#291 Why Most AI Projects Fail to Deliver ROI, Sinohe Terrero, CFO and COO, Envoy

Speaker A: The purpose of today wasn't to talk about fundraisers, it was to talk about AI. Why are AI implementations failing? Why isn't AI delivering the roi?

Speaker B: It should Full workflow automation is not really a thing right now, but task automation is really a thing and we've been able to leverage that brilliantly here at Envoy, something that was taking us about two hours to do on a monthly basis. I took it down to two minutes and I was like, well if I could do that for this one thing, what else can I do?

Speaker A: Grow CFO is where finance leaders grow together. Join thousands of like minded professionals using Grow CFO to access the combined knowledge and experience of the finance leader community. You can join us today. Growcfo. Net. Hello and welcome to the Grow CFO Show. I'm your host Kevin Appleby and uh, today I've got with me Sanoue Torero who is the CFO and COO of Envoy Senoe. Welcome to the Grow CFO Show.

Speaker B: Thank you for having me Kevin. Looking forward to this conversation.

Speaker A: Sonoe, tell me something about you.

Speaker B: I'm now what I consider a serial sort of startup cfo. Been doing this for a long time. Really enjoyed the operating mode of growth stage companies live in California. I've been working in tech now for sheesh, 20 years now.

Speaker A: So you're CFO and um, COO of Envoy. What do Envoy do?

Speaker B: So Envoy, we are a workplace technology platform so we do a variety of things that really help companies manage their spaces. So anything from visitor check in, mostly security. So right now I say that we're mostly a security and compliance company if you will because some of the biggest companies in the world use us to make sure their places are secure. We have a visitor login so when visitors come in you can check in. If you've ever signed in on an iPad, you've probably used our product Envoy. We now have emergency notifications desk M allocation tools. So basically everything that has to do with the physical space and the management of it, we have the software for it.

Speaker A: Brilliant. Quite a space to be in. Most of my experience is talking to fintechs so it's nice to be talking to somebody today who's not a fintech and um, seeing a little bit of different aspect of the world. But you say that you like being through early stage fast growing companies. What stage of the lifecycle is Envoy

Speaker B: at in terms of a life cycle? We are a series C company. We raised our series C actually back in 2022. So it's been now three years since we actually did that close. But we are in that stage of a company where we're growing into a good size ARR and have about 6,000 customers across the globe. And so from a sheer sort of breadth of coverage, we're pretty big. We punch above our weight, but we're still pretty much a developing company. We're about 250 folks, uh, most of us here in San Francisco.

Speaker A: Brilliant. I love that sort of organization where you're still small from the people point of view, but you're big in terms of what you're actually doing. So that fundraising process, was that something you were in the company at the time it took, right?

Speaker B: Yes, I was.

Speaker A: So, lessons from doing a Series C.

Speaker B: The lessons is make sure you time it right. I think that we were in the middle of COVID and the company was doing really, really well. And I think as a cfo, you have to make sure that you're trying to see 2/4 out, 3/4 out. And I saw some turbulence coming to the market in terms of valuation multiples and other things that were happening because everybody was riding high. But I've been in this for a long time and the ebbs and flows that happen. And I think the biggest lesson was, well, there were two. One was time it properly, of course, so have the foresight. But two, make sure that your story is tight and that the investment community can see that you can operate a company well out and efficiently. One of the things that really worked out for us, it wasn't just the growth. It was the fact that we were able to do it, showing the market that we can manage our cash flow, that we can manage our business, that we're not just out here burning money. And I think that really helped.

Speaker A: Yeah, exciting times. But the purpose of today wasn't to talk about fundraisers, it was to talk about AI. And you approached me with the, uh, topic of why are AI implementations failing? Why isn't AI delivering the roi? It should. So, Sonoe, why isn't AI delivering?

