EP 019: Startup of the Week: Mandel AI

0:15Nick Gospodinov: Welcome to this episode of Art of Procurement's Startup of the Week series. I'm your host, Jothi Hartley. These Startup of the Week conversations allow us to shine a light on the tech startups challenging and revolutionizing the digital procurement ecosystem. And in just 20 minutes, we dig into each organization's origin story and value proposition and explore how they solve business challenges in a unique way. Our provider and startup of the week conversations are the only formats at Art of Procurement that focus specifically on a company's product. And today we are introducing Mandel AI. Mandel AI is an AI supply chain coordinator that helps manufacturers and distributors manage supplier communication, detect disruptions and automate procurement.

1:10Nick Gospodinov: Founded in 2024 and based in San Francisco, California, Mandel connects to existing email and ERP systems to streamline procurement workflows. And to tell us more about Mandel AI, I'm delighted to welcome Nick Gospodinov, founder and CEO of Mandel AI. A little bit about Nick. Nick studied electrical engineering at King's College London before co-founding an open banking payment startup that generated so much traffic at launch that it crashed the Bulgarian banking system. system. He then went on to work as a developer at Kraken, the cryptocurrency exchange, before being accepted into Y Combinator.

1:56Nick Gospodinov: Following that, a conversation with a Fortune 500 executive about how global supply chains were still run on email and spreadsheets sparked the idea for Mandel in 2024. Nick, welcome and thank you so much for being our Startup of the week. Thank you for the warm welcome. Okay, so let's dive in. For those who are listening who may not know what Mandel AI does, can you explain a little bit about what you do? Yeah, absolutely. In simple terms, we are an AI assistant that coordinates manufacturing and distribution supply chains. So we focus on direct procurement workflows. Essentially, we connect to the email inboxes of buyers where the communication between them and the supplier and logistics partners happens.

2:47We detect disruptions, we execute RFQs and purchase orders, chase suppliers, validate documents, etc.

2:55Nick Gospodinov: Okay, so that's quite a lot that you're doing. It sounds very simple, but if we can go back to the very beginning of when you first started out, what was the inspiration to start? What was the gap in the market? What was the the problem that you wanted to fix what inspired you yes well so um i was fascinated first with kind of like manufacturing and engineering and that was like part of the reason why i studied engineering in the first place but um when we went into white combinator we were kind of still in the payment slash finance space but we were working with some legacy companies and manufacturing and distribution and although we decided to pivot from that product we stayed within the industry and started like learning more.

3:41And that story that you just told earlier about me meeting this Fortune 500 executive, it was after a conference. If I remember correctly, we were having lunch together and I was like pitching one of our ideas and we just kind of like explored different things in supply chain. He mentioned how they're dinosaurs and they still communicate with suppliers over email. And I was like very surprised because I thought that this is like a problem that programmatically could be solved by me connecting with like an API on the supplier. side or us connecting like the two ERPs but it happens so that even the largest companies in the world still have majority of the work happen outside of the ERP rather than in it and so that will be usually be like an email a spreadsheet a pdf that arrived like a couple hours ago so it's a lot of like manual tasks that need to be done that's with large language models that could be

4:32Nick Gospodinov: solved yeah that's interesting I think about my own experience and how it's almost like you're operating parallel systems you've got the um the official way of doing things in in in whatever platform that you you're supposed to work in but there's this tendency to to go towards emails and spreadsheets because it's a comfort zone kind of um it's a path well trodden so yeah interesting we'll we'll explore a little bit more um as we as we go on this conversation um what what do you think what what in your opinion makes you different from say others that are operating in the market um what are the points that somebody that's operating procurement and supply chain will visibly see that's different yeah absolutely so a lot of the players as i mentioned focus on the already existing erp workflows just enhanced with ai so something that i actually recently was talking with some of our clients is that there's a lot of fluff in the market when it comes to adopting agents, right?

