nine years. Project engineer and a project manager for most of that, right? Delivering large capital projects on the pointy end in some really awful parts of the world.
00:01:41
Speaker
So it's all that problem there. I did take a really weird sidestep in my career after being a project manager for nine years, which was I was a special advisor to Theresa May, the prime minister of the UK for 18 months, long months.
00:01:55
Speaker
And in that time I wrote the UK's national strategy on AI. So that was kind of like my weird bridge between being out in the field managing large projects and then starting and then running an AI company.
Transition from Government to AI Entrepreneurship
00:02:09
Speaker
Can I press pause then, Dev? How did you get, how did you get, how did you become a special advisor to Theresa May writing an AI strategy?
00:02:18
Speaker
It kind of fell into it. The last thing I did at Shell where I worked on the acquisition of PG Group, basically I got to work with the CEO, the CFO of the chairman of the group. And then one of them was like, Hey, what are you doing next in your career? And I was like, I don't know. And I'm like, have you ever thought about working for the UK government? I was like, yes. Nice.
00:02:43
Speaker
been my dream all along. I remember introductions were made and I just showed up on Theresa May's third day in the office. Writing the government.
00:02:58
Speaker
Writing that sort of document and then founding an AR company feels like the venture equivalent of
The Role of AI in Construction Scheduling
00:03:02
Speaker
insider trading to it. I wish it was that limited. Sadly not. I think all it really gave me was perspective. I was just like, I can cut through bullshit pretty damn quickly now. Remarkably, 60 of them, it's still a lot of bullshit in the world.
00:03:22
Speaker
in the AI space, that is. It just helped me navigate the space really quickly. When founding a company, clock speed is so important. You have to move really fast by figuring out what you're going to do and how you're going to do it, and then just do it. So I don't know. Call that information arbitrage. But yeah, it certainly wasn't financial arbitrage.
00:03:45
Speaker
I wish the current government would get it cutting through the bullshit. Right, so to get, I guess, a little stuck in our audience is typically project managers and engineers. What does AI schedule analysis do for them?
00:04:02
Speaker
That's what it says in the gen cam. It will analyze any schedule to be able to answer three key questions about any project and any schedule. How long is this project most likely to take? What is the most probable outcome? Question one. Two. What might go wrong with this project? What opportunities do I have to improve this project? Question two and question three.
00:04:26
Speaker
What do I do about all of this? Prioritization, help me come up with strategies, scenarios, plan through the maze of project delivery, becomes challenge three. And all of that is driven through a system that can run the analysis path automatically.
Positioning AI Tools in the Construction Industry
00:04:42
Speaker
Previously, people used to do this analysis, known as a quantum schedule risk analysis.
00:04:47
Speaker
QSRA for short, run these long workshops. Lots and lots of guesswork goes into that, right? I've been passing these workshops before and I'm like, I don't like that answer. Maybe we should change our guess. It's not only just, not only is it incredibly time-consuming and cumbersome, it's let out not correct. You just don't get valid answers through that process. There's tons of research that can show that today.
00:05:13
Speaker
And so like, David, if I can jump in, I guess, uh, one of the things that I say internally in our team, like a time is construction project managers and teams are like drowning in this like sea of softwares being like trying to be like foisted on them, uh, with names that they can't remember. And like logos that are all slightly different versions of a blue word that they won't remember either. And positioning is no effort to me on purple. So.
00:05:39
Speaker
They've just pointed to his logo at the back. We also have a blue logo, so we're also guilty. But like, so positioning is super important. Nowadays post AI hype, I guess there's a bunch of software out there claiming they've now suddenly become not just construction software, but construction software plus AI. And you guys are like the OG of construction AI.
00:06:03
Speaker
I guess, how do you place or position and plan and what you do as well as possible so that it has cut through for like a project manager or someone that's going to make a decision to use your software?
Empowering Project Managers with mPlan's Data Insights
00:06:15
Speaker
How do you position it? Is it replacing something? Is it adding to them? Does it open up new capabilities? What's the thing that sort of gets the cut through? The most frequent cut through we get, we cut out bullshit. The more formal word of that is called assurance.
