Subscribe
Share
Share
Embed
AI is changing how we work, but the real breakthroughs come when organizations rethink their entire foundation.
This is AI-First, where Box Chief Customer Officer Jon Herstein talks with the CIOs and tech leaders building smarter, faster, more adaptive organizations. These aren’t surface-level conversations, and AI-first isn’t just hype. This is where customer success meets IT leadership, and where experience, culture, and value converge.
If you’re leading digital strategy, IT, or transformation efforts, this show will help you take meaningful steps from AI-aware to AI-first.
Colin Stoner (00:00):
People aren't late, people haven't missed it, but the sooner they can figure out how to leverage it and how to make it part of their day-to-day, the better off they're going to be.
Jon Herstein (00:10):
This is the AI First Podcast hosted by me, Jon Herstein, Chief Customer Officer at Box. Join me for real conversations with CIOs and tech leaders about re-imagining work with the power of content and intelligence and putting AI at the core of enterprise transformation. So I'm curious, I obviously want to dive into the AI topic, but I want to start with a big picture question, which is when you think about just over the last couple of years, just the surge of conversations that have been happening around the possibilities of AI in the enterprise, not personal use. What are the opportunities that for you really stood out the most at first? When you first started understanding the potential for AI in business settings, what were you like, oh man, we got to go do this?
Colin Stoner (00:58):
I guess when I started these conversations around the office, when it sort of became this, oh, we can tap into this API and start throwing some documents in, asking questions. Once I got over trying to talk to it or have it talk back to me like a professional wrestler and get to work, I would go and ask people, "What do you hate about your job?" Which is a fun conversation to start with someone because they're going to look at you very skeptically when you ask that. But the whole point was trying to remove the stuff people don't like because that's a great spot for AI to come in. It doesn't have feelings about this stuff. There's no bias for it. You give it to task, it goes out and completes it and gets its little carrot at the end of the stick. So I started there, but that leads to a ton of administration, replacement, a lot of workflow stuff, a lot of things that they're not sexy.
(01:48):
It's not necessarily the most fun things to solve or throw AI at, but when you can see the time savings and see that person no longer have to open a letter with certificate of insurance in it, enter it into the system, file it away, and it can just be done over email using AI. I mean, it's a huge leg up.
Jon Herstein (02:07):
Yeah, I remember the first few times playing with some of these tools and you're like, "Oh, write a limerick about a construction project or something." I'm like, "Oh, that's really fun, but that's not really helpful today."
Colin Stoner (02:16):
That was kind of my first two or three months where it was just like, oh, this is really fun. And then also, what am I going to do with this?
Jon Herstein (02:23):
Right, right, exactly. Yeah, I think everyone went through that similar sort of journey and I love that framing. And I feel like we recently had another guest on the podcast who did the same thing, which was, what do you hate about your job? If you're in a job and there are aspects, and you're probably not talking about the people doing the physical work for the most part, but people who are managing projects, the supervisors and saying, "What aspects of this job do you not like? " What you probably hear a lot is paperwork, checking things, QA, that kind of thing, versus actually being on the job site and getting things done. So what were some insights that came out of those early conversations about what people didn't like doing where AI could play a role?
Colin Stoner (03:01):
Yeah, the certificates of insurance was a huge one. And more than anything too, that was just a massive, massive waste of paper. So every single project we have, we subcontract out, every subcontractor has to maintain a certificate of insurance throughout that project. And then for a lot of our projects or even most of them, we have to maintain that certificate of insurance for the decade following project completion. That's a lot of compounding paperwork over time. So we would just have these stacks of open envelopes with COIs sitting on the receptionist desk waiting to get processed. And we went through and redid our subcontract agreements a couple years ago, we decided to take a real hard look. And this was kind of our first use case of, can we do this? I mean, when you look at the Accord form for the COI, it's very well structured.
(03:46):
I mean, this is not visual intelligence. It can't see the table. It's still just using RAG and pulling the text out of there. But it's structured in such a way where it's kind of a key pair relationship. And then you can have them throw the project code into the description on the COI, and then we can sort of stitch all that together and pull out 98% accuracy across ... I think we did 1,300 certificates of insurance over the last 30 days here. So it really scales well and fast, and it's like immediate relief for some people.
Jon Herstein (04:17):
The first win feels like get rid of the paper, so just probably not losing track of things, that sort of thing, getting the COIs associated with the right projects is probably a pretty key value. What about searchability of it? Let's say you've got to go look for one of these a year later or 16 months later and say, well, did we actually have the coverage we needed to have or there's an issue that comes up? Has it helped with the findability of the right information?