5:36Because we treat the agents as, I'll have like a sourcing agent, I'll have a negotiation agent, and we treat them like very functionally, right? But when it comes to supply chain procurement, a lot of the time, the tasks depend on the scenario that they are executed in. So they're kind of like, they won't be a super predefined RFQ that I need to execute. It won't be a super predefined negotiation that I just need to execute. It will be like a little bit of here, a little bit of there, and then I need to kind of combine them, maybe do it like in a spreadsheet at first and then send it out to the supplier or a lot of these like kind of like different scenarios, right?

6:12So the way that we do it is by you essentially give a playbook of how we run negotiation, how we run sourcing, et cetera, to our agents, and you know how to execute it based on the situation. But we don't have to burden our users by, oh, develop your own agent, or here's our AI studio, here's how you connect them and things like that. We have like a fundamentally different way of working. And as I said, the biggest difference is we focus on the communication layer in the first place. So we kind of treat the first source of information in the company when it comes to supply chain procurement to be the communication layer rather than the ERP because usually you first get it from the communication layer, which is either emails, meeting notes, or anything else.

6:55and then you bring it back into the ERP where you actually work with that data.

6:59Nick Gospodinov: Okay. So you're actually addressing the challenge or the problem or the issue at source as opposed to later on down the line when it's been modified somewhat. Absolutely. And what do your customers say when they give you feedback? What do they say that reassures you, that gives you the confidence that you are doing the right thing? Or what was or what continues to be the aha moment where you know you're doing the right thing will continue on this path? Well, for me, it's like, I like first principles thinking. So you break down problems into their fundamental level and think our supply chain is going to get more complex.

7:46They are. As we adopt more agents, there's going to be more coordination, not just between humans. There's going to be more between bots and humans and bots and bots and humans and bots etc so there's many configurations and uh if that's fundamentally true then somebody needs to solve that coordination problem right so the thing that we call the coordination tax and for 50 years we've been solving this with supplier portals with edi and a lot and even with rpas but that's kind of like forcing the other side to do something for me i've even heard of like manufacturers that are finding their suppliers for not using their portal and like emailing them but it's kind of like why would i be forcing my supplier that i'm trying to build like a relationship with to do something for me which of course is going to ramp up costs on their side right because it's it's wages it's many hours time that they need to now go into your portal to update information but they also have other customers so they need to update their portals as well so it gets like very complicated right but us even on this call we're going to exchange a lot of information through natural language and we have these like systems now with AI that can actually parse that natural language, right?

8:50Understand it, act on it, shift that data from one system to another. So why not have that orchestrator in place for both sides, so we can actually communicate naturally, but everything else happens in the background.

9:02Nick Gospodinov: Okay, that's really good feedback then. And just staying on that for a second then, so your customers give you a lot of feedback, and they're helping you develop in very short space of time that you've uh you've been around um but what what has shaped the product most do you think we stay very close to our customers i don't think there's been a single customer deployment where me or another person on the team has not flied out to to meet them in person so even if it's like a small customer we always sit beside them at least for a couple of days to actually understand how their workflows like work what do they do what is the random spreadsheet that there's like five people on the team that essentially that spreadsheet runs the whole company so how is that built why is it built like why are you doing these kinds of things because there's always a reason for it right we call it like manual work there's a reason it exists right so we need to fundamentally understand what drives that company and the great thing about ai is that because of the fact that every company works differently.

10:05In the past, we've done like very complex integrations and very complex systems where there's like custom built or hire consultants to build on top of. Now with AI, we can kind of like make it more flexible, right? But no, but we also need to teach these agents on how my specific organization works. And that's, I think, the knowledge transfer between buyer and an agent, buyer-supplier relationship and that agent, kind of, that's, think the biggest problem to be solved and our customer feedback has been invaluable because just like by us sitting as close to the customer as possible and trying to ship together with them like we had this case like uh very recently where one of our customers was extremely surprised because we sat with one of the buyers on their desk and we saw this like very weird excel table that they had where they were tracking like rfqs and quote prices and everything and she said like it would be so great if, um, like, uh, the, your AI can go gather these quotes and they allow me to just like simply export them and I can merge them in that table.