00:06:29
Speaker
So algorithmically led assurance is the primary one. If it's a project manager we're speaking to or commercial manager or above, we're basically like, listen, we can give you more confidence about the outcome of your project than you will possibly ever get from any other source, period. That seems to work quite well. We do have different
00:06:50
Speaker
When we're talking to planners, when we're talking to project controls, the message is very different. It's much more of a message of empowerment. You guys spend hundreds of hours trying to come up with these things, and you just get shot down by project managers that don't like the answer that you're providing them. We can give you a tool that will back it up. Your answers now are backed up with 750,000 projects of data in the back.
00:07:20
Speaker
It's no longer Dev's opinion or Jason's opinion or the project controls team's opinion versus, you know, everyone else. It's like, well, this is backed with data. That's, that's a story of empowerment. Um, and that seems to work very well with the controls team and the planning team that we work with as well. So not, so not something like the classic advice is, uh, it's always better to sell to pain. Um, but you find that like this positive story and empowerment and confidence was the word you use for project managers. That's like the.
00:07:50
Speaker
Yeah, there is a serious pain in doing a QSRA and then it gets ignored. I don't think anyone likes doing that or enjoys the feeling when that happens. So we all think to that pain there. It's the pain of insufficient empowerment, right? And say that we could give you the tool to get you out of the basement and onto the top floor.
00:08:11
Speaker
One of the, um, one of the things that comes up a lot when it's amazing how many teams that I talk to, or we talk to, and everyone's interested in, what can AI do for them? And people are really talking about like thinking about it at the moment on the, I guess, from our background and our exposures on the contractor side, one of the things that people are trying to wrap their head around is like concrete examples of, you know, what, how would I use this here? How would I use this there? So like, if we were to, I don't know.
00:08:38
Speaker
If I can just throw us sort of scenario, I'm on, I'm on up, you know, an infrastructure project, project manager for the contract or the client you choose, whichever is the better, the better fit. Like what's like one or two concrete things that I would use or do like the, the, the, the analysis would do or give me if we were to really boil it down for like a, I guess someone on the ground.
00:09:01
Speaker
So let's take the example, you're in project controls on the contractor side, big infrastructure project, call it early days of execution. You're out in the field, you've got people out there doing stuff. You're far enough into the project, but
00:09:18
Speaker
Things are not going to the plan that you originally came up with. But it's already a year old now. At that point, so what Mplan will be doing is we'll be analyzing the month's progress updates as they come through. You're making progress. You're making changes on the schedule. All that stuff is, I call it business as usual. And we are going to be able to tell you what in the future might go wrong next and the quantitative impact of that. So it's sort of like a ranked, a continuous ranked list
00:09:48
Speaker
of the key things you need to think about in advance of time. We call it snowball modeling. So if you've had the analogy of the snowball top of the mountain in construction, that is, you know, if your rig operator, if your rig, if your piling rigs are delayed by one day, every project manager, every project manager I know will be like, eh, fine, we'll catch up, right? Maybe you will. Maybe you won't. When you don't,
00:10:12
Speaker
The effect that that leads to is that the next thing after the rigs come in will be delayed by five days. The next thing after that will get delayed by 10, 20, 30. And then the snowball keeps getting bigger and bigger and bigger. And by a certain point, you have lost control of your project. We can tell you where the snowballs will start. So that one-day delay, you become pedantic on it. You're just like, we cannot let that even little bit slip there.
00:10:37
Speaker
And equally, it's like, well, we know that there's this wall coming up in front of us, or this giant fire is going to light up six months into the future from now. It is way cheaper and less stressful without putting implement it, putting plans in place now to mitigate that risk than it is to like discover that that risk is eventualized. And then now you're all in a frenzy to try and resolve it.
Preemptive Solutions with AI in Project Management
00:11:02
Speaker
I put my project manager hat on and I think about similar scenarios. It might be similar in the past where early on in a project, things aren't going well, pressure on the critical path. And we start inventing strategies to work around the pressure on the critical path. And I imagine when we used to do that, there's a lot of
00:11:25
Speaker
There's a lot of onus and responsibility put on the planning team to make those adjustments in a way that is not totally making mistakes and missing things. And does having something like end plan give me more confidence in the adjustments that we're making by like, I don't want to use the word big brother, but I guess double checking in any way the changes that we're making that we're not adding risks to the, to the, to our sort of on the fly. Uh, here's the new critical path.
