← AI to ROI
AI to ROI
AI-Native ERP vs. Legacy ERP: What's the Difference? with Santiago Nestares, Founder & CEO of DualEntry
00:29:33
Loading summary
Transcript33 lines
- [00:00]
A
Foreign.
- [00:08]
B
Welcome to today's episode of the AI to ROI podcast. Yes, formerly known as the Metrics Measure up podcast. Today I am joined by Santiago Nestaros. He is the founder and CEO of Dual Entry. And I'll be covering four topics with Santi today. Number one, traditional ERP versus AI native erp. What is the difference? How to approach ERP migrations, Where does the ROI actually reside in an AI native solution? And a little bit about the auditability of financial decisions driven by AI. So with that, Santi, I hope it's okay that I call you Santi. Can you give a brief background of your journey to becoming my guest here on the AI to ROI podcast?
- [00:57]
A
Thank you, Ray. I appreciate you having me. And yes, Santi, sounds great. I always say that our journey for dual entry actually might have started before dual entry without us knowing, because in our last company, we were using a legacy ERP that ended up being a catastrophe. And we jokingly said at the time, if we ever start a new company, it's going to be one to build a modern version of this. And ChatGPT3 still hadn't launched at the time, but there's certainly a need for something modern and intuitive. We were stuck in this implementation for over nine months. We had a team of 12 people in finance that we had to parachute into this implementation. You can imagine hundreds of thousands in costs. And there was a joke internally that if you do like an ACV to NPS ratio in our software exposure, it would be the highest. We were spending 10 times more in ACV for legacy ERP than the next software spend that we had. Yet it was the lowest NPS score in the entire stack. So the ratio was through the roof. And we jokingly said, if we ever build a new company, it's going to be a modern version of this. And here we are.
- [02:05]
B
There's a new metric for me, the ACV to NPS ratio. I'll have to think about that one.
- [02:11]
A
Okay, there you go.
- [02:12]
B
So there's a lot of talk out there about systems of record. Now, by the way, nothing is more of a legacy system of record than erp. Right, the legacy erp. So even the word erp, it freaks me out a little bit because a lot of what I'm seeing today is AI native platforms being more accounting than erp. But let me step back from that. Maybe you can address that later on today. But what differentiates an AI native ERP from a legacy ERP where they're already adding AI to their legacy ERP systems out there?
- [02:48]
A
Yeah, well, I mean, adding AI to a legacy system is like the equivalent of trying to run an on prem system with a CD on the cloud. It just never worked out because AI is not really a specific feature. You can't say this is ERP and this is ERP plus AI. Those are just, you know, it's not a feature that you just plug into it. Some people think Copilot is like the version of that feature, but it really isn't. Like Copilot is just another interaction layer. AI is really the use of these magnificent models that we stumbled upon as humanity to make more gooey, more gooey, less deterministic guesses about what the next step or what the next action from an accountant should be. So that shows up everywhere in the product. And to give you the silliest example is in dual entry. Every time you click on a dropdown, AI is already precomputing. What are the things have you chosen in that screen? And can I guess based on those things, what is the most likely dropdown choice for this dropdown? So that's like a small example. That's not a big feature. You're not going to advertise that on your website, but it's just everywhere it's embedded in it. In order to do that, you need to make sure that the data and the context is available. And I'm sure we can talk about that more later. But whether it is to run a report and to tell the system, hey, give me a report segmented by these departments, by these regions, that's an interaction layer. It could also be like it pre writes the flux analysis. So it's already doing a lot of the commentary work for you. It's not a single thing. It'd also be categorizing. Categorizing is probably the most obvious example, but it is. How do you rethink something? By the way, some UIs are even different in the A of AI. How do you rethink a system from the ground up that is already taken into consideration, that a non deterministic layer can give you the best suggestion and what guardrails you want to build around it, whether it's permissioning or approvals, so that you can ensure that not even a 1% transaction error rate can make its way through the system?
- [04:42]
B
Santi, let's. There's these two terms and you've thrown out one of them and that's deterministic. And people talk about probabilistic, right? And a lot of the people listening here might know the difference between the two. But can you talk a little bit about moving, you know, deterministic versus probabilistic?
