AI in AEC: Stop Chasing Tools. Start Fixing the Business.

Hello and welcome to Design Under Influence.

I am your host here, Alex, with architea.

My co host is Liz.

Liz, say hello.

Hello.

That's a good hello.

Our job here at Design Under Influence is to deliver technology wisdom and help to the AAC community.

That's all we do.

And these days, Liz, technology almost equals to two words, two letters, right?

What are those?

Oh, you get like we planned it.

Yep.

AI.

Exactly.

And we just launched a new product in that space helping companies deploy AI tools.

And I've had a lot of conversations with people that are and I have to talk them off the ledge.

What's happening here is folks have a wrong approach the technology and it's hurting them and making them feel like they're left behind, they're not doing what they supposed to do and they're going to be AI'd out of existence.

And I want to refocus the conversation and help people understand how to deploy AI tools within their AC firms.

Okay.

And we're specifically talking small to mid.

Nobody cares about large people.

They know what's up or they think they do, but they have the resources to play around and stuff like that.

I'm talking about us, the smaller people, smaller businesses who work 14 hour days and don't have extra resources.

So the way, let me position this, okay.

So the way a lot of people are thinking about this right now is all in terms of tooling.

Oh my God.

This new cloud code came out.

Open claw.

Oh, I have to be on this.

This is, this is where it's at.

We need to optimize everything.

How do we use Claude for Excel based budgets?

Those are questions I get.

What's the best AI for bim?

Right.

How do we automate everything?

Those are the questions I get.

And it's very difficult for me to refocus the conversation on what the core principle is.

And the core is improving your business.

It's really identifying what needs to be improved and improving it.

And we'll talk through that framework.

But what is your knee jerk reaction list?

What are you hearing from your peers?

Yeah, I think that there is a lot of that fear of missing out and wanting to adopt, but not having time to adopt and not really understanding what does adoption mean and trying to do everything all at once and it becoming more of a hindrance or a burden than a help, which is what it's supposed to be.

All right, I agree.

It's hard to talk.

The first, the first 10 minutes of conversation is basically trying to educate someone why they should be calm, cool and collected and not get a little frazzled about them feeling behind.

But let me lay out this framework and let's work through this together, because we all fly in this, putting this plane together while we got pushed off the mountain.

And so that's what we're doing.

But there's a proven approach to this, and our recommendation is the core framework of this is to really come up with a North Star business problem first.

We call it North Star because that's kind of, everybody's looking at it, it's the shiniest point in the sky.

And that takes.

We'll talk about how to get there.

But when we qualify a North Star problem, I'm saying it's something that, the way we think about it, it needs to improve our business between 100 and 300%.

So whatever that core problem is, it's very important to identify it first.

But before you start running around and solving it and testing all these different tools, our recommendation is to break it down into multiple components and focus on one component at a time.

Optimize, understand, build, test, rebuild, document, deploy, get feedback, redeploy one small component of that North Star.

And we can use examples like, for example, like what's some of the North Star things that you see other architecture firms should aim for.

Would a proposal turnaround time be one of them?

I would say proposal turnaround time, but not just that, but also being able to harness the data that you've got in all of your systems in order to create a better proposal.

Because I think like right now, a lot of firms, we use our like the stomach feeling about how much time will this take?

What will the cost be?

And that might be like step one is, can we create a more quick proposal?

Then the next would be, can the proposal be more robust, as in building on data we have in our time tracking system in our from the past.

How much time did we say it would take?

How much time did actually take?

How much time should it take?

And then the third, I would say on top of that would be maybe be using all of that too.

And this is a, maybe a bad word in the industry, but to go to a fixed price.

So how can I use all that data to create a quick proposal that's accurate, but also will make us money and make our clients happy?

Because it's understandable and it's a reasonable price.

And we can point to, if someone says, why does it cost this much?

You can easily point to three different projects of a similar size that took X amount of time to produce.

So, so it's like data backed proposals that anyone can understand and can be produced quickly.

Great example.

Thank you.

So let's unpack.