00:11:54
Speaker
Correct. Very much. Um, there's the two roles we'll play. Like, so one is you could use a big brother, right? Like in a nice way, not the creepy spy way. Yeah. Friendly big brother. Everyone likes the big brother. It's like, Hey, Hey buddy, I'll look after you. I've got all this experience in my AI. I think we found your new tagline. It's like the friendly big brother. All the space. Um, the.
00:12:24
Speaker
It's about one of it. The other
Leveraging Emerging AI Technologies in Construction
00:12:26
Speaker
is just suggestive, right? It's like the project team will use the word earlier that's like the project team have to kind of invent their way through mitigation, right? And, you know, I did ex project manager myself as well. There's a lot of like bullshit in that. You just kind of like wing it. You're like, this kind of feels right. Let's just do that now.
00:12:47
Speaker
You're like, you know, this is a multi-million dollar decision we're taking here, right? And it's like, well, does anyone have any better ideas? You look like the highest paid person here, so we'll just take that. We'll just take what you say as gospel. And so what Enplan does is that it will see what you've done. It'll know what other projects in the past have done and say, hey,
00:13:10
Speaker
I can foresee that if you go down this pathway, this is where you might end up. Here are other pathways you could follow. What it won't be able to do is tell you which one is the best one. It's sort of just giving you options, right? And say, here are the options that thousands of other projects have gone down the path off and the results they got to.
00:13:29
Speaker
You pick. Your risk contingency part might not be large enough. You might not have the right relationships. You might have a supply constraint that Encline doesn't know about. We're just saying like, hey, others have done it this way. You want to take it big from there. Yeah. So to double click on that, because the question I was really wanting to get to
00:13:48
Speaker
after we went to the example was like, there's a ton of, in AI, in all construction software, there's a ton of hyperbole. People have lovely websites, they say what they do. And like, there's a, there's a delta between like the hyper reality. So like, yeah, you were quite upfront on what it won't do there. Let's go into that scenario. I'm on a project. I've got pressure on my critical path. I thinking if I do, instead of going down a path through
00:14:15
Speaker
uh building the structure and then i don't know now i'm i don't want to expose the project that i'm thinking about in my head if i'm very if i'm too specific but if
Addressing Conflicts and Collaboration in AI Analysis
00:14:26
Speaker
i was to change my path and uh and go at an alternative way that into like one of my second or third critical paths will push stuff off the critical path end plan might suggest it what do you mean by end time will suggest other routes and what does that actually look like if i'm
00:14:42
Speaker
on the ground if I'm the planner or the project controls thing. We call it the driving path. It does play on the word critical path. What that means, I can't finish it right now.
00:14:59
Speaker
You use the hands of the mechanism to illustrate. Imagine you've got multiple critical paths and each one has a certain likelihood to impinge at various points throughout the network of execution delivery. We can forecast what the most likely, how risk propagates through the network. So we're basically going to be able to say, we know that the way this structure here, the way that it's been planned to be executed,
00:15:27
Speaker
has this likelihood of being delayed based on all the data we've seen in the past and the context we've gained from understanding this project. And we do that for every single activity in the entire network. And then what that creates is this thing called the driving path. It's an interactive tool that allows project managers and planners to interact with how will risk propagate through the network. If I'm going to move off the critical path and now start focusing on the fourth order underneath it, this is what might happen if I do that.
00:15:54
Speaker
You can play out what-if scenarios in there. You can also just play out what was I missing inside my own thoughts, right? Like, I thought that the substructure would be the super hard thing and, you know, we need to get our best people on that one. Does anyone think the same thing? Or, you know, maybe it's the electrical systems that actually are going to come and bite me in the butt later on. And it's not even on my radar today. So doing that in a very intuitive way is what the heart of our product does.
00:16:20
Speaker
And so if I'm, if I'm to dumb it down to my, well, let's say for others, but mostly for me, can I think of the driving part as like a risk weighted critical path? Yes. Cool. Got it.
00:16:36
Speaker
Cool. Does this create a difficult relationship or I guess open a can of worms when it comes to a contractor submitting a
Staying Ahead in the AI Landscape
00:16:48
Speaker
schedule? They'll do their own analysis, but then does the client take the schedule and then do use a tool like Endplan and then basically come back with this endless list of reasons why the schedule isn't good enough because Endplan says you can do all of this stuff with them.