- [05:00]
A
Yeah. So, you know, deterministic is when you have a mathematical equation algorithm, a computer system, call it whatever you want something in the middle. And no matter how many times you give an input, the output is always going to be the same. And you can do that a billion times, a trillion times. Always going to be the same output. Probabilistic. It might be almost always the same output, but not always the same output. And that almost makes a big difference when it comes to accounting, because you can't get accounting right almost always. You have to get accounting right almost always. Always. Now, if I were to tell you and saying that's probabilistic. So there's a probability involved. And most of the AI models are probabilistic, which is why we have to create guardrails around them in order to make sure that we get it right every single time. Now, if I were to come to you and say, Ray, what if I had someone, an accountant in your team that is going to get things right 99.9% of the time, not 100, but 99.9, you'd probably first say, well, that's anyway better than the. Than what I'm getting it right today. And then number two, that's fine as long as I get to supervise it and create guardrails. If the transactions are above this number, then I need to approve them. If they're not, then maybe I don't need to approve them. So that's the job of a quote unquote rigid ERP behind the scenes to make sure that we create those deterministic guardrails around the model.
- [06:13]
B
So let me ask this because once again, I have relationships with all of the ERP vendors. I had a solution showcase day where six of them actually demonstrated their software to our audience. Santia, if you were to say hey to the listening audience, here's three or four key ways to identify why, you know, if it's truly an AI. First platform versus an AI add on platform, kind of. What's the three to four things you'd say? Ask these questions.
- [06:44]
A
That's an excellent question, Ray. And it's going to totally date this because I think the answer for this is going to be totally different in a year, even six months from now. So I'm hesitant to tell you. Like, you know, they have these five things that are, you know, today the thing is agents, but maybe tomorrow is something else. I think that the biggest indicator is how young is the team and the DNA of the company? And then that's not necessarily, you know, a commentary on the age of the team, but it's rather like what is their mindset? How you know, if you've been running ERP a certain way for 20 years and you build these legacy systems, most likely you're an old databases, you have old architectures, you don't have the mindset to try new things because you have so many clients that if you were to try something new, it might piss off 99% of your people. So there's a lot of things that come from being a legacy, which makes it an innovator's dilemma for why you actually can go and change things. The perfect example is we believed first that in the next state European migration was a key element to making a modern day AI first experience. But I was wondering, like some of these legacy players, they actually can do that because if they do that, they lose a big portion of their services revenue which they make by keeping you in this consulting stage of implementation. So the first indicator I would say is like, how does it feel? How does the team and what has the trajectory been? How quickly have they gotten to where they are? Because you're not going to change your ERP next year, you're not going to in six months. If there's a better vendor, you're going to go and swap it out. This is a long term relationship that you're setting yourself up for. And the best indicator of where this relationship is going to end up or how good this product is going to be is how quickly have they gone here and how quickly they've gone at adopting what the latest technologies are. As an example, in dual entry we have a benchmark every time. We just launched ChatGPT 5.4. We ran it through the benchmark and we noticed almost a 10% uplift from the prior model. And then we immediately run a quick test and incorporate it into our system so that our customers are getting the latest models behind all these AI features across the product. You're not getting that from some of the legacy people. That level of curiosity and avant garde that comes from it, you're not getting from these slower moving teams.
- [08:55]
B
It's funny that you talked about the newness of the organization. There's this famous old saying, I know it when I see it with all the companies I'm working with and go visit. If it's a more traditional SaaS company that's trying to become an AI first versus an AI native software company, I just feel it when I walk into the office, when I talk people. And sometimes it's hard to clearly define it, but there's this energy and thought process. It's just different. In fact, in your category, a lot of people have been talking about one day close, zero day clothes. But when I was doing my research and talking to you initially, you have another kind of new phrase, and that's next day migration. Because people are really looking at, man, this has been my system of record for 15 years. I've got so much legacy knowledge, my data model has been extended. So tell us a little bit about how you use AI to accelerate your migrations.