Okay, so proposals is something that we feel are Northstar at my, let's say imaginary AAC firm.

Okay.

We found that we can improve the outcome for our customers and our company by 100%.

We can at least double or triple the value that we bring to the clients and our own value.

Okay.

And then within that proposal, big hairy goal, so to speak, North Star, there are three components you just helped us identify.

One is time turnaround time.

Two is that data utilization.

How do we get, how do we enrich that proposal in order to come up with the ultimate pricing, correct pricing, fixed or not fixed.

That's a decision that's purely onto the business.

But sure, but I think.

And then from here we go deeper.

Now each component needs its own, like we need to break it down into further, for example, time.

I don't want necessarily you and I to create a whole thing here, but like in what are some of the sub components of time you think or turnaround time?

Well, probably a template, some kind of template file that has your basic language in it, a way to fill that in quickly and accurately.

And then some kind of calculation tool if it's in Excel or some other thing that's helping you guess.

This project takes six weeks.

We need two people, two and a half people per week, basically mapping out the time to get to a number at the end.

So some kind of calculation framework.

So a template and a forecaster.

All right.

Intelligence.

Right.

Intelligent template that adapts, that has the right formulas and already pre, pre existing language that's been vetted.

And also a forecaster that allows you to quickly map things out.

Yes.

Okay, so those would be the two components that I would start writing down in this one larger proposal component.

Right.

And we, the way we do this for workflows is we do the same for say, RFI responses or BIM coordination conflicts or anything like that.

Now in business it's all about ticket response time and quality of resolution.

Okay.

So when a customer has a problem, our job, our whole sort of existence relies on this one core focus we have, which is we got you.

Someone competent is on the case.

We are coming up with resolution and we'll let you know as soon as possible.

So that's someone competent is on the case.

We got you.

It's essential, the essence of it.

And so for us, we deliver.

We, we build our North Star based on that concept and then we break it down to small components.

We can use ARC IT example later, but suffice it to say, what Liz and I just did in the X in a very short time is the central of AI adoption strategy.

Do you agree?

I do agree.

I agree.

I think it's this the way you should think about any kind of adoption of any technology.

It's not what can the technology do, it's what problems do I have and then match the technology to solving them.

And that's the hardest thing, right?

That's the hardest thing.

That really takes some whiteboarding.

It takes multiple team members who care pulled in and bought in.

So I'll give a little tip.

One of the reasons you want to bring other people into this, aside from the obvious being you get a group think you get a lot better decision making.

But it's also bringing people into this journey with you who will continue this journey without you.

Yeah.

On their own.

Future leaders, future thought leaders, folks who will get the momentum of your improvement going.

Yeah.

And I think that's the piece I do want to spend a bit of time.

Let's talk about how once we identify the component tree within the North Star, how do you guys have done very good job on like getting a handle on AI and deploying it within your business.

So I'd like to get your thoughts on how to approach a component of a problem.

So we North Star is here.

We've broke into five, six components.

How do we attack each one?

How do you start?

Wow, that's not easy.

We're building a plane.

Let's do it.

Let's do build it while we're flying it.

Yeah.

For me it would be.

You have to understand what is happening now.

What is the current workflow and what parts of that are good and we want to keep and what parts are bad and we should either optimize or throw away.

Because I think that there can be a mistake if you just solve it and make.

Let's say you talked about you have to have your team of people who are going to be the early adopters and the champions and the ones pushing this forward.

But if you're making everybody work in a completely different way than they're used to, it's harder, much harder.

But if you sit down and you look at.

Okay, we do steps one through five, step one is good, step four is okay, and step five is terrible.

And even at the end, you throw them all away, at least now you understand why you threw them away.

And it wasn't just for the sake of cool technology.

But you learn the process.

Then you look, at least in my world, replacing steps, perhaps, or whole groups of steps with AI or with a tool that can do that for you.

And so instead of, like I mentioned before, want to do everything all at once at the same time, we're going to fix this whole thing.

Maybe that's too hard.

Maybe.

And I'm guilty of that.