00:17:02
Speaker
And then you end up in this back and forth where we now have so much information, it's hard to get to an end point. And as you say, it might suggest lots of things. And then you're in this battle around, like almost trying to reason or argue with a suggestion from a tool, trying to get that agreement on the program or sign up on the program becomes quite tough. Or do clients not use tools on someone else's schedule because that's, I don't know, not right.
00:17:30
Speaker
No, no, they do. So there are two scenarios here, right? And mPlan, our split is roughly 50-50, so we have half our clients that own our operators and half our clients are contractors. So let's start with the contractors. So if a contractor is using mPlan, we've seen it with Kia here in the UK.
00:17:55
Speaker
where they're using Enplan a couple of projects. The one I'm allowed to talk about is the Oxford Station Redevelopment by Network Rail. They're using Enplan to give the client assurances that here are not just marketing their own homework, which would typically be the norm. You'll just say, here client, here's my updated schedule and here's my updated forecast. And the client's looking at it like, really?
00:18:21
Speaker
Is that actually where you're going? You know, what are you not telling me inside this? And then so this cycle of mistrust is kicked off, right? And you end up with the back and forth and it's, it's just, it's just a crappy situation to be in. I used to be on the client side, you know, trying to like, contest what the contractor was telling me. And the best I had really was like,
00:18:39
Speaker
my intuition that the contractor's trying to hide information from me and make my job to go and dig it out by hook or by crook. Really crappy environment to work in, by the way, right? And that sends itself to, so what they're doing is they're saying, hey, here, hey, network rail, we've done this analysis. We've done it through this method. Here's what the analysis was saying. We're giving you the top bits and we're telling you what mitigations we're putting in place against those things.
00:19:03
Speaker
So, you know, in a way that kind of reduces the assurance or the review burden that's going on in the project, but doesn't go away, just reduces it. If a client's using it, yes, we have seen circumstances where it does get a little confrontational, right? Because the client's saying, I now have access to buckets of information. And then you, you know, who has more information starts becoming
00:19:29
Speaker
a source of power, we do not endorse that in any shape or form. So what we suggest to our owner operator, Ren, is it is healthy to challenge your fund factor, right? But go to them with, you know, the five things we have found
00:19:46
Speaker
that have the highest impact that if you were to find a mitigation strategy to them would give you the best chance of delivering this project to time and budget. Bonus points if there's some contractual incentives behind that as well. The best-in-class systems, the best-in-class frameworks where the owner is using AMPLAN is inside alliances, because the opportunity for contractual conflict is
00:20:10
Speaker
a significantly higher than, um, you know, the traditional mixed price lump sum. Cool. I had a, I did have another question. I was, I, um, I was quickly trying to trademark the, uh, uh, friendly big brother.com domain.
00:20:25
Speaker
So yeah, so the world of AI tools, strategies, techniques, capabilities is expanding rapidly. Some of the strategies that exist now all have emerged as like successful and powerful probably were nation or somebody not even existed when you were getting going. How do you internally keep up with
00:20:50
Speaker
You know, is it this LLM? Do we use an LLM? Like, how do you keep up with the pace and change? Are you just experimenting with everything that comes on the block? How do you internally keep up? Great question. We do the least startup like thing you could imagine, which is we have a dedicated
00:21:07
Speaker
five-person research team whose job is only to do this. Not only do they just keep a horizon scan and see what's out there. I mean, I call that less than 10% of their job. 90 plus percent of their job is to push the boundaries of what is even technically possible today in the research world.
00:21:32
Speaker
You're right, LLMs have suddenly become very much in vogue, and we at MLM do use LLMs. That part was like the 10% part, right? Like, how do you find out what we could do with existing technology? Cool. Great. Dedicated resource to do that, by the way.
Demonstrating ROI with AI in Project Management
00:21:49
Speaker
The 90% of the effort, though, is like, well, what's even further than this? Because if we're thinking about how to use LLMs, all we're doing is catching up. But what's beyond all of this stuff?