- [09:58]
A
Yeah, so it's funny that you mentioned that that was our phrase because we were sitting down with our marketing team and we're like, we actually don't have a zero day or one day close phrase because we don't have the concept of closing. And this permeates in this little things like in the legacy ERP systems, at the end of the month, you have to run the inventory simulation, you have to run the eliminations. It's actually a background task that goes in and adjusts the elimination entries. And if by any chance you or somebody in your team changes a transaction and forgets to rerun that, the eliminations now don't make any sense and don't match the correct transaction. We don't have any of that. Our transactions are atomic. When you edit a transaction, all the eliminations that are pertaining to that transaction get atomically means all as one unit edited. So it's all these little things that allow for our customers to run a continuous close. Because even if you close, quote unquote, in one day or zero days, the month is 30 days long. And it if something happened the first or the second day of the month, you're still waiting now another 28 days to be able to get the answer from your books. So that to us didn't make sense. And by the way, you see this because companies then go ahead and build data pipelines in parallel and they have all these dashboards. And now that with vibe coding, it's so easy to do these dashboards. And it's like. But these are just crutches to the fact that your accounting, which is your true and accurate data, is not continuously closing and giving you a continuous insight. So we didn't have that zero or one day close phrase and we had to come up with some other sexy part of our process. And that's how we came up with next day migration. And look, implementation Takes a little bit longer than that because we involve humans and we guarantee the results and we help you. There's also some partnership part of that that is like we're trying to give you the best advice for this new stage of your company. So it goes beyond just the migration. But we identified that especially in our experience, the riskiest part of your whole implementation, going from a starter cloud or a legacy to your new AI native ERP system, is that the migration takes a very long time and it's error prone and it's manual. And most importantly, most legacy people, they do it at the trial balance level, so they actually leave behind all the context. So the beauty of AI is it can take context and use that to extrapolate what the next step or what the next categorization should be. Well, most of these legacy players are leaving behind all that context in the last system and just bringing into our balances. We felt that is not a good starting ingredient for an AI native erp. So we want to fix that. First, every transaction gets migrated. The second thing is it needs to be done quickly because when we're going up against legacy systems, we are getting compared to people that have brands, so they have boards, and the boards say, look, nobody got fired for hiring IBM. Why wouldn't we not go for legacy? So we have to earn our right to win the trust and say, look, we're actually going to migrate it to you during the demo and give you access to a sandbox with your live data, which by the way keeps on feeding as your team keeps working in the old system so that you can run reports, show it to your management team and say, look, this is what dual entry feels and test out every edge case things that you might not even think about it beforehand during the demo on your own time, you can test out all these things. So this was how next day migration was born. Now, a misconception is that we're using AI for the migration itself. We do use AI for a lot of the mappings because you're coming maybe from multiple GLS into one, you get a map, classifications, departments, all those mappings. We do a first draft with AI. When we do the implementation for you, we guarantee those outcomes. So we have a human check them and you can also check them yourself and only when you click you're good to go. Then the data starts coming in. Programmatically, that part is fully deterministic, meaning there's no AI in how we migrate those transactions. It actually took us a long time and it was very meticulous of our engineerings had to understand all the origin systems and say hey, when this case shows up, this is how we're going to treat it in dual entry. Because it's not like starter cloud says, hey, this is our code base looks like. So we had to trial and error it. And today we have the seamless migration engine that oftentimes in a few hours your data is live in dual entry.
- [13:54]
B
Let's talk a little bit about that migration resourcing because I've talked to some companies like, hey, our implementation practitioners are truly finance and accounting practitioners and then we hear a lot about the whole forward deployed engineer. So if there's some things we need to do on a customer specific implementation, we can try to get it back into the core product right away. How do you balance the type of resources on your implementation and migration team? Santi?