Colin Stoner (04:41):
Yeah, and that goes back to having Sentinel as that system or record for us. Got it. All those COIs get pushed back into Sentinel associated with the right projects. They're all visible. Stored in books, of course. Box is a tremendous extension of Sentinel for us for all of our project files and content management within those big, messy projects at times. It really just streamlined everything. As the first one, it's kind of funny for me to just sit here and think, yeah, that's still easily the biggest one and it's still my favorite one. I'm sure that's going to change someday because again, not that fun to talk certificates of insurance and processing them, but it's an incredible weight off of people that don't have to think about that anymore.
Jon Herstein (05:19):
Well, and it's real business value. It's actually solving a set of problems for you with a thing that's incredibly important for your business. The things that people demo around AI are cool, they're sexy, that kind of thing, and maybe business value as well, but this is one that maybe it doesn't demo as well, but actually solves a real problem. And I think that's what's most important.
Colin Stoner (05:40):
Yeah. Behind the scenes, everybody loves it. Try not to be telling everybody about it just because it's a tough one.
Jon Herstein (05:45):
Maybe you don't brag about it at cocktail parties like you do, but they do love it. Well, so you mentioned that, but you've also got, I would imagine in all these projects, you've got things like drawings, change orders, contracts, agreement. Are there any other areas that you're exploring next or have already implemented?
Colin Stoner (06:04):
Other one we did recently was processing. So we're a little bit different. We can get into the ACH game. We haven't found a really good vendor relationship on that side. And also we've noticed a lot of people having issues sending ACH payments to the wrong addresses coming from largely impersonation man in the middle kind of attacks. So we kind of just avoid that. We still send checks. But the little trick with that is those all print out and then I need to get a digital version of them. So we take those scanned checks, same sort of thing. I set up the ability to parse through each page of the check, identify the check number, use that same system of record to tie it back to the project, identify which subcontractor and which subcontract this payment is associated with. And then we attach that to the lien release in the system and have that sent out through, of course, Box Sign.
(06:57):
And it streamlined so much. This was another one where I started talking to the team and I'm like, "Man, I don't know that this is the right..." A lot of this is, is this a good use of our time? Is this where we need to be focused on? I bet. And then I got a little bit down the road, pulled a couple strings. I was like, oh, okay. Now I can really, really see the value in this. And what it's going to do ultimately is just allow us to kind of agentically send, receive and collect our lien releases. And I can just build an agent to be able to handle all of that for us rather than our admins who bless our hearts. All this is just clicking through different screens and they can be freed up to do other things. I have one admin that kind of appreciates this level of busy work that she confessed to me and she says she might miss it, but I think she'll get over it pretty quickly.
Jon Herstein (07:45):
Well, and I think there's always, I mean, we'll probably get into a little bit of the people and culture side of things, but I would imagine for someone in a role like that, there may be that fear of like, oh my gosh, well, if this can do all these things that I do every day, what does that mean for my role? And job security and all that sort of thing. And I would imagine certainly the way we think about it is it frees you up from that drudgery of that kind of work, and now we can actually use your brain and use your talent to do things that are higher value for the company.
Colin Stoner (08:13):
Yes. And there's a ton of that, especially in construction. I think you get people from all walks of life around here. There's plenty of people that love the idea of more technology. There's plenty that are scared of it, terrified of it, don't understand it. So I mean, I feel like I have to walk a pretty fine line with that. I've certainly embraced it a ton. I do kind of laugh at this sort of misnomer of AI is going to bring on easy mode. I'm going to get paid full-time, wages for only 24, five hours a week of work. And I'm going to tell everybody my last 12 to 16 months since I really started involving CloudCode in my day-to-day, I've never worked more. Now that's not because I have to, it's because I love what I'm doing and I'm extremely passionate about it. And CloudCode helps me build so many things so much quicker than I ever could before as kind of a part-time contributor to this project, but it's not all easy mode.
(09:09):
I do know that.
Jon Herstein (09:09):
And what about knowledge sharing? If you think about all the documents that we've talked about and some of those needing to be on site, some of those maybe not, but available for reference or archive purposes or whatever. Have you solved that problem of knowledge sharing between sites and all these vendors and partners and the home office?