Speaker B: Well, I think as I speak to other colleagues and other CFOs, I think there's a couple of reasons why AI is not delivering in some companies. I think, number one, I don't think people are distinguishing what AI can do. And so it can lead to a lot of, I think, disappointment or people going uphills. Instead of really identifying where you need to deploy AI, I break it down into actually three. Three different ways that you can deploy AI. And if you do that well, I think you can benefit. One is augmentation, so you can Augment things like I'm sure you augmented yourself for this interview by using ChatGPT or Claude to do something. The second one is task automation and this is more for finance. I think that right now full workflow automation is not really a thing right now in most of the work that we do, ah, at finance. But task automation is really a thing and we've been able to leverage that brilliantly here at Envoy. But our approach has been to microscopically look at our work and then identify the areas that can be automated. I think anyone that's coming in and trying to do a full sweep automation of anything is just going to fail at this stage of where we are in the maturity. And then the last one is what I call application development which I spend a lot of time on, which is literally building full blown applications for your teams that can replace other applications that you've been using, not just from a cost perspective, but really from making it curtail to your business. Right. So now you've really fine tuned it to your business. And I think a lot of companies still don't have those application developers internally. And so the lack of in house application developers and the lack of identifying where AI actually fits in your stack I think is the main reason why people are not getting roi.

Speaker A: Now you describe yourself as the weekend vibe coder. Tell me more.

Speaker B: From the minute that cloud uh, code came out I ah, just had a curiosity about it because when we sit on boards, right we CFOs, the board is going to come in and they're going to have a very strong opinion as to what other companies are doing with AI. And so from the beginning I understood that I needed to understand it really intimately in order for me to have sophisticated conversations with people that are more exposed than I am to different companies. And so I really started trying to learn everything that I could. And in that process I just started identifying applications and things that we could do internally here where I can save a lot of time. So for my team I started building a bunch of applications to ingest information. So for example, the first application I built was a way for us to automatically reconcile our asset account from our bank. So we had this process where we would take the Excel sheet, we would take the PDF, add some numbers to an Excel sheet, bring in some other stuff, do some reconciliations, tie the short term, long term doing all that work. And I just said, well the machine can read the PDF, it can organize the information, it can create your journal entry automatically. And so something that was taking us about two hours to do on a monthly basis. I took it down to two minutes, and I was like, well, if I could do that for this one thing, what else can I do? And so I just started looking around and saying, well, where are the opportunities to save time by deploying these applications to automate? Because the machines can read much faster than we can and can organize data much faster than we can. So literally the last three months have been me looking around and seeing where there's an opportunity to save time or do things better and then just create, literally building applications from scratch, which is pretty exciting.

Speaker A: Should the CFO be doing that?

Speaker B: I think that CFO should be doing that because you have two things. One, I think you need to lead by example. And two, particularly initially, you need to understand the technology enough to understand the security aspects, because it's very different than a spreadsheet. We were trained on spreadsheets, and in a spreadsheet you can look through the formulas and you can do your own audit of a spreadsheet. But if someone is building an application and handing it to you and you have no idea how that is working, you're really flying blind. And I don't think that in this stage where we're at, uh, right now, in terms of application development, the CFO has the luxury of flying blind and just taking things on face value. Because for as awesome as these applications are, they are bugging, right? There's a ton of bugs, there's a ton of things that can go wrong, and you need to be able to diagnose the issue, at least from afar, to be able to give proper guidance. And only then can you then say, okay, we're going to go build this. Hey, Joe, you build that. And so from my perspective, I think that it's critical for the CFO to have their hands really dirty in this and really understand it at an intimate level.

Speaker A: I must admit that I've been doing some stuff in this area as well. We started building a very small application in grow cfo. The idea initially was, oh, wouldn't it be nice if somebody we were training on a program could go and turn their camera on, present to the camera, and the AI could come back and give them a score and tell them how well they'd presented. Great idea. And it started getting bigger and bigger and bigger to, well, actually, we've got all the tools here. We could completely rewrite the learning platform. So we did.

Speaker B: It's a beautiful time to be a builder and to be a tinkerer and to be a thinker. It's a beautiful time because a lot of the power is back in our hands, at least to think through that solution. So for us we've really embraced it.

Speaker A: Getting some of the more detailed bits right, though, uh, I found the problem is as soon as you come across an API, the system seems to have a problem figuring out how to use the API to the other system properly. It's great when it's just coding stuff in its own system and it can read all its code, it's already written, but it keeps hallucinating about how these interfaces are supposed to work, which is, as you say, things are buggy, they need a lot of testing.