- [14:22]
A
So because we were so efficient on the tedious part, which is the data migration, which by the way most people don't know, there's actually two data migrations. There's the initial data migration and then you have to migrate the stub which is the data that came in while you were migrating the first part of the data. So there's actually two migrations that most legacy ERPs have to do. So if you save up all the time and you reduce the risk in there, you actually now have tons of time and budget to spend time with your customers saying, what do you actually want to solve with this new face of your company? We set up rules, sometimes we even advise on what are the best practices for GL accounts. Because we see companies that have very long jail accounts and actually what they want is to use some new dimensions or segments and reduce the amount of accounts to keep things more simple and manageable so we can play that partnership role. And then oftentimes when we're playing this partnership role, we realize, hey, I have this very quirky thing about our system and I'd love to solve this because this might save me 20 hours a week. And that's when we take that feedback quickly and incorporate it into our product team to get it released quickly. And we felt that these legacy systems, because they're so detached, there's an implementation firm that goes to an implementation partner program. It's so far apart that that product feedback never happens. And by the way, they're not incentivized. Why would the consulting firm tell NetSuite hey or sorry, the legacy system tell them, hey, this is how you can improve product so that I can bill less hours. That makes no sense. They want to keep billing. So we want to align the incentives. We want to say, look, we're not charging you for implementation. Our job is to get you live and happy and we'll make money if you stay in dual entry for many, many years. Otherwise, we're not here to keep you in this loop.
- [16:07]
B
So what is your average time to migrate and Deploy? Are you 30 days, 60 days?
- [16:14]
A
It depends on a few things. It depends which system you're coming from, which integrations, because we like to be involved also in the integration transition. We like to unplug it from the old system, plug it into the new one. In some very simple cases, when coming from starter cloud, we've gotten people up and running in one or two days. It's very quick. In some cases it might take two to three months. I would say the average by most customers get up and running fully implemented in less than four weeks. And internally we track that and our goal is to get it down to one week. But we're not, we're not there yet because we'd rather spend a little bit more time than necessary to make sure you're successful. We'll also stay with you to the first close. So we'll do the first close together. Make sure you're familiar with all the tool. So all that is incorporated in this four week timeline.
- [17:00]
B
Gotcha. Okay, so you are on the AI to ROI podcast and you're selling to the executive who probably cares more about return on investment than any other executive, the chief financial officer. So are there some specific finance workflows that you can really point to and say, this is really measurable ROI that our clients are reporting from using dual entry.
- [17:25]
A
Yeah, I mean, by the way, like, next day migration is the first ROI they'll ever see because they'll see like, okay, we're up and running quickly. We didn't have to set everything aside for six months to do this migration. So that's already a huge win on day zero. Look, the way I describe it is like companies do things and then they account for things. Like they journal what they did, the journal entry comes from that. And what you want is to make so that from the moment that the action is taken, there's the least amount of touch points until it's already accounted for. And it happens in the least amount of time. And that generally happens from two data sources. Either you're getting it from the financial institution that might be like the bank feed or the credit card feed, or you're getting it from an external integration. Maybe you use Brex, ramp slash, whatever, one of those. So the first thing, and it sounds very obvious is like we want to make sure that we're integrated in real time with these tools. You'd be surprised how many banks and how many financial institutions legacy erps are not integrated with. So already by doing that, you're already allowing the customer to do an ongoing close as opposed to waiting for the month and to upload their CSV transaction. CSV with all these. It sounds obvious and I hate to say it sounds so mundane, but actually saves people so much time. Then once you do that, you can already start creating rules or using AI to get from the data gets into the data is accounted for in as few steps as possible. So sometimes rules are the best way to do this is if it's deterministic. But there's always curve balls. For example, in a transaction, something's happened one off. Or you might have something like Uber late at night where it might mean that Uber might be. It might be that you took transportation home. So that would be the GL account transportation. Or it might be that people ordered meals to the office through Uber Eats. So that might be meals. Those are different GL codes. Although it might sound and look the same in the bank feed. Well, AI is exceptionally good at these cases because it knows, ok, usually they first order the meal and then they take the Uber home. So that's usually a clue. Sometimes there's clues in the bank statement line itself. So it's like, can AI now account for this so that I don't have to? And then usually just leave the approval to the accountants where they say, look, AI got it right, I'm just going to approve it. So now you've gone from the thing happened to the thing is accounted for in as few steps in as little time as possible. So you have real time information and you've saved the accounting team tons of time. We're seeing most of my there. The second area is like, okay, now that everything's accounted for is how do I actually extract the insight? Because accounting is not just like, we're not just doing it for the fun of it. We're doing it because we want to understand our business better and then direct it in the right place. That's really where the biggest ROI comes. Like, yeah, you save costs. Like cost saving only going to get you so far. If you determine that certain region or a certain sales or a certain skew is performing very well, you can say, hey, let's double down on that. Let's put more AES or spend more marketing dollars towards that SKU. Or if you determine, look, this territory is not doing very well, maybe we should cut our losses. All those things are actually going to drive meaningful order of magnitude changes to your business that are much larger than perhaps just a few hundred thousand here and there that you're going to cost. So we'll do both things for you. But the ultimate goal is to get the account to the second bucket and extract the insights so that they can take it to management and say, and by the way, it used to be that you only segmented your books in two or three different ways because it takes time to categorize things. But with AI now you can segment in eight different ways or nine different ways. We have no limit in categorizations or segmentations. And then you can also use AI to mix and match these because if you have nine, it's nine factorial times the amount of insights you can get. So if you actually start matching, okay, this region with this skew or this territory with this salesperson, you start matching all this department. If you start matching all these things, AI can do this for you. Create matrices and then say, by the way, we found these pockets of values and these pockets of risks or areas to cut losses. That's the ultimate point. ERP is just the means to the end. The end is to direct the business in the correct direction.
