So I have, like, a general question. What prompted you to, like, look around, look for a solution and work with us? Right. I think one of it was essentially I and Imperial Treasure... I mean, I was working with Denville when he first came in as the CFO so I've been involved in Imperial Treasure for quite a while,

right? From the PE ops standpoint. So me and him kind of put in the UI Path part of it, but that was obviously pre-AI, right? I think when we started, it was probably about 30 grand or something like that, and it has kind of doubled in probably, I don't know, two, three years. I think that's kind of how UI Path is playing the thing,

right? So what prompted me was essentially, okay in this day and age of the whole AI part of it, was there a solution that probably would be cheaper and a little bit more... How shall I say? They could self do it in that sense, right? So I was thinking, "Okay, there should be something." I was looking around Twitter

and stuff like that, and then somehow other, I think you guys launched or you tweeted and stuff like that. Somehow other it came onto my feed and I was like, "Okay, this looks interesting." That was why I think I reached out to you guys. Probably this was what? This is probably six, nine months ago or something like that.

And then it kind of all went... I kind of went silent for a while, I was talking to Denville on the sidelines throughout the months and essentially we were like, okay, there was a CFO conference that we organized, that Nevis had organized probably, what is that? Maybe about four, five months ago or something like that?

So essentially, I think I got to talk to him and kind of suggested, "Hey, you know there could be something in this Mediar platform. Potentially, you could get out of UI Path," and then you know... So I reached out again to you guys, like, what? Four months ago or something like that.

It was more that I think AI was going to be able to do this, 'cause I was playing around with browser-based and I think around that time Claude came up with the whole computer use agent, right? So I was thinking, "Oh, wow, this is possible." That's why I think I started looking.

Because I think when Claude launched that whole computer use agent thing, it was like, "Oh, wow, AI could also do this." And then I was scrolling around, and then somehow I think I found your tweets and essentially checked you all out and made first contact.

So have you tried Claude computer use and browser-based for this specific use case? Right. I think the computer use agent thing, I was helping a colleague at that time to try to do it, right? But it was obviously a little bit more technical, so I was thinking, "Okay, I don't think this is gonna be something Imperial Treasure can use," right?

But in the meantime as well, I think I was trying browser-based, right? So I've got my own little company and essentially I was like, "Okay, let me try using the whole gen AI part in browser-based to actually create an invoice on my Xero account," right? And essentially it did.

So it kind of convinced me that, wow, this is possible. It looked very long. I think it took 10 minutes to create an invoice, but it looked possible. So then I was thinking there's bound to be startups or whatever it is that potentially are doing this, and I think a lot of them was using more the whole browser UI type of thing, right?

And then after when I came across you guys, it so happened that you guys are more specialized or actually the platform was able to do Windows applications. So it's like, oh, okay, that would be good for use cases across the Nevis portfolio company.

So initially it was Nevis portfolio companies. Then when I talked to Denville he seemed interested, so then we kind of started working more towards replacing this. And, I mean, it's been a quite good thing so far. I mean, they are now managing to save like, what, 80 grand a year, right? So that's tremendous actually.

Although I must say the obviously the original idea for using UiPath was it's gonna do many processes, right? But somehow other, I know it kind of got stuck onto just one, and obviously, you know, paying that amount for one process just doesn't make economic sense.

So I mean, good, and now it looks like Chai's kind of getting on this, Denville's looking at it, and I think your pricing model, whether it is that $1 per execution type of thing or a usage-based metering type of stuff, I think it would still be probably if you are below X number of transactions or workflows, it probably still makes sense.

Pay as you use. But I think at some point there will be, "If I pay this much, I get unlimited." Like in comparison to UiPath where you just pay for the license, right? So you pay a lot just for the license, but you can use as much as you can.

Every time you do something in the automation, you know how much you're gonna save in this specific automation. You can always calculate, is it worth taking on more usage, right? Because the alternative? Like, how much do I save?

So I think in this case they bundled up two semi and two autonomous bots. Obviously in the case of Imperial it probably was a bit too much. But unfortunately they only do bundle type of stuff.

So you mentioned that the original plan was to have more automation in UiPath, but it didn't materialize. What do you think was the key limitation factor about that? I think several. One is the resourcing. Chai and the tech/IT is okay, right? There are people there. But it was more in the finance department or in X/Y department, the resources there are not very techy.

We needed somebody in the functions, for example, in Denville's finance team, to be able to drive this, because it's actually not a tech problem. It's actually what process, which parts of it can be automated, which parts might not be able to be automated.

When we did this with UiPath, the initial reception was skeptical. People were like, "Are you sure the bot can do it?" That's why it took probably six months or something like that to even do the first process.

They didn't really want to do it themselves. So ultimately there was an SI that I went through, who was sort of reseller for UiPath and also does professional services there. How much did it cost to implement actual one single process? Probably about 20 grand Singapore dollars.