Speaker B: Yeah. My personal view is that I see this as a today problem. I think that the future, quite frankly is going to be in MCPS and APIs being the broad distribution of information for CFOs and then you get to create your custom dashboard, your custom erp, your custom everything. And I can see that in the next two years that it's just going to get so much easier to do. Right now I'm developing my custom dashboard because you either have dashboards or spreadsheets. And a lot of times I'm comparing dashboards to spreadsheets and I literally am building a combination of a, uh, dashboard product that has all the graphs that I need for all the things that I look at. I have this thing called Morning Coffee, which is my morning dashboard. And I'm also implementing a way that I can bring in Google Sheets stuff so that I can see whatever my FP and a team is doing in one sort of pane of glass, which I think is really the future. And it's going to just make me so much more productive and I'm excited about that.

Speaker A: Yeah, that uh, is going to be fantastic. And I think, yeah, it is dashboards or spreadsheets, I think we're still living in that, uh, in between world at the moment. Our new system, one of the things I'm responsible for is all of the CPE reporting that goes into NASBA to confirm that we've issued certificates and so on. I've got a wonderful new dashboard built to do that. NASBA still want the information in an Excel spreadsheet though. There's still a long way to go. So you're doing all of this at CFO level and I get that bit. About what? Well, you need to understand how it works. There's the governance angle and so on. But what about bringing the rest of your team up to speed? How are you addressing that issue?

Speaker B: Well, at Envoy. We have a pretty much company wide mandate slash initiative on AI across the board. And it definitely starts with the finance team. And right now, so what we're doing is all of our processes, we use a platform called Scribe. I'm not sure if you're familiar with Scribe. Scribe is basically an AI tool that records your workflows, sort of like step by step. Every single process that we have has been scribed by my team. And the team and I, we sit and look at those scribes to see like, where is there an opportunity for automation? And we have a monthly meeting with the team to talk about what have we automated. And it doesn't need to be full applications, like what I'm building. But I'm saying, like, what skills are we adding to Claude to facilitate work? I'll give you a great example that my payroll person did, which I was amazed because this was an AI reluctant person and she leaned in and one of the things that we do is we pay health insurance in America. Everybody gets a deduction out of their paycheck and then we have an insurance company that sends us a PDF with a statement. But you almost could never actually reconcile. Like have has everything been recorded on both sides at the same time? Right. Uh, you have like this running liability account that you hope is correct and you reconcile it every quarter or whatever. And it's tedious and painful. And we took that from tedious and painful to. She built a skill where now she's feeding both sides of the equation, feeding the payroll information in detail deductions and feeding the PDFs from the companies, I should say. And now the system just creates a reconciliation that we were able to identify $40,000 that we were owed. Then we sent that to our broker and she didn't have to do much. Right. All she had to do was like oversee it. So we're looking for opportunities like that where things like reconciliations that took a long time before can now just be done automatically with the machine. So we're looking for all those opportunities for time saving in finance specifically. My approach is what can save us time because our work is so repetitive. Meaning, like we do it every month, we do it every quarter. And I say, well, if I can save X amount of hours a month, what does that equate to on an annual basis? And it's turning out to be a lot of time.

Speaker A: Indeed. So we started on the premise that AI isn't producing the ROI that it should do. That sounds as though AI is producing a lot of roi.

Speaker B: Well, I think AI is producing some roi, like I said earlier, for the people that are identifying the problem with the first principles mentality, I think where AI is not creating an ROI is in the teams and companies are looking for full blown solutions and then they're deploying these, whether it's products or internally developed stuff that's just not, it's a little too ambitious. And I think that's where I think the ROI is not really in play. Also, I don't think everyone is an AI capable, I will say person. So identifying the people that can build is the other thing that we've had to learn and quite frankly are continuing to learn. So I think there's plenty of opportunities for roi. I think right now though, a lot of companies are not doing it well and partly as I speak to my peers and other CFOs is because the mandates are pretty strict coming from the top, right? So the IT department is dictating what you can use and what you can't use. And um, one of the things that I've learned is that there are different tools right now available and you got to use the right tool for the right job. You can't use a hammer when you need a screw. And if the IT department is saying just use hammers, well, you're just going to lose a lot of time, a lot of effort and there's not going to be any roi.