- [21:12]
B
You know, a lot of times always had to do hypothesis of. I think there's some correlation between this one field or one variable and this outcome variable. Right. So with dual entry, can you actually use some level of automated correlation or causal algorithm to say, hey, we found this unknown relationship between this data point and that data point?
- [21:33]
A
Yeah, I mean, causation is like a whole field of science that in math we can get into. But, but I think the beauty that the GL has is you get the full picture because you have all these expense management tools and you can see whichever you're spending more on. But the word ROI means you're getting the expenses and you're determining what the return is. So you need to see the full picture. What we are really good at is segmenting that picture and saying, I have a hypothesis that this salesperson is really good at selling this SKU in this region. For some reason you can throw those three filters or matrices and then see that cell and say, actually the full roi. We only spent this much money in marketing, this much money in advance for that region and in that sku and we actually got this many in sales so it's actually like a great roi. We should spend more money there where you would have never been able to see that picture if you were only seeing revenue or you were only seeing expenses. You need to see the full picture to get the ROI calculation. And before you would have never categorized so many things because. Because it was just so time time consuming.
- [22:33]
B
Yeah. You know, Sanjay, I can't believe this. We've already 25 minutes into our conversation, so we have five minutes left. And I have so many more things I wanted to ask you. But let me ask you this one because I'd lose my AI podcaster license if I didn't talk about context. Context graphs and AI explainability. So just tell me a little bit about how you ensure that there's real context and explainability audit trails in your software.
- [22:59]
A
Ray, I love this question because when we were starting dual entry, the first things we built was like permissioning layer so every user had very clear permissions and approval workflow builder so you could visually create an approval flow and then an audit trail. And some of our investors were like, you guys are an AI erp. Why the hell are you spending time on all this stuff? Go build some AI stuff. And I say, well actually the thing about AI is that it's a black box and it's non deterministic and you can't explain it. You really don't know why GPT or Gemini or Claude responds to you in a specific way. There's no way to know. They don't even know. So it's like you can't go to an auditor and say, hey, look, I just did this this way because AI did it. You can't blame AI. So for us it was important to create all these permissioning layers and these approval layers where you can say, and then let the AI roam for free. Look, categorize the transaction, draft the journal entry, I don't care, because it's going to be in draft state and then it's going to be routed through the approval flow, almost like a very junior accountant would have done it. And then the more senior accounting team can now actually verify and decide, yes, we want to approve this or no we don't want to approve this. And most importantly, you're actually now giving me feedback as a software because you're saying if you approved it on the first go, that means that whatever prompting or context I gave for that transaction or was really good, but if you correct it and say no, this actually was not correct, you're now giving me information. How should I do it next time so that I make sure it's more accurate. So we don't actually have to explain AI because everything gets traced back to a human, whether it's an approval flow or a rule that was created. And it's all right there in front of you. You can see AI suggested X, human changed X for Y and then ultimately approved it for Z. It's all there. So it was a bit of a counter approach that we took. We went back a few years into deterministic land to then be able to push forward into AI first land.
