
How to make AI agents work at B2B SaaS
Online
Most B2B SaaS companies have tried AI agents. Few have scaled them. The difference is rarely the technology. It's who owns the agents and how teams are set up to build them.
Watch recordingEvent
information
Most B2B SaaS companies have experimented with AI agents. Far fewer have made them part of how sales, customer success, product and engineering, and finance actually work.
In this lunch webinar, Leilei Tong, Head of Customer Success at Abundly, shares what we've learned from working with B2B SaaS companies: where agents create real value, how to scale them without losing control, and what separates successful adoption from stalled pilots. Expect concrete examples and practical takeaways.
We'll cover:
- Where agents create real value: high-impact agents across sales, customer success, product and engineering, and finance
- Governance that scales: how to put ownership in the teams while a central function stays in control
- Lessons from the field: the most common pitfalls and how to avoid them
- Getting started: a practical checklist for your own organization
There will be plenty of time for questions.
Who should attend
Leaders and team leads in B2B SaaS companies.
Speakers

Leilei Tong
Head of Customer Success, Abundly AI
Recording
45
Watch recording
45
Enter your details to access
Summary
Most B2B SaaS companies start their AI agent journey the same way: a few enthusiastic employees build promising pilots, often on their own laptops, while a much larger layer of shadow AI goes on unseen. The results are fragmented. Pilots are chosen for convenience rather than value, nobody measures their impact, and agents become orphaned when their creator leaves. Research shows that only about a quarter of companies running agents keep a central inventory of them. The challenge is moving from these isolated pockets to real transformation, where every team works with several agents side by side with humans, as in a finance team where billing and collection specialists each have dedicated agents reporting to them. Three ingredients drive success. The first is a single approved platform where agents can be governed, shared, measured for value and controlled for cost, with clear ownership flowing through the org chart. The second is enablement of domain experts rather than central IT teams or consultants. The people closest to the work should build the agents, and they need protected time to finish them, not just to start. The third is a fleet plan: a roadmap for each team that maps processes and prioritizes high-value quick wins. This is often the hardest part. Done well, this can happen in about three months. Examples include one customer with 113 production agents and 52 trained champions in 16 weeks, and another with production agents in all 11 departments. Supporting practices include leadership training, champion networks and showcasing wins at all-hands meetings. Common risks are shadow AI, lost momentum from lack of time, and focusing on token costs instead of demonstrating value early. Companies also shouldn't wait for perfect data. Instead, start with the agents you need and solve their specific context requirements. The key takeaway is to ask yourself a few honest questions. Do you know which agents exist? Can team leads see what they cost and deliver? Does every team have a roadmap? And do your builders have time to actually finish what they start?
Transcript
How to make AI agents work at B2B SaaS - 2026/10/22 11:45 CEST - Transcript
Attendees
Adrian, Alex Constantin, Andreas Viklund, Anke Nechwatal, Åsa Martinsson, Barbro Pettersson, Benjamin, Carlo Beschi, Carsten, Cathrine Bore, Chas Academy assistant, Christer Öberg, Claes Scheffer, Claus Broholm, Daniel Apler, David, Dmytro Shervarly, Domenic Vossen, Editerats assistent, Elina Saarinen, Emelie Alm, Emil Stridfeldt, Erik Häregård, Erika Boberg, Erika Olsson, Erland, Erwin Siu, Ewa Woloszko-Klos, Felicia Fazzi, Felix Bemm, Fireflies.ai Notetaker Domenic, Fredrik, Fredrik Bergström, Fredrik Ernander, Fredrik Lundberg, Fredrik Nyberg, Gabor S, Goofy Salen, hamza rabah, Hanna MG, Hannes Palm, Helene Haegerstrand, Henric Johansson, Holger J, HRM Software, Janet, Jerzy Stawicki, Joel Rönnqvist, Johan Jacobsson, Jolo, jonas jaconelli, Jonathan, Joost, Josefin Wulff, Juan Barberis, Kamila Kuchalieva, Karin Pagels, Keiward Pham, Kristina Stojanovic, Krisztian Niesz, Leilei Tong, Lennart Boklund, Lillian, Lina Andersson, Linda Nilsson, Linn Markryd, Lone Koch, Lotten Skeppstedt, Madelen Porserud, Magnus Sundin, Mar B, Marc Ottersbach, Markus Michelbrink, Markus Onyango, Martin Börlin, Martin Klose, Mikael Engbom, MÜMİN HİCDURMAZ, Nidhi Shetty, Nis Holst, Oliver Schäfer, Oscar Ohde, Per Brandt, Per Olof Beckemeier, Peter Boman Byström, Peter Oerhalmi, Peter Vestergaard, Petter Åström, Petter Nilsson, Pontus, Powered by Dstny, Powered by Meetric, Reclaimit, robert brunbäck, Robert Ingemarsson, Robin Svensson, Rune Østbye, Shuchi, Shuchi Agrawal, SIKI, Sofia Palmqvist, Stefan Åkesson, Stefan Grenstad, Stelios Apostolakis, Susan Meszaros, Susan's Fathom Notetaker, Thomas Sjövy, Tobias Westerblom, Tor Ganslandt, Vasanthiha Ravichandhiran, Ziki Khan
Transcript
00:10:00
Leilei Tong: Hello. Hi everyone. You are muted.