I think UiPath wasn't so bad in a way. It was relatively friendly, except that Chai and all that didn't really go into this. The Mediar project probably is a bit different because I kind of drove this, and Chai was more incentivized personally to get more involved.

AI-first tools like Mediar, you can make a lot more things with a smaller team. And also definitely I think the cost component here was good. The whole entire SaaS industry is probably feeling this.

When the whole AI thing came on, then it kind of democratizes the access to such things. Yeah, it's like 80-90% from moving from SaaS to AI-first SaaS. It's a pretty good saving.

And then the other component is that, yes, okay you go with UiPath but you can't implement UiPath most of the time yourself, so you look at vendors that are gonna implement. And then end up having like a six months project that's gonna cost like tens of thousands of dollars more to get started.

And then I remember that you also pay for the maintenance of that? Yes, yes, yes. Like 2K per month, right? We were talking to Chai that he was complaining that they would only accept emails and they would respond in unpredictable fashion.

I think the whole part about when I was discussing with you guys about process discovery was, you know, let me let loose an observation tool on the finance department, or X department, Y department, right? And it'll be able to collect data to be able to see which ones are quite good for potential automation.

So I always say that for legacy kind of companies, it's always better not to give them a blank sheet of paper because a lot of them do not know what good looks like. They've been in that company forever, right? 10 years, 20 years.

As soon as you cross-department, then the other department also needs to be convinced, right? So the more cooks... This is not even cooks, man. You've got to convince people that this is possible. And I think there is potentially still that fear that, "Will I be replaced?"

When I was playing around with browser-based, I didn't feel that it was good enough. The auditability of it, meaning that what did I do, what was the result, so on and so forth, that was a little bit lacking.

There are things about logging. There are things about detailed steps. So I didn't really pursue much. I was looking for more of that, but with the addition of things like logging, observability, so on and so forth. Then when I saw you guys, it's like, okay, I think it takes quite a few of these things.

Understandably, it's a startup. Understandably, there will be hiccups and all that. So which is why the business themselves also needs to be able to say, "Okay, I'm willing to do this." And at least for Imperial Treasure, 80 grand savings is worthwhile.

Yeah, now you're paying just 5K a year versus you were paying like 80K. 80K difference. And just to summarize what you said, so when you looked at AI-first browser-based and Claude computer use type of products, they were very innovative and they opened up what's possible.

They could do things, but they were not very consistent. You didn't have a lot of control over it. You want to have like a 50-step long process. You don't know if it's gonna do it consistently every time. You don't know what's the output of each step and what's the result.

No visibility, no consistency, and not designed for B2B type of use cases in the first place. I think to be fair, those were the early days of the product. And I think since then, obviously, the LLMs have improved. The time horizon that an agent can do things is now a lot better.

I think you guys already have those elements to an extent, and that is good. I think that probably compelled me to, "We need to try this in a bit more of a serious manner." And great, we took about two months to get from objective scoping all the way to going live.

So I think that is very good, the speed of it as well. But also I think kudos to you guys. You guys definitely stepped up and I think you definitely improved a few things.

I don't know how simple it is to do this if Chai does this, right? Because when we were discussing it over the past two months, there was a lot of edge cases and I'm not sure how you guys had addressed those edge cases, whether it is a product thing or whether it's more a prompting thing.

So whenever AI wants to build a new workflow, the knowledge base pulls up relevant, most useful examples that were prompted for typical problems. And typical edge cases like, oh, unexpected pop-up or when you press this button, actually you can get rejected.

I guess the more your customers increase, the more that knowledge base gets built up. So essentially, it gets smoother for the next one, right? Like a shared, hopefully anonymous kind of data. Because what you're more interested in is the behavior of the platform, not the data itself.

I know that you also used n8n. And built up like a portion of the workload through the n8n. But why couldn't you get to the whole thing? So I kind of suggested to Dan, "Okay, let me take this on. I'll try to do everything via n8n," which was quite okay, right?

And then there seems to be APIs for the SAP B1 part of it and all that. But I think as I went on, it was probably more... It became a little bit too technical for my liking, especially dealing with freaking SAP APIs, which is not as well documented as something like Stripe.

So kind of doing what we ended up with. But then I think the whole basic premise was if I can let Chai do it in a platform which has all those observabilities and all that, that would be better than doing n8n plus this, plus that.

So you knew about n8n. You knew it's the popular tool. You tried to do the whole thing end-to-end through n8n. But then, n8n doesn't specialize in desktop automation in the first place, so you have to rely on API-driven integrations.

And once you started to dig into the details of API, it turned to be way more technical. So if SAP hears this, they're going to crap on me. In the sense where I say it's not documented. It is kind of documented, but it's not friendly enough for someone like me who's semi or 10% technical.

You really probably need to get fully trained on, have used the SAP SDKs and everything else for some time to be able to figure out what to really do. This is like 20 years ago kind of method of doing stuff.