Speaker A: Does the IT department get in the way when you're doing this? There's always a new AI around, there's always a different tool coming along that in theory you can put into place quite quickly. Are, uh, other rules that you're sensing coming out of the IT department that are really restricting your flexibility in doing things.

Speaker B: So there were in the beginning, one of the things that we've done, and I actually lead the AI Council internally here, and one of the things that the AI Council has done is create clear policies for AI use. We have a, uh, data governance matrix that basically guides everyone to what data they could use in the different products or use cases, if you will. Like what could you upload to an AI LLM just for querying? What can you use in an internally developed software that you're creating? What can you use if you're doing something external in terms of feeding it to the AI? So I think for us it's taken again the proactive approach of making sure that we get in front of IT in order to make sure that we're doing it properly. And then our security team is part of that AI Council to Make sure that we have proper security when things are being internally developed.

Speaker A: Yeah, but that's opening a whole can of worms around AI governance. Big challenge. What do you think some of the key things are that we've got to do to make sure we're governing AI properly?

Speaker B: I think clear frameworks and I think back to your question on the IT creating. I think making sure that everyone is on the same page as to what is allowed and what isn't allowed. So, for example, for us, back to the governance thing, you're allowed to do whatever you want so long as you don't use our data. But the minute that you're asking for any of our data, it has to be in some of our sanctioned sort of channels and ways to develop. And so we don't want to tort creativity. We want people to be creative. We want people to be using tools. We want people to be showing us what new tools are out there. We just don't want them using our data or our customers data. That's a no. No. But if you're building something internally, again, the team is reviewing a bit of the work that's being done and then afterwards we're defining like, what's good use and what's not good use. And look, I won't act. We have it all figured out. There's a lot of things that we have right now back to the ROI inefficiency. There's a lot of, in some teams, duplication of work, meaning multiple people are building the same thing because they can. We, we're still trying to get our process internally to figure out, well, when do we say which one is the best? And then we apply that as a company. But our approach right now has been, hey, let's just let people be creative. These things are going to come into play over the next month, two months, three months. Let's not stop people from going out there and learning.

Speaker A: I think that's something we've got to be very mindful of at the moment. This is early days and we have got to be creative. And, um, I don't think anybody's got it right yet. We're all learning. One of the great things about doing a podcast like this with you is I think you're a little bit further down the learning curve than a lot of other folk. And there's some great stuff that we can share and it's all learning from each other, which is fantastic. So in terms of day to day use yourself, what are you tending to do with AI? I know, I know you're vibe coding stuff. How does it affect on top of that your day to day job as a cfo?

Speaker B: Uh, it's making me a lot more productive. I have a couple of skills that I've set up in Claude cowork and also scheduled tasks. So for example my interview process, I interview a lot of the senior people here. My interview preparation and debrief is pretty much automated. I have a skill that comes creates my pre briefing document and then what I do is I record my interviews with granola and then I feed the granola back to Clark Cowork for that same interview and it knows who I interview, whatever. And then it gives me a post briefing document which again identifies some things that happened in that interview, some things that I might have missed, things that were covered, and I think from the perspective of interviewing it's made me exponentially better. I have another scheduled task that runs in the morning that basically looks at my previous day's emails, looks at my previous day slack, looks at my previous day meeting notes for my granola, and then says hey, here are the tasks that came out of that. Here's what you missed, here's the urgent things that you might have missed in your emails, here's the things you gotta respond to. It also like highlights specific senior leaders or the board that if I ever get an email from them it, it flags it for me. And so that's the first thing that I see in the morning. Right now I have another skill that basically looks at the previous day's activity, whether it was customers acquired or customers return or anything that moved around. And it basically gives me summary of exactly what happened with a bit more detail. So I've been really using it and pushing my team to use it for augmentation. A lot of times as CFOs we get these emails with two lines and then a spreadsheet that you're supposed to click into and you're on the road like you're not going to be able to do that. Now everything comes to me with already a summary of that report and some insights that are built by AI, which just makes my review process just way more, more fluid and I could do a lot more on the go. So I'm using it extensively.