Colin Stoner (09:28):
I think I'd be lying if I said I solved it. I mean, just it's kind of an evolution, right? Moving all of our content to the cloud with you guys, what, 13 years ago, I think 13 or 14 years ago, that was step number one for being able to have better access to that information. We're not in a Windows file share environment anymore. Everybody has their project permissions. We can maintain that, keep that level of governance in there. The next step with that that we've been, I guess, implementing is building a knowledge graph. So the system of record we have also references box files, whether it's a RFI attachment, whether it's an extension of a submittal item as a product manual, whatever. So we got to the point where we decided we wanted to create a knowledge graph and index a lot of project documents that are in our system of record.
(10:20):
So played around with that a little bit, saw that, oh my gosh, I don't totally understand. I mean, we kind of look at this like Voodoo. We don't totally understand how these vectorized searches work other than it does something like this, I think is the way it's generally explained to me. But so when it does that, and then it can pull in that specific submittal item manual from Box files where it can find the architect's response to that RFI and the shared folder for that. It's incredibly, incredibly powerful and just extends beyond that layer of this. It gives much more richness and depth to the answers that we're able to bubble up to our users.
Jon Herstein (11:00):
This is actually one of the things that we've been hard at work on is this bringing together kind of classic search and finding the documents that are most likely to relate to the thing you're trying to find and then using AI to actually go get the insight from that. And so
Colin Stoner (11:14):
We're
Jon Herstein (11:14):
Coming together. And that plus the knowledge graph that you talked about probably means you can stop having people dig through and dig through the folder structure to go try and find the right document and then read the document. It's like, this is just my question. This is all the stuff and the knowledge I have access to. Tell me what the answer is and cite the source. Tell me where you found the answer. So if I actually want to go read that agreement or go read the certificate, I can do that. I can go to the source. So you're making it and you have made it easier to get information, to get insights, to answer questions, that sort of thing, which is great. How are you balancing that with security, access, making sure that if a supervisor's looking for an answer, they're only getting answers that you want them to see and not answers they shouldn't see.
(11:58):
Maybe safety's not a great example, but maybe you've got firewalls between projects. So if you're working on this project and it's confidential, you don't want other people to know what's going on there. I know you have some confidential clients. So how are you thinking about that access, security, privilege, that sort of thing?
Colin Stoner (12:14):
That one's kind of baked in from ground zero for us. A project team can only see their projects in Sentinel. And once you're in your project, then you may join the box folder as a collaborator at that level. So we kind of insulate against that right away. So then with the chatbot within these projects, it's only specific to that project, and we just carry the scopes all the way from the user permissions at that basic level within Sentinel and bubble them up into the chatbot or bubble them up into the way that they're able to view the box files. I call it the secured file for budget and cost specific stuff that isn't necessarily open to the entire project team, so they'll only have access to that as well. So bless my CTO on this one. He built this perfectly from the ground up and it just has not an afterthought, but we know what it's doing.
(13:01):
We're pretty confident in how to approach it.
Jon Herstein (13:04):
So your system of record is essentially your construction management system and then that then cascades out to anything else?
Colin Stoner (13:10):
Yep, correct. It kind of sits at the top and everything inherits from it from there.
Jon Herstein (13:14):
Makes sense. Okay, great. So I want to maybe shift the conversation a bit from technology to a little bit more about leadership and culture and change management, because this is something that I think we all grapple with. And let me start with the change management piece, which is whether it's AI-driven tools or otherwise, how are you approaching change management? How have you approached change management, particularly with a population, as you say, that are not all technology friendly or technology comfortable. How have you approached that? And are there any lessons learned for other folks in terms of do this, don't do that. Everyone struggles with this. So what have you learned?
Colin Stoner (13:52):
Patience.
(13:54):
That is the easiest answer. A couple good antidotes here, but in the fall, I hit the ground with Claude Code coming out about this time last year, maybe a couple months before. Enthralled, love it. It's perfect for the way I need to do my day-to-day. My CTO saw it as a lot of Cody asked to review, they didn't ask to review before, which was a fair criticism because it was a lot more lines of code coming out of me. But so I want my guys to use this. I want them to see the benefit in it, but I'm not going to force it down their throat. It has to be on their terms. It has to be on their time, especially when we're talking about software engineers because at the core of ClaudeCode is this hanging over like, here's the thing that's going to take your job and displace all of software engineering sometime in the next couple years.
(14:43):
So that's kind of square one for these guys a lot of the times when they look at it. So I just keep doing my thing. They see what I'm able to do, how I'm able to build it. And then I was hanging out with my CTO last fall and at the office having a beer after work, just chatting and he says, I need you to know that I wrote zero lines of code myself today. It was all cloud code. And that was one of those moments where I was like, oh, okay. Yep. Yeah, it was rewarding for me because I knew playing the long game, he's going to come around, he's going to see it. There's no way. But I try and apply that everywhere else. I mean, this is one of the most technical people I know having a little bit of resistance to that.