- [24:47]
B
Got you. Okay. For the last couple minutes, I want the audience to get to know you a little bit on a personal basis. So I'm going to do three rapid fire questions and one is very germane. What are two or three of the variables that's key to a CFO ensuring that he or she gets ROI from their AI software investments? Not just do entry, but AI software investments overall?
- [25:12]
A
As long as you can say it's AI, you're all good. No, I'm kidding. Look, does it save you time? That's the biggest ROI human cost or the biggest cost of every business. Does it actually save you time? A good proxy that I always ask our customers is do you feel you're doing more intellectually challenging work? Because you'd be surprised how many people are just doing mundane, repetitive tasks that are not intellectually challenging and people don't actually like those. That makes your time go by slower. It's not cool. And it usually is a sign that you're not working to your fullest potential and you're not getting the most ROI out of yourself and contributing the most that you can. So it's like, that's just a good proxy, but it's ultimately that I save you time out of your Monday tasks so you can spend time on the high value strategic tasks. And I think in the erp, what most legacy people are missing, or I think generally is like, we've been so caught up in the concept of closing the books and getting the data in and getting it done and checking the box, that it's like, checking the box is not the point. That is the means to the end. The point is to understand your business so that you can actually look forward and make a better decision. So the best ROI is like, did you actually make a meaningful decision for your business? Did the company change course because of something that you got from your AI tool? Like, are you doing something now? Much better, better roi. And I think people miss that. And that's actually the whole point of why we exist.
- [26:31]
B
Got you. Last question. I know you sell to the Office of Finance cfo, but every executive is looking at investing in AI software. Who owns Measuring the return on investment for AI software investment. Is it just a cfo?
- [26:45]
A
I mean, it's the company, right? Like company profitability is the company's goal. If companies were AI models, that would be the reward function. You want companies to grow more and make more money. So I think everyone in the company is incentivized to say, saying, how do I do more with less? And usually we're living in this beautiful revolution. It's like sometimes doing more with less is investing in AI and having these wonderful softwares save you time so that you can scale much further. We have companies that are $300 million of revenue now with two people in finance. So I think it's everyone's responsibility to say, how are my budget, my resources allocated the most effective? Because if the company does better, usually everyone else does better. So.
- [27:26]
B
Okay, last question. There's a lot of angst out there for early career people. Hey, I started as an FP and a associate, right? So what recommendation do you give these? Just recently graduated or you know, maybe. I've been in Corporate America for 612 months. What's your recommendation right now regarding them and AI?
- [27:46]
A
How freaking lucky are you? You're living the biggest positive revolution that's happened in the last two decades. Lean into that uncertainty and lean into that fear because these changes are good. It's leveling the playing field. It used to be that expertise was held by a few people, so you had to learn it from someone. Now expertise is universally available. You don't have to learn 606. You can just ask dual entry or ask GPT. Explain to me how this transaction would fall into 606. So expertise is now universally available. You're instantly upgraded to a more senior account off the get go. And then the second thing is like you're now 10 times more valuable to any company. It used to be that companies couldn't hire accountants, so they had to wait until they became mid market to hire their own internal accounting firms. Now they can, because you're 10 times more valuable. You're going to learn these tools 10 times faster than other people that are used to their own ways. So I think if you're an early person and finance or accounting is like, you're living the luckiest time of your life. And lean into it, go capture it, learn the latest tools, stay up to date because you're going to be 10 times more valuable.
- [28:49]
B
Santiago Nastaros, founder and CEO of Dole Entry, thank you so much for being my guest here on the AI to Roi podcast.
- [28:57]
A
Thank you, Ray. It was so fun. Thanks for having me.
- [28:59]
B
Okay, and to our listening audience, if you're enjoying these discussions we have with amazing founders, entrepreneurs and practitioners who are deploying AI and their company, it mean the world to us to go ahead and subscribe to the AI RI podcast. Go ahead and give us that five star rating. And hey DM me and let me know who else you'd love to see on the podcast.
- [29:16]
A
Thanks, Santi thank you Ray.