I want to fix the whole thing.

And my husband, who codes, he often is reminding me, no, you just can fix a little piece and then see if it works and then add on to that, and then add on to it.

You don't need to get frozen in the fact that you don't know how to solve step three.

So therefore everything's broken.

That's a perfect mindset.

Let's provide an example.

So we just did something for arc it that it's going to be applicable to your business and anyone else who's listening.

Okay, I'm excited.

So service delivery is everything for us and everything for our clients, and everything's for you.

That's ultimate goal of what we do.

Who delivers that service?

It's the people.

So the people is the main component of it.

So you gotta identify.

Within our North Star of resolving tickets on time and making sure clients technology works and they focus on their business is the fact that we need trained quality people who say, I got you qualified engineers on the case.

We're going to come up with resolutions shortly.

Okay.

For us to do that, we have to have quality people.

So the first line of defense for us is the level of one engineer.

So within that large North Star, we identify a component like, hey, we need to give people tremendously valuable feedback so we can see their growth through the organization or get them moving on to different opportunities where we are not a fit.

That's crass, but that's.

Oh, that's true.

We deliver service to our clients that need to be excellent.

Otherwise somewhere else might be different.

But so for us, it's that we do a quarterly review.

And quarterly review with L1 level 1 engineer is a core opportunity to give feedback.

So we looked at that process and we said, okay, what is we.

What are we doing currently, as you said?

And we had exactly.

We mapped out exactly what we do.

And that's no different than anyone else.

You seldom send them a questionnaire.

A manager goes through like 8 12.

Sorry, 12 different.

12 different qualitatives.

Things like your first response over the last quarter was this, here's my 1 to 10 score.

Here's where you can improve.

So the manager takes half a day to go through that thinking and then we delivered.

Then we get together and we explain it to the employee.

Hopefully they have some takeaways and we move on.

That's in a nutshell.

The way we've reoperated it now is AI does a lot.

So all the one on one conversations with the manager weekly are all recorded in Fireflies.

We're using that intelligence to build.

We're using AI to be able to interpret all those conversations and the feedback that we're given over the last 90 days.

Then we put them into the format and we ask AI hey, can you please look at review all that data and put it in those 12 different questions value metrics and give us feedback on how each one can be improved.

Give us your score.

Then the manager reviews that information.

Then we layer on the KPI performance which we track over the last 90 days and we compare them to their peers.

So that becomes two components of the review process.

And then finally we have them fill out the self assessment.

And that self assessment also gets into that same process and via specific prompt engineering.

We came up with a methodology that is consistent and extremely.

I just got off an hour and a half call with an employee.

I don't know if you have been to employee review calls.

Yes, I have.

Dreadful.

For everybody involved, this one was not takeaways.

The suggestions.

The next quarter goals were so precise, so well thought out.

Everyone was engaged in that meeting.

That was amazing.

That's one piece of the component.

Now as we do more of these reviews, we document the process better, we enrich the data other for example, we can pull all the tickets and bring that into the data set to get to provide better feedback.

So this is how we feel companies should go into this brave, this new AI world is let's solve it.

We have new tooling to solve good old problems.

And one thing I would add to that is that you didn't touch on would be that you've eliminated bias.

That's a good point.

So if you have a manager and employee who don't get along and they bring this to hr, for example, HR can trust these evaluations because the AI doesn't have a problem with anybody as long as you write thank you in your prompts.

One cautious thing here is it's all we're innovating as we go, right?

We don't take this at face value at all.

So here's the challenge, right?

And I had to teach my managers to think that way.

It's what GPT or another tool turns around for you, is this as good as your prompt?

Right.

As good as the data that you feed it.

But it also does not have context.

No, it does not have a lot of things.

So you need to use this as a framework and not as a like a God given truth that you will deliver, you will spit out and repeat.

That took a little bit of doing.

Yeah.

I could think of situations where perhaps someone's metrics go down, but you as a manager know it's because their mom has been sick.

So they've been taking care of them and so they haven't really been as focused.