(15:29):
How's the superintendent going to view it? How is the admin or the accounting staff or even the project manager going to view it? So it's a journey. I had slides yesterday at our superintendent meeting and I called it the AI journey. It's not something that's so binary like, "Oh, we do this now or we do that. " It takes so much imagination to be able to get the most out of these tools. And I think that's sort of a square one starting place for how to instill confidence or how to proceed with AI is use your imagination, push it, push it. And I try and instill that on our people as much as possible.
Jon Herstein (16:06):
Right. Yeah. There's something that I've heard very frequently over the last, I don't know, six months to a year, which is don't give up if your first attempt at doing something with AI fails because, and we all know this, they just keep getting better. So the thing that you tried three months ago that was a dismal failure, you talked about hallucinations in the first couple of years. If you ask that same question today, you'd get a very different answer and a much better answer in almost every case. So I think part of your message to your teams is like, yeah, even if it didn't really feel that useful to you when you tried it before, give it another shot. So I want to churn a bit to the future and maybe also ask for a bit of advice as we go for your peers and start with saying if you look out, I don't know, it's so hard to say how far out you can reasonably look and have a valid point of view, but let's say three to five years, how do you see AI continuing to transform construction operations, construction projects, your industry?
(17:05):
I'm
Colin Stoner (17:06):
Going to see it right now with you guys with the Box Automate platform. I'm starting to see that come from other construction specific vendors as well, having this agentic approach, agentic view to the data that already exists. I think that's going to be incredibly powerful. I think I just made it three or six months past. We are right now. It changes so fast. For having my pulse on this industry sort of day-to-day, I do not have a really great answer on where this is going. Other than hold on tight, play around with it as much as possible. The more fun you can have, not necessarily doing work stuff. I mean, I love sports, I love markets. So this fall, I built a bot that would go out and scraped Kalshi markets mostly for sports. Tell me what's going on there. I could build some models around it.
(17:57):
Long story short, this thing didn't do very well. It went broke with all of my fake money that I gave it twice. So the models, maybe lacking a little bit of data, I need some more points in there. The best thing you do is just get in the weeds and play with this stuff as much as anything. And again, I think I mentioned it earlier, but the sooner you get in, the more prepared you are for the next incredible version, upgrade, whatever that's coming, and you're able to leverage that right away. Sort of the way I've tried to position Novo going forward with this is just put us in the best possible position to be successful. I think we already have had a lot of dumb luck getting to this point, having our data staff completely integrated and ready to go. You have other companies trying to play catch up and do that, or they're spread across multiple systems of record just because of the nature of this business.
(18:47):
And being able to avoid a lot of that saves us the trouble. But man, three to five years is alone.
Jon Herstein (18:54):
It was not even a fair question. If you're talking to folks who are maybe not as far along or maybe you don't feel quite as well prepared for everything that's happening now, is there advice you'd give them now? What would you say to a peer, a CIO at another company who's not as far along and say, "Well, here's what you should do starting tomorrow."
Colin Stoner (19:14):
Yeah, I think the best you can do is lay a foundation. You have to have good groundwork, a good base to build all of this on. Just like we were lucky with building out Sentinel and going that route, we were incredibly lucky to partner with Box. You guys came along on that journey step by step with us, even all the way to Box AI, being able to extend our, what, 27 plus terabytes of data in that system. It transformed what we were able to do so quickly and being able to build and have a partner like that, I think it was critical. So all you guys are core for all of our documentation, for users, for projects, for everything. Being able to have that right partner, the correct company to run with and grow with and just have that foundation built. Don't rush to get that foundation done either.
(20:00):
You have to do it right. You have to structure this properly. If you're doing it on a programming level, your models better be really well thought out, really well related. And by the way, you can have AI help you with that.
Jon Herstein (20:13):
Right. That's a great point. Even if you're not super advanced on AI, you can use AI to actually help you get some of these pieces in place and even make recommendations around things like you want to migrate content, what content do you migrate and what were, what's critical, that sort of thing. We've seen the same thing. And one of the things we talk about a lot of the challenge of all of this unstructured content sprawl that organizations have built up over years makes it really hard to get the value of AI because that means you've got to point AI at all these different things. So I think to your point, and obviously we would love for it all to be on box. It doesn't all have to be on bots, but getting into a place that is AI friendly, that's robust, that is secure and integrated, now you'll start to accrue the benefits of AI on top of that.