Speaker A: When you're asking it to read emails and delve into the spreadsheets that are attached to the emails, is it generally coming back with the right answer?

Speaker B: Yeah, by now because of the way that the MCPS are working, like it's really pulling in from the emails pretty directly. It's not like when ChatGPT first started. Now the connections and the connectors directly through whether you're using Gemini or whether you're using Claude or using Codex or whatever you're using, the connections are right through to the data. I've yet to see an, uh, error. If anything. What I found, Kevin, is that it finds things that I missed and that for some reason my email, I have this thing where like some of my email just gets automatically like archived because there's a bunch of filters and sometimes it filters the wrong thing. And what I found more often than not is that it highlighted an email that I just missed because it was improperly archived. And I'm like, oh my God, that was in my email. Rather than the other way around where it shows something that I'm like, okay, this is three on fire. And then I go and I'm like, oh, it was false alarm. Like I've yet to have that.

Speaker A: So thinking back down into the work, I know we mentioned getting the rest of the team used to using AI and um, rolling that skill set out. Do you think we're going to see headcount reduction coming along as a result of this or just people working better?

Speaker B: I think it really depends on the size of the team, to be honest. I think that in small companies like myself, it's less about the reduction in headcount because we run pretty lean as is and it's more about getting a lot more out of the people. I think in much bigger organizations where you have multiple people doing specific jobs, I think that there's definitely going to be job disruption. I think there's no question about that. If you have a hundred accountants, you're going to get down to 60, that's for sure. But if you have six, you can't really get down to four. Right. Like we're not there yet and we're not going to be there for a long time. What I'm trying to do is really get my team to be doing more value add work. And what I've been preaching to them is, hey, we're business. We're sort of like business owners now. We're business consultants, if you will. So it's less about can you reconcile this bank rec, which now the machines can just do automatically? And it's more like, can you identify what's happening in our bank and our cash flow in a more insightful, meaningful way than we've been able to do before? So I think it's going to be liberating. But I do think that Bigger teams are going to be reduced for sure.

Speaker A: In terms of core systems, do you think you're going to be changing core systems in the near future because of AI? Do you think you're stuck with any legacy things that are going to cause a real problem getting data out of and into?

Speaker B: I think companies are definitely going to struggle with getting rid of legacy systems that they have. But there are a ton of systems that we are just uprooting and just completely doing away with across the whole stack. Not just in finance, but in data. For example, we've displaced now like four, five tools, saving us hundreds of thousands of dollars because we just realized we can build it ourselves in accounting. I build a flowcast application for us where that's what we use not to close the books so we no longer need to even think about spending the money for flowcast because we have a pretty good, pretty robust closing application that does everything that FLOQAST did for us and Densom. We're displacing things like Asana. If you're in Asana, Monday.com and some other applications, I think this is going to be hard for you to hold ground because companies can build project management applications fairly simple. Other than that, I think for us we're looking at every tool and saying can we displace it? Because the thing about it is all those tools are great, but they do way more than your company needs. You can build a tool that does 75% of what they do, but that's 100% of what you need and is fraction of the cost.

Speaker A: And there's always something with those tools as well that though they do maybe 150% of what you actually need, there's that little 5% that you would like that they don't do at all.

Speaker B: That is exactly it. You can really customize it to do what you need. For example, another application that I built that we use is I built a company wide OKR project management platform and I sat down with folks and I enjoy this and I was like, okay, what makes this the best and what things do we wish we had in the other products that we can add here? And it's pretty remarkable how we can get better software.

Speaker A: One that I've looked at and in here in grow cfo, I'm head of tech partnering so I'm dealing with a lot of fintechs all the time and there's probably 20, 25 companies we regularly partner with now. Keeping up to date with what's going on in all of those is not easy. But I've got a very nice little routine now and I've coded that whenever we've got a new partner, does some research on them. But it repeats the research exercise on schedule once a week, throws me a report out and says what's changed since last week? Oh, Company X have just announced a fundraise. Brilliant, nice shorthand report. So I've got in front of me every week a very straightforward guide to what's going on out there in the fintech world. Which I never had out of our previous systems.