And that's okay, you know about it and we'll go hopefully go back up when things settle for them.

That's a very good example.

Exactly.

The performance has dropped over time and we know why and we're just trying to support that person.

So our the way we phrase the phraseology in this review would be considerate of this fact and Absolutely.

What's one thing you.

I know you guys been cooking a lot of things.

What's one thing you've been.

You told us last time, but I want to hear this again.

If there's maybe something new that you've implemented.

We are cooking, we are trying to cook a better way to do our time cards basically.

And there are a lot of tools out there that should help with this.

But I can tell you that since I became any kind of consultant, because architects are consultants.

So as soon as I got out of school, the bane of my existence has been time cards.

Hate doing time cards and especially if you are jumping between different projects is really hard to know how you spent your time in a day.

And by the time the end of the day comes around, you're exhausted and your brain is fried and you don't want to sit down and make a time card.

But time cards feed into billing.

Billing feeds into the economy of the company.

It also feeds into what we were talking about before making proposals.

Because you need to be accurate about the time you spent on something so that you can bill it, but also so you can add that to your data pool for your future proposals.

And to capture all of that accurately now seems to be like lots of little pools of ways to do it.

There are like Big Brother apps that sit on your computer and watch you and try to learn if you work on this or you open this file, it must be this project, and so on and so forth.

But then how to take that data and clean it up so it will then.

Because we can make our time tracking system read the data, but it's messy data.

And then when you try to produce a bill from that or an invoice from that, it's even messier.

And we all know the embarrassment of sending an invoice to a client that they question, it can get sticky unless you have the data to back it up.

And if you can answer questions easily, then no problem.

So when you have a team like now working crazy hours to get a deadline done and they don't have time to stop and write down thoughtful notes about what they've been doing, they barely have time to eat or sleep.

So now this will come to a head, I think at the end of this month.

But we're trying to build and teach some AIs how to read our activities and interpret them into usable data that then our time tracking software can use.

And then the end result being that you just review it instead of having to create it from the start, you just review the data and edit it accordingly.

And that's like a maximum 20, 30 minute thing one time a month rather than, or five minutes one time a week rather than multiple hours of trying to figure out what the heck you did all month.

It's huge.

It is huge.

And this is a huge north star for us because if we can get that off of our admin plate, we will have so much more time to do the other fun stuff we want to do.

Build systems that do cool things or talk to new client or whatever else.

Seems like a lot more fun than making time cards.

Oh my God.

I want to get my fingers into this and start kneading.

So what are the components?

Have you, have you start breaking it down?

I know you guys been crazy busy caron fire kind of thing with this latest deadlines.

By the way, I really appreciate that you made time for us today.

Sure.

And I know it's hard.

I know you're going to send.

You're going to bill me for this time too.

Probably, probably.

But then you can send me a bill back for your brilliant idea.

My time.

Yeah, we just got to wash it.

Yeah, yeah, we're going to just pay taxes on it.

That's a great idea.

But in all seriousness, if you broke it all out, what would you think?

The first component that you could focus on is the.

It's the recording of the time.

So I think it will be.

And we use, as we've talked about before, we use soft drive computers.

So we control them and we do have some people use their private computers.

But asking, would you mind installing one of these tools that watches what you do and you have full control.

You can turn it off when you're not working, but would you mind doing that?

And hopefully everyone would say yes about that because it doesn't record like a video.

It just says this file was open and then this.

You were in your email app and then you were in this.

So it's.

There are ways to protect your privacy.

And I guess the question.

There would also be a lot of privacy discussions to have about.

Oh, yeah, for sure.

If it's installed on one of our soft drives, we can just say we shouldn't be doing private stuff here anyway, whatever.

But if we're asking a consultant or someone else, please, would you mind?

That could be a sticky discussion.

So I would say that would probably be.

Step one was collecting what's happening.

So that's the first component to solve.

I think so.

Okay.

We can break it down further, but I can tell you right now, my wife is a professional project manager and her company, it's all they.

It's done.

It's.

What do they use?

Nobody asked them questions.

I have no idea.