00:22:01
Speaker
What is the world not even seeing right now? So in our world, what we call that is large graph models. That means if you think about a schedule file, it is a beautiful source of information where it contains words, language, large language models, LLM, it's basically just text. Schedules contain text, they contain time, and they contain relationships.
00:22:24
Speaker
And then if you want to rake it really complex, you have the monthly progress updates, you have like grass evolutionary structures as well, like the grass morph as projects go on. We've been spending a ton of time and money learning that.
00:22:41
Speaker
so that we can generate graphs. That means they've used Gmail before, or you've seen autocomplete on Gmail. You know, it kind of knows what your next sentence is going to be. You just press tab and it starts filling in, right? It's not perfect, right? Never used perfect. Imagine that for scheduling.
00:22:59
Speaker
you're typing in and it's like oh I kind of know what you're up to here and just starts forwarding you know ahead of time of the graph system but there's all sorts of applications of that things like today it is nearly impossible to truly detect cupidity inside a schedule so what would that mean if stupidity would be
00:23:20
Speaker
when someone suggests that you can put the roof before the walls go up, right? Like you could put a finished start relationship on that and like Acumen Fields would be like, yay, tips on you. You know, that's a dumb thing to try and build that. Or if you can figure out how to do it, then you need a raise. But the these types of technologies can then start like saying, hey, you know, you forgot the scope to do this. You've missed out on
00:23:49
Speaker
The better one, I guess the more practical thing that we've seen is missing scope, right? Where someone will be optimistic, like, oh, let me take a five-day review from this person and then we're good. We can look at the green light to go into the field. It's like, actually, well, no, you're going to have to do that. Then you're going to send the document to this other person and this other person over there has to stamp it and then it's going to come back. And then you might go into the field. And that's, those are classic schedule misses.
00:24:18
Speaker
Um, but they're not someone's fault. It's just like really hard to like be able to figure these things out when you're on a project team with limited experiences on your hand. Let's call it like the next frontier of research and how we try and capture this. Yeah, cool. Carlos, back to you, man. When I guess organizations, um, you'll get through a procurement exercise, you produce some sort of business case, you're going to be focused on return on investment. There's the obvious like time saved for planners or project controls leads compared from
00:24:47
Speaker
something manually to do something with anything. I can imagine ROI is broader than that. It gets a bit more wild because of like end dates, for example, and everything else that comes with it. How do you approach that and how do companies sort of digest that? What is the approach?
Overcoming Skepticism and Objections to AI
00:25:04
Speaker
The truth is it is hard to truly quantify the full ROI of using mplan. And the reason it's difficult is we are probabilistic future estimators. We're saying everything we do is a probable event in the future. So if it does or does not happen, you can't arbitrate it. You can't say, let's go build two projects, one with mplan, one without mplan.
00:25:28
Speaker
You just can't do these things. So, without trying to make a science project of ROI calculators, what we do is we typically, with new organizations we work with, we do these things called backtests, which is like, let's play a game that you actually purchased an enzyme license eight years ago.
00:25:50
Speaker
will train our AI models on 80% of the project that you have in your organization and you hide 20% away from us, 20% goes into the test and will pretend like it's 2016 for 500, 1000 projects, right? And we forecast and we just say like, look,
00:26:12
Speaker
had you bought us back then, we would have told you all these things about the projects. What actually happened to those projects? Did the things that we thought we would have told you, did they eventually? Did the projects get
00:26:23
Speaker
Do they go early? Do they run on time? Were they late? How good are these forecasts? What would you have done had you had this information back then? That kind of creates this retrospective ROI. It's a very strong indicator. It shows you that the thing is not just randomly guessing. It also feels like an A-B test with conventional QSRA.
00:26:44
Speaker
Yeah, yeah, yeah, it does. Yeah, correct. Very much. It's also a handy way to get hold of thousands of schedules pre-purchase because you get to build that bank right by doing this test each time with that. Oh, you're such an evil, you're such an evil person. That's a good thing. That improves the model.
00:27:05
Speaker
Yeah. And with every organization we work with, the model gets better and better. The services get better and better. The forecasting gets better. So yeah. We like to think of it as everyone winning.
00:27:15
Speaker
There's one last quick question for me, because I'm conscious of time. But I guess one of the things that, going back to the conversation that I've had with people that relate to your product and AI generally, what are the typical common objections that you have to face and overcome when dealing with whether it be a contractor or a client?