11:09It will save me like five hours a day or something like that. And we just sat and like, by the end of the day, we actually build that functionality for them. And they were like, it's just mind blowing for them. Right. So, so yeah, it's, uh, yeah, to us, it's just amazing because I don't come from like a procurement and supply chain background. But I do come from like an AI and engineering background. And so I just love fixing these problems that are very complex and very high stakes. And we mostly also focus on direct procurement. So the cases that we see are very complex because let's say you run like an RFQ or a purchase order.

11:45These usually are massive, very complex with a lot of specifications, a lot of back and forth with the supplier. It's not just like, oh, I sent out an order for like three things and I receive a confirmation for three things and everything is perfect. Usually there's quite a lot of things happening between like the initial RFQ and the final delivery of these materials to your facilities.

12:06Nick Gospodinov: You're right. Yeah, there's a lot of back and forth and all of that, every single time there's chat or email or conversation, there's information being shared and that can be quite rich and sometimes if that's misinterpreted, that can lead to a very different outcome, of course. Absolutely. That's really interesting. I mean, you talk about some of the shaping of the product. What has surprised you about maybe some of that feedback or just some of the development, the journey that you've gone on? What in building a procurement startup and you say that you don't have a background in procurement, which is interesting because you've got an outside in perspective.

12:50Nick Gospodinov: So what has surprised you about creating this product? well the initial idea right because i think like all of our competitors focus on the known things in procurement which are like i have the erp i have the known like sourcing events i have all the known processes i just need to automate them with ai as you said we kind of like started outside in and then how that's how we learned about the industry from like simple discussions like oh we still run things on email so we're like okay let's just build apply ai to emails in supply chain that's how we kind of started and that's how we learned more about the industry but uh what surprised me maybe most like and that's like in the past few days and it's very recent so that's why i'll say it um actually agent adoption is extremely low right now uh you asked me like how are we different to our competitors and what they're going to see in the market um people are still very early on the journey in terms of adoption and supply chain procurement for ai agents people are like still on the level of simple co-pilots maybe integrated like microsoft copilot in their in their company and now everyone is like pitching multi-agent systems very complex like things running your whole procurement autonomy things like that and i think for a procurement professional their head is just going to explode from all that information right so it's about like simplicity uh now and simplifying these kind of things and making rapid deployments i think that's one of our biggest advantage by the way is that we recently actually integrated with one of the largest hourspace distributors in the world within a single day and by the end of that the our agent has already caught like multiple quotes it proposed a negotiation opportunity uh picked a picked a winning bid for like an rfq and things like that all autonomously straight from the inbox just by connecting a couple of the inboxes without any european integration just from getting all the context that is initially surfaced through this communication layer and uh building up on top of that like kind of source of truth so so yeah i I guess like adoption has been my biggest surprise recently in terms of, I thought the industry is further because I'm in tech.

14:52And of course I see what my competitors are doing. I see them like raising big rounds, we're raising big rounds. And then it's like just competition, competition. And then you actually go in the industry and there's people like there's, I recently talked with like a company that's over 20 ,000 people. They, until last year, didn't even have like a formal purchase order process. They were calling their suppliers, can I get this? And the supplier will be shipping it. And then you will trust the buyer that they will be following up on the phone, not even over email. So there's quite a lot of interesting things happening in the industry.

15:24And it's, I think I'm very excited to be building in this space.

15:27Nick Gospodinov: That's really interesting. Yeah. And it's quite surprising. It's surprising and it's not surprising to hear your reflections on the procurement space, I suppose. If I could ask you about what do you see the biggest friction point for, say, a procurement team that is looking at your product? What would your product address first, would you say? So the biggest friction point for sure has been trust, because if you're coming from a non-tech industry, which is majority of like manufacturing and distribution unless you're one of the VC backed manufacturers that are building like one of these products and you have an internal tech team is just understanding of is AI going to train on my data?