It's a medical stuff, so it's all compliance stuff and all that.

But it is built into their systems to their computers.

Just the fact that, same as you said, you switching between projects, you're doing clinical studies and stuff like that, it's all super important to have all that data.

And so it's there.

Okay.

But yeah, I think it's solvable.

It's just we haven't found the right tool yet.

So we're testing a couple different things at the moment.

I think you.

So what you've done is you've identified exactly what that first piece or first component of this and this.

Hey, data.

This data needs to exist before we start misogyny, before we go to open claw and build ourselves an agent that converts some hopeful data into a time card input.

That's all next right down the line.

Yeah, yeah.

And worrying about that now is like, it pauses you from real progress.

It does.

And what we've also discovered is that the first step, we've realized that we don't want these tools necessarily to try to learn what we're doing because they're not very good at it.

So we would prefer to find a tool that just collects the data, the raw data, and it looks at all the things it compares to what our actions are on the computer, but it also looks at our calendars and what's in our calendar.

It compares it against the actions and then also could look at Zoom calls and transcripts because we get UF fireflies.

We have Fathom.

Fathom sends us a transcript after every call, looks, reads those and compares against.

And then also would look at any other thing, like emails that are coming in, stuff that's going on in addition to our actions on the computer.

And then we want to take that raw data and those things that exist now don't really know what to do with it very well.

We've tried to teach them, but they're more.

Less bad.

So then I think the next step where we've gotten to is to actually process the data and process it in a way that the AI doesn't come back with.

Here's the answer.

But the AI sorts and organizes it.

This stuff I'm 90% sure about this stuff.

I'm 80%.

Here's the stuff I have absolutely no idea about.

So that you're teaching it, but you're also understanding the outshore is the agent about all of the things it's saying to us so it doesn't come back.

As the overly confident intern that just dumps all this stuff in your lap and you're like, but which parts of this is good?

And I think that's the next step that's fuzzy and we're still trying to figure it out.

And then the last step of getting it into our system, that is easy.

That is a CSV file that's organized with certain columns.

That's no problem at all.

You'll see.

Maybe we've tested that part and that part we could probably also.

Now it's a drag and drop situation, but I think in the future we could also probably have an agent do it for us once it's approved.

Gotcha.

Oh, yeah.

So that's final, like final step where you drop data, verify data into actual time card.

Yeah.

And then you drop it into the system, which is already set up to receive it in a specific format.

And we've tested that and that works great.

So, yeah, it's, I would say collection and processing and verifying before it goes into the final system is what we see it as the steps.

So I'd want to help people put the core structure into this effort of once you have the component, how do you evaluate what tools to use?

How do you evaluate if it's working?

So here's the workflow, here's the core kind of process we came up with.

You start with a prototype, you stress test it, you refine it some more, validate that inputs are right.

Like what you said.

I'm not going to trust it.

Let's make sure we validate it every single time.

It's just too important for AI to assume stuff now.

We document the whole thing, then we see it work, then we train the team, then we roll it out, then we get more feedback, then we make improvements.

So the team feels they have an input, they're not just being told what to do.

Then we measure adoption.

It could be on a huge scale of hundreds of people, departments, or it could be on a small scale, it doesn't matter.

The steps are the same.

And then only when that's done, I would say I would put done on that one and then move to the next one.

Now you can do things somewhat concurrently because while you validating and waiting for your team feedback, that takes a couple of weeks sometimes these quick cycles.

I wouldn't want to sit around and just hope because if you do have time, you probably prototype into the second component.

What I'm saying is don't run after the big thing.

And you guys did it exactly right.

So you have one thing that it's in flight that's almost done, a key data collection thing, but you already have ideas and processes for downstream stuff.

Yeah, and that's key too because if, for example, the, the if you have a fixed point in your stream, your data stream, like for example, our billing slash time tracking software is our fixed point.

We're married to it.

That would be a huge disruptor to change at this point.

So we're not going to unless God tells us we have to, basically.

So knowing that's our fixed point, it almost once we mapped the direction towards it, we could go backwards.