Reflecting on AI's Transformative Potential in Construction
00:27:38
Speaker
There are two. The first and most common by far is
00:27:43
Speaker
I don't believe it can ever work. How do you know the soil conditions of my site? Like, how do you know the mood of my superintendent? Like, I mean, these things matter, right? Like, they, projects get food on, on the back of these things. How do you, how does your AI system know all these things?
00:28:09
Speaker
The short answer to that is we don't. We don't know these things explicitly. All we know how to do is find patterns amongst projects that might seem or feel similar to yours. It's how the human brain operates, right? Like if you've seen 10 building projects come together and then you get put on to the next building project, you will use those experiences to bring. So if you've seen five building projects and five
00:28:36
Speaker
infrastructure projects and you get put on to building number six, you use your building projects experiences to govern
00:28:44
Speaker
what you would think about building number six, your other experiences would be less relevant. You can make it more intricate than that. So the function of the algorithm, the topic of our 400 page fashion document at a certain time, if you ever need bedtime reading, happy to share, is the ability for an algorithm to automatically generate context of an activity with no human input. So it looks at an activity,
00:29:09
Speaker
and says, I can generate context of what that means, like the activity will take paint. And our algorithm will automatically figure out what paint means. That's part one. Part two is we backtest. It's like sort of like, just imagine you don't believe anything I just said to you just now, which happens very often to me. Then you're like, well, we'll just test it and show you, right?
00:29:31
Speaker
If the thing doesn't understand your project and that you've got funky soil and weird politics in the thing that you're doing, then it won't forecast it correctly, period. The forecast will be wrong. We'll show you that the forecast is not wrong. The forecast is really freaking correct. And so that becomes unequivocal.
00:29:50
Speaker
Yeah, I can believe like I can, I can believe the story for sure. If I put my project manager hat on or pretend to play one in real life that like, okay. You're like, uh, you you've got this capacity. Like if I was a project manager or planner and I've seen 700,000 projects, okay. Yeah. I will, I'll know a lot more. But the flip side of that is like, if I did, if I wasn't seven thousand.
00:30:13
Speaker
projects and still alive and I was on a project, I would be able to touch, see, feel, be in the room. I'd have more sources of input and not just like looking through the lens of the schedule. And is that, is that something that, because that probably touches on this idea of like, what are the, one of the points that I've definitely had this conversation before of like, you know, so much of what's happening isn't in the schedule.
00:30:40
Speaker
type of conversations. It was like a conversation you end up having quite a lot.
00:30:44
Speaker
Okay. Yes, it happens, but not, I wouldn't say a lot. Perhaps the best part of NPLAN's forecasting engine is that if it doesn't think it knows what's going on, for whatever reason, either you're doing something super unique, but it was never seen before, or it just thinks, I can't generate enough context to really be able to provide enough confidence here, it will either decline to forecast,
00:31:10
Speaker
Or we'll spread it uncertainty, right? Like that paint job could take you five days, could take you 500 days.
00:31:18
Speaker
or anything in between that. That's basically saying, I don't know. Something human beings could do a better job of sometimes. So in a way, the system won't hallucinate or it won't bluff when it doesn't know something. So if there's information that is beyond the context of the schedule, so it's looking at the schedule and looking at sometimes progress and saying, hold up. The way this thing is tracking doesn't seem like anything I've seen before. It doesn't seem like
00:31:44
Speaker
or even pre-tracking, right? And you have no progress updates to a project. It'll just be look at it and be like, I don't know. I just don't feel confident about making this forecast in a way that can be relied on. So that happens to our clients as well. Yeah, yeah. That's super interesting. Sorry, Karl. I've gone over to back to you, right?
00:32:02
Speaker
It's interesting the, um, the angle of like, you're giving people the experience that they don't haven't had the time to gather the sort of gain. So like it is an experience industry. People are old when they're senior. So this should in theory accelerate to like other industries where you can have young individuals who they'd be great at reading these models, but it's really old and friendly big brother. Yeah.
00:32:29
Speaker
That's the one. Right. I think we're over time. Dev, thank you very much for coming on. Really interesting conversation. So yeah, that was a really good chat. Thank you. Yeah. Thank you very much everyone for listening.