16:16What if I allow to actually the AI to roam around my organization and all of a sudden it takes control, right? We even had like these kinds of questions. What if AI takes my job? There's quite a lot of these like trust factors, both from the procurement departments and IT teams. So I think that's the biggest barrier that usually slows down our deals. Then we've done like a lot of work to actually ensure that these people understand the risks behind the AI adoption. Why, like, for example, for us, like, because we're on the communication layer, right? We need like access to your emails. And so why would I give you access to my emails?

16:51What are you going to do with them? What if like, what if there's like an engineering drawing that is like extremely important to our company it leaks and there's quite a lot of these kind of questions that we've needed to work through the past like year or two to actually ensure these people understand like what we're doing and why why it's good for them yeah it's understanding and also it's being able to

17:13Nick Gospodinov: trust and and commit to and have the confidence in working with you knowing that the guardrails and the firewalls and everything is secure yeah good good point um so if a customer um is talking to you about about your product what are the typical roi um factors or kpis that they're looking looking for um what are they looking to to improve or solve when they come to you so when they come to us it's usually around uh saving amount of like hours of work when it comes to many follow-ups data extraction putting that data back into the erp but they very quickly discovered that there's like quite a lot of deep roi when it comes to for example quote expiration costs so because of the fact that you're sending out rfqs getting the quotes in all kinds of formats for example then you need to put them in the table reconcile the materials and the specifications so they're actually comparable then choosing the bid to win having probably like a meeting with the team to actually uh choose the supplier to win and then going back to the supplier negotiating etc that can take so much time that by the time i actually choose the supplier the quote might have already expired and they say oh we're going to search as you like two three four percent and so throughout like the whole year if you have like a billion dollar supply chain you're probably losing quite a lot of money from that uh other things when it comes to purchase orders because of all the manual work surrounding them and you're the only way for you to actually scale is with more headcount uh if you're if there's like any kind of disruption you may need to expedite way more frequently than you would if you had like a system that actually proactively manages and monitors your whole supply chain and all your operations to actually ensure everything is running smoothly.

18:59And when there's like something going on wrong, it actually escalates and finds the right person in the company to solve that problem. So you don't have to like expedite things like that, right? There's quite a lot of things that you figure out very quickly. Now, one of our last clients deployments, they said it's kind of like a kid in a candy store when we went to their facilities, because the amount of things that you can actually do with our system is a lot to the point where we started getting dragged into meetings with accounts payable teams, with logistics, on the export side, on the import side, customer support and things like that.

19:35So we're trying to stay laser focused when it comes to direct procurement. But I mean, we've had deployments, not one where it's been like expanded very quickly into indirect procurement just because of the fact that the natural adoption of the product.

19:51Nick Gospodinov: Interesting. Yes. Yeah, that's that's everybody wants a piece of your product and what your platform can offer. Very interesting. Just thinking about how your product fits within the rest of the procurement tech stack. so a procurement person supply chain person we talked about it earlier on in the conversation about how they may be operating with different systems as well as the email and the spreadsheets that we all know and love um so where does your solution fit like how would they how would a user interact with mandel ai it's our system starts where all the other systems end so the moment that the purchase order goes out everything is chaos right so it would be supplier chasing understanding if there's like a back order and we need to actually react and switch our supplier if all the milestones of that purchase orders are met is it in production when is production supposed to end have they actually met us have they notified us that the production has ended have all the batches like split and maybe we need to now do a partial shipment to meet the deadline there's quite a lot of these kind of questions that the only way for you to actually understand would be through manual work right through a person actually having to remember maybe no down own, like, so put it somewhere in the system.

21:05And it's not only about purchase orders, it's about quotations, it's about shipments and all the work that is done through these communication channels after the work of the purchase request, for example, being turned into a purchase order, and then going out of the system. That's essentially where all of these systems end, and the problem has never been solved. So that's where we start. But you can treat this as kind of like an orchestrator between these systems because we would still end up connecting with all of these systems for us to push data back, for us to inform our agents about any kind of updates that were not happening across these communication channels.

21:39But I just like to say that like we can go live within a single day just because we are laser focused on the communication layer first and we treat it as the origin source of any kind of information in your supply chain procurement cycle before it goes towards these systems. So yeah.