Yeah. Makes sense. So like with Mediar, you get more focus on non-technical users. And also, there are other things. So you build part of the workflow with n8n, but then the next phase of this, the whole entire thing should go into one platform if possible.

So you want to replace n8n completely? If possible, I think that would be good. Then when they need to change something, it's literally just... Because I'm thinking if Chai is not there, then I'm gonna have Denville's people or maybe junior IT.

I'm not exactly sure whether a finance guy would know that, wants to bother about knowing it. The good ones are not going to work for F&B. Chai is a rare case where he's an IT manager and a software working in F&B with legacy software. That combo is rare.

So all this, you know, it just so happens that it was quite a good combo in a way, but if Chai's not there, you want to do n8n plus Llama, it's going to be very tough. So everything onto something easier, like a Mediar thing, would definitely be the ideal platform.

How would you compare n8n and Mediar since you have experience of using both? I would say n8n is a lot more visual, right? In the back, it's obviously all JSON. It's very visual, right? It's nodes, boxes with arrows, decision trees and all that kind of stuff.

In the execution or the observability and monitoring portion of it, n8n has the history of every time it runs its execution. The whole entire flow. You can actually go into each node. When you double click, you can see input, output, input, output. So I think from that perspective, it is a lot more friendly.

Compared to the Mediar one, I think again you guys are obviously not version 5 or version 10 in that sense. It's still pretty daunting, right? There's a lot of words, there's a lot of "I need to look at things." So definitely not as user-friendly, but I'm presuming that this will eventually get to something more... less daunting for finance people.

So how was it collaborating with our company, and onboarding, getting started working and then moving into production? How would you describe the experience? I would say it was quite good. I think my only... I don't know whether I call it a challenge. It's more a time zone kind of problem, right? We are 13 or 15 hours apart.

And getting that timing and to an extent, the whole piece of responses, right? So it's a kind of like a 24-hour response time to an extent because of the time zones. I think that probably is a little bit of the not so ideal.

But besides that, I think it was great. I mean, you guys definitely tried to understand the problem, and secondly, you know the product, you own the product. So that's awesome. Any changes or whatever it is, you guys can just do it and potentially the next day things get solved.

The only thing that will say was, it's just not so ideal because of the time zone. But you guys definitely stepped up to it, and I think whenever there were problems, you were already quite proactively mitigating or addressing those platform related challenges. I definitely saw a lot more maturity and stability as we went on the project.

So that was good. Thank you very much. Yeah. We did add quite a bit of features on the platform side. As you mentioned, we usually ship new features to the product itself, like overnight. If you're missing something, "Oh, I need this extra field or this extra button," we're gonna ship it.

Would you recommend us to other clients now? Yeah. Definitely. I would definitely think so. One, I think the product's quite kind of there now, at least for a V1 or a GA version. And then obviously it will improve a lot more over time.

The like I say, I think when we do the onboarding, I also want to be there as well. Because I think I kind of want to learn this so that I can partly do it, and there are elements whereby maybe if we keep in contact, I can kind of pass those things over to you guys with my other clients or other prospects.

Because I actually do think this is awesome, you know. And the mere fact that AI is used to do this, I think is very good. The consistency doing this has been quite okay.

But you were saying that the agentic horizon is still quite short, right? Like, what are we looking at? What, 20 steps, 30 steps, and nothing really more yet to be kind of stable? Yeah. I would say just on the number of executions that goes down. Because in B2B, you want the workflow to be 100% stable, right?

And because they're on the scale of 10,000 executions, there are hundreds of edge cases. And even no matter how well you prompt AI, it might misinterpret that and do something that you would need a human 10 hours to fix. Because it will post something into SAP and you didn't find it.

You will find out about it when you do your tax return and they will fine you for like 30K because you did something wrong. So our approach is we build deterministic workflows with all edge cases, but they are deterministic. We build them with AI, and AI has the knowledge base.

At the end of the day, it's like human who supervises the building part, and then once it's built, it's very solid and it handles 100% of the situations. Which is why I guess it shouldn't be too many steps, because I guess the context window is not so large as well.

We do support step by step too. So 100%. It will be quite good. I don't think it will be a flagship client, but essentially it also shows, especially elements of F&B. I think there is quite a lot of things that can be automated in F&B.

Chai mentioned it's like 500 people company. Yeah. It is quite a lot. I mean, a lot of it is obviously service stuff and the cooks part of it. Head office is actually quite big as well. I think it could be about 50 people at head office.

So quite a lot of automation possibilities in that sense. The only problem is, one, is the mindset. Secondly, the pocket size of F&B isn't always big, which is actually quite good because from the perspective of Mediar, it makes automation a lot more affordable.

Whereas if we stick with the UiPaths of the world and automation anywhere kind of price tags, those are pretty much unaffordable. Yeah. That's a good point. Let's finish today's call and I'll send you an invite for next Wednesday. Thanks, guys. Thank you.