Speaker B: Exactly. And your augmentation, right, the ability for you to do your job is now exponential because you could have done the Google searches and the Google updates or the Google alerts or whatever, but that took work and now it's just.

Speaker A: And I never did because that took hours. Even if the Google alerts were there, it took hours to go and read them and remember to read them and so on. So it's that lovely thing, and you mentioned it earlier, about just having the information in one simple, easy to digest, uh place the uh, your examples, the email summaries and so on. I think that's a fantastically powerful thing that you can now pretty much design your own personalized dashboard that has on it whatever it is that you want to have on it. Not what the project management system thinks is important, not what the time management system thinks is important. It's personalized to you and the limits of it are only your imagination.

Speaker B: Yeah. And I think, look, one last thing, that last cool use case that at least for me. Back to the augmentation. We actually have a board meeting this week and we create a pretty in depth pre read deck that we sent to the board. And it's interesting how much the human eye can miss because I put it through, I have like this consultant reviewer of decks and I ask it to, hey, make sure that everything foots, make sure that throughout this long deck everything is consistent. And it's funny because the things that I highlighted, some of it were small but now the deck is that much more accurate, right? And so things like hey, you said 4.78 here, but in this other slide you said 4.76. It's like, okay, which one is it? So I can go back and say, hey guys, we're quoting two different numbers just in two very different parts of this deck. And before, I mean it would have been really hard to actually find that because it was kind of a needle in the haystack. Error. Back to the augmented cfo. My board deck is way more Accurate. I'm a much better cfo.

Speaker A: And I had an interesting one from another CFO I interviewed where taking that board deck a stage further, and he was feeding the board deck to a virtual board within the AI. He'd got Personas set up of all his board members and he'd give them, particularly the finance part of the board deck, and he'd say, okay, how is the board going to react to this? And the AI would come back and give perspectives from some of the individual board members that were likely to come up in the meeting. And, uh, that was triggering him to think about, well, if that question comes up, how do I prepare for it? I thought was a very neat application.

Speaker B: Yeah, I think that's great, actually. I think I'm going to try to copy that.

Speaker A: Yeah, yeah, yeah. He was having a huge amount of success with it. And I went like with a little bit of the concept afterwards. I'd always had this idea of your own sort of mental internal board that you chucked ideas at. They were sort of reviewing things and I had sort of half a dozen characters on my board and I went and got ChatGPT to go and research the characters that would have been there and form profiles. So can I ask questions of these people and how do they respond? And, um, providing it found a decent online presence for the particular Persona that you were putting on your fictional board, it worked quite well. And again, it's coming down to the limits of this stuff is your imagination, which I think is fantastic. Where do you think we're going to be 12 months from now?

Speaker B: I think that 12 months from now there's going to be a lot more data availability for that customization. And we're starting to see even Salesforce now. They're going the MCP route. Salesforce knows is their software's not great. The power is in their data. And now that people can build, I think that the access to that data is going to be sort of their real asset. And So I think 12 months from now there's going to be a lot more in terms of banking accessibility to data. You could see it now even like ChatGPT and perplexity just are allowing individuals to connect through plaid to get their financial information directly in there and get all kind of reports, which is amazing because now I don't have to go to my bank to even see what's happening. I think in 12 months we're going to see a lot more of that in the finance world, where banks are just going to facilitate the data in a way. That you're going to be able to do a lot more. And quite frankly, I think it's going to be an amazing time for us because we won't have to deal with the tediousness of bank reconciliations and all that stuff. It's going to be way automated. And I think that's going to just give us a, uh, further superpower and where we put our time. So 12 months. I'm pretty bullish on that.

Speaker A: Interesting. Sonoe, that has been an absolutely fascinating conversation. Thank you hugely for being this week's guest on the Grow CFO show.

Speaker B: Oh, uh, thank you for having me, Kevin. I love the topic and I think it's a great time for us CFOs to be leaning into this and see what the future holds.