21:58Nick Gospodinov: okay if i was to ask you what is the most underrated or what was the unsung hero feature or functionality in in your product what's what's the one that's in your platform that is it it punches way above its weight well probably the escalation system because it's uh it sounds like oh just like the ai just monitors like emails and when it spots something it escalates but our system has caught like some very interesting problems in the past where a human would not have caught it um so for example we did like a deployment in a fortune 500 company right around uh the trump tariff situations and one of the suppliers was emailing so we're about to apply a surcharge on this material group of like let's say 15 % starting next week because of the tariffs.

22:51And then the agent, what it did, and it was not programmed to do that, and that's the beauty of actually building in an agentic way, is the agent went online and checked for tariff exemptions. And it came back and actually saw that there's a tariff exemption for like another three weeks. And so it escalated to a person and said, well, this supplier, he can't be surcharging you next week week if it's justified by the tariffs because there's tariff exemption for another three weeks. And so that's like the beauty of agents, right? That's something that the human would probably by the business of their day, probably just accept it.

23:28And even if they only if they had time, they would actually go and like actually go search up online. Or a couple of months ago, we worked with a very large pharmaceutical distributor in Europe where they had like a purchase order go out for like 6 million euros for like a lot of line items and they had just like negotiated a contract that had like certain discounts that were expected um in the in some of these materials that they were ordering from that supplier the supplier sent back the new invoice on the six million euros of materials and then the discount was not there but the agent because it had context of that previous negotiation it actually saw okay so they promised this but now it's not in the invoice and it escalated to a person we found like a 1 % total price discrepancy in that invoice.

24:16So it was like 60 ,000 euros that we surfaced immediately to the buyer who could actually fix that problem before it went to like to the year end financials and then it gets spotted.

24:27Nick Gospodinov: So helpful. Yeah. Yeah. Okay. Thank you. Just as we're about to wrap up here, Nick, if somebody's listening to this and they want to, they're interested in getting started with you, what advice would you give them? how can they get started with with your platform well first of all book a call with me or someone on my team but um it's very simple as i said because we don't focus first on day one to actually integrate with all the existing legacy systems in qrp etc it's very easy to get started literally within the first day we can plug in your emails and the ai will start tragging between the different inboxes so for example if you sent out a purchase order in one inbox get a confirmation in another inbox or like in that same in a separate thread all of that is going to get caught by the agent so it will be our goal has always been to get to roi we say internally within 15 minutes but within the first day at least so actually when you plug in these inboxes for you to quickly understand do i have like an invoice with any kind of issues maybe a drawing that got rejected and the engineer here needs to go back to the supplier and see and prove the changes approach so that I've never got confirmed.

25:42So all kinds of these like ambiguous problems that you would see in your day to day where an agent might actually be able to help you without even having to go and program. And that's why I'm like kind of against the whole multi agent movement movement where you build like separate agents because we treat them as like functional problems where like the unit of design should not be an agent. It should be like a scenario in which or like a playbook on which the agent can, depending on the situation, use.

26:10Nick Gospodinov: Okay, that's useful. So where can people go to find out more, to learn more, and to indeed contact you to book that call? Yep, you can go to our website at mandel.ai, so M-A-N-D-E-L dot A-I, or find me on LinkedIn, or my email is nick at mandel.ai, pretty simple, so all of these options are at your disposal. Perfect. That's great. Thank you, Nick. Well, Nick, thank you so much for this opportunity to learn more about you, your backstory, the company and Mandel AI's value proposition. And thank you to everyone who joined us to learn more. Take care, everybody. Thanks for listening to this episode of ProcureTech Insider.

26:56If you found this conversation valuable, you can subscribe to our ProcureTech specific newsletter, The Stack. Every Thursday, you'll get intelligence on ProcureTech companies, market developments and insights from practitioners delivered right to your inbox. To subscribe, go to artofprocurement.com slash subscribe. That's artofprocurement.com slash subscribe.