Colin Stoner (20:56):
I think the secure and integrated is critical too, and it was a bit of an oversight for me, my initial assessment, but you got to be careful where you put this stuff. You got to be careful who's allowing to interact with it. That foundation is critical. Do not rush it. You're not going to miss the boat, but also be sure to be thorough and do it right. And again, the ability, even if you draft up your models or sort of build that baseline, you can vet it with the AI. It doesn't have to write the code for you. It can do the analysis. It can do the audit portion of it. It can provide suggestions. It does not have to do all of the baseline work for you. So I think that's still really important to understand and maintain. If you're just having it write a ton of code for you, but you don't necessarily understand what it's writing, that's where it's going to get dangerous.
Jon Herstein (21:40):
Right, right. Okay, great. Great tips. Thank you for that. And I want to wrap with a little bit of a rapid fire thing, which is in my role in customer success, I think a lot about three domains, value, culture, and experience. And we actually touched on aspects of these, I think, throughout our conversation, but I'll just hit each one with one question. You give me a relatively quick answer on each of these. So on the question of value, how do you define the business value of AI initiatives in your organization?
Colin Stoner (22:07):
The business value, again, efficiency gains as much as anything. I think consistency and documentation is sort of a second one, and I think that's going to dwindle over time as we get more into a visual world, working more with drawings, working more with photos, 3D captures, that sort of thing.
Jon Herstein (22:23):
Great. Now, the topic of culture, what most determines whether new technologies like AI are successfully adopted?
Colin Stoner (22:31):
Here, it's kind of a reality. I don't know if that's a generational thing, but you get people talking about a cool feature, a new tool that we built. It permeates amongst a project team. They talk to their friends elsewhere in the company, and then it kind of goes from there. So I'm not a big social media guy. I don't understand it that well, but it's definitely a reality.
Jon Herstein (22:54):
Do you change champions that help bring that forward and talk to other people and that sort of thing?
Colin Stoner (22:59):
Yeah, definitely. I have my influencers, so to speak.
Jon Herstein (23:06):
So you're
Colin Stoner (23:07):
Not- I got to get in-
Jon Herstein (23:08):
But you know how to influence people through others.
Colin Stoner (23:10):
Yeah, you got to get it into the right hands first.
Jon Herstein (23:13):
Exactly. Right. If you think about your 12 or 13 supervisors, you've got one or two that are just like, "Oh my gosh, this has completely changed how I get my work done." That becomes, in a way, a lot more trustworthy than you saying the same thing, even if you're
Colin Stoner (23:24):
Super
Jon Herstein (23:25):
Well-respected because they're doing the job and you're not.
Colin Stoner (23:28):
And half the time when I'm up there presenting, they're not paying attention anyway, and that's just the way it works. It's not made different anywhere else.
Jon Herstein (23:35):
We're paying attention to your AI journey conversation, I promise.
Colin Stoner (23:38):
Maybe a little bit, parts of it.
Jon Herstein (23:41):
All right. So the last one is a valid experience. What criteria do you use to determine when an AI capability is truly ready for enterprise deployment?
Colin Stoner (23:49):
I think the beauty of being able to build for one company versus being somebody like a box where you have tens of thousands of customers is they have a little bit more ability to screw up and roll that back. So there is plenty of, wonder if this is going to do what we need it to do, being able to vet it that way. It's mostly user feedback. We listen to them a ton.
Jon Herstein (24:11):
Right. So try something, fail fast, listen to the users and iterate as you need to.
Colin Stoner (24:16):
A little bit of that early Facebook model, for sure.
Jon Herstein (24:19):
Yeah. Well, this has been a great conversation for me. I think for folks listening or watching, they're going to get a ton from it. So Colin, I just say I really appreciate a few things. One, you as a customer, the long partnership that we've had together, you as a vendor who's built amazing spaces for us to work in. So thank you for that as well. And then mostly, thank you for this conversation and for sharing your insights and your knowledge with others.
Colin Stoner (24:43):
Well, thank you as well for being a customer as we kind of started off full circle. I really appreciate the time and being honest with a fun opportunity. Thank you.
Jon Herstein (24:52):
Thanks for tuning into the AI First Podcast, where we go beyond the buzz and into the real conversations shaping the future of work. If today's discussion helped you rethink how your organization could lead with AI, be sure to subscribe and share this episode with fellow tech leaders. Until next time, keep challenging assumptions, stay curious, and lead boldly into the AI first era.