We could start looking at, okay, what talks to it.

Is there anything out there that already talks to this?

Or if we wanted to make something talk to this, how do we do that?

And then that.

Also, instead of having a hundred options for the Big Brother software, we get it down to three or two and we start with testing those.

And so I think it's important to come at the problem from multiple angles so that you don't start from the beginning and work your way to the end and realize, oh, now I'm stuck.

And we should have thought of this other connection previously or at least not solve it.

I don't say that you have to solve all the connections, but at least do enough research to say, is it possible to solve this connection?

Yes, there are many options.

Okay, we'll solve it later, but we know it's feasible.

I like the fact that we sometimes anchored into particular Tools like for us, it's the IT tool set is, is.

I wouldn't say rigid, but like the main system of record is like not something.

Yeah.

Not something we're going to change because we're trying to change one small process in order to get to our North Star.

Now sometimes that would dictate a big change because you were already probably you're using old or something and you're already behind most of the time you're right.

You have to connect to the team.

But for that tool, core tool set.

But that core tool's got APIs.

There's ways to do it.

I wouldn't want people to stress about that.

Last.

Call it last mile.

The last mile is everybody tries to solve for last mile before you solve for the core problem.

Yeah.

And I wouldn't say solve for the last mile, but make sure that last mile exists.

Or there are many ways to complete that last mile.

Like for example, what if our software, time tracking software, we find out that it's going to be decommissioned for something else in a year from now and we built all this.

Like that's kind of stuff you need to figure out.

Is there any roadblocks we haven't seen that will stop us, that we need to.

That will change the way we're thinking now?

And if, and of course you can't always solve and you can't know everything in the future, but you can at least try to foresee potential problems and at least foresee potential solutions and then put that to the side and start with your first like block or to Sprint or whatever you're going to break it up into.

Yeah, I think this is good.

I think this is, I think we've given people our current roadmaps and roadmaps we lead our clients through on how to optimize.

There are multiple.

There, there are a lot of different elements to this.

For example, like how do you actually document this?

Where do you document, how do you run projects?

How do you make sure that your AI tool set is secure?

How do you know what memory is?

It keeps.

How do you tell it what memory it should keep?

Because without memory you're losing out a lot of value.

Yeah.

With that memory, you exposing risks the wrong person access the wrong thing.

You know, I know if you heard about this, Grok revealed somebody's name, some actress name, whatever.

Doesn't, doesn't.

It's unrelated here, but okay.

Somebody asked what's that actor's actress's actual name?

And Grok said what it is?

And then nobody knew it was supposed to be a CP secret and all.

So it's a big brouhaha now and I feel for her, but it teaches us not, not to rely on these things.

So all these things in future podcasts, Liz, you, I, and hopefully Megan when she's out from the deadline, think about these things, discuss our findings and help people think through this way.

In the meantime, if anybody needs help with a little bit more about this, first of all, make sure if you need help with bim, Aurora BIM is your destination.

Happy to help.

Yeah, Super AI enabled.

These guys know what they're doing.

And then to help deploy AI in architecture, design, engineering firms, here's Arc IT ready, ready to help.

Do we have all the solutions in place?

No.

Do we think about this day and night?

Absolutely.

Yes.

Yes, we do.

And I would just also say don't be afraid.

Afraid of just trying AI in little places.

It doesn't have to be this big system that solves or improves your company a hundred, three hundred percent.

It can be little tiny improvements just to get your feet wet.

Yeah.

Do I think this is like in this, in martial arts?

The key, the real key, everything about martial arts is do it's not think.

It's not move like a shadow.

It's not sword fighting.

It's getting up and doing, just doing.

And it's consistent with this approach too.

Liz, it was a pleasure.

It was fun, like always.

You can find us@getarchit.com you can find Aurora Bim at aurorabim.com is that right?

Yeah, that's correct.

That is a good domain name.

We did.

All right, thank you very much for watching, listening and thinking through what we discuss.

And we'll see you all next time, next week, hopefully on two weeks.

All right.