00:15:00
Åsa Martinsson: Yes. Hello everyone. We I can see that some people are still popping in. That's great. So, we will start in a couple of minutes because people tend to come from other meetings when it is lunchtime. Uh but while uh you're waiting, please pop into the chat where you're based, which company you're representing, and uh an emoji uh for your mood today. And that's of course optional, but it's just fun for us to see where you're from and how you're doing today.
Leilei Tong: I really like Daniel's marketing of Upsala, the former capital of Sweden. I'm from Upsala, so it's just the heart for me.
Åsa Martinsson: Excellent. And I'm also thinking Lelay uh would you like to take over uh sharing your screen and…
Leilei Tong: Glad to hear that. Sure. Or I can do it when you're done.
Åsa Martinsson: sharing the presentation?
Leilei Tong: Is that easier? Then you can control it yourself. Or do you prefer for me to share?
Åsa Martinsson: Oh well uh I think I well it doesn't matter. Uh I think it's better if you share because then I can see what's happening in the chat also.
Leilei Tong: Okay. Then I will share.
Åsa Martinsson: Thank you.
Leilei Tong: Uh I will take over from Osa. Yes. See. All right.
Åsa Martinsson: Excellent. Now I can see some of the people. Fun. So some people from Sweden obviously because we are based in Stockholm. Nice to see you. some familiar companies also,…
Åsa Martinsson: but fun. Here we have some from Australia. Welcome. I don't know what time it is where you're at the moment, but very happy to have you.
Leilei Tong: Oh, wow.
Leilei Tong: Okay. What do you think was that? Should we get started?
Åsa Martinsson: Yes, I think so.
Åsa Martinsson: I can see that the number is just rising, but let's start anyway. So, great to see everyone and uh welcome to our webinar. We are going to talk about how to build AI agents at B2B S companies and uh I am your moderator for today and I'm a part of our uh customer and uh uh commerce commercial team um and I am working as a sales manager there and with me I have Lele would you also like to introduce
Leilei Tong: Yes, my name is Lelay and…
Leilei Tong: I lead our customer success team. So I see a lot of what's happening with our customers that are uh mainly B2B SAS companies. So I'm here today to uh bring some of the learnings that we have from working with our customers.
Åsa Martinsson: Fantastic. And uh…
Åsa Martinsson: if you would like to uh just click to the next slide. Perfect. Before we start, it's also just good for you to know that we are recording this webinar and that is uh so that you will be able to listen to it afterwards and we will of course send out an email and around some etiquets for this webinar.
Åsa Martinsson: We are would love if you could stay muted and…
Åsa Martinsson: instead save all the questions until the end of the presentation uh because we have should have a couple of minutes at least or many a lot of time to go through all the questions. Uh so that would be highly appreciated.
Leilei Tong: You can write them…
Leilei Tong: though in the chat um when you think about them so you don't forget about them.
Åsa Martinsson: Yeah. Yes.
Leilei Tong: So when you have a question in your head just write it in the chat. We'll just probably take it at the end uh since it's a bit easier to manage it that way.
Åsa Martinsson: And also my experience is that we tend to answer your questions along the way or let's hope that. And before Lelay goes in and…
Leilei Tong: Thank you.
Åsa Martinsson: dive dives deep into today's topic, I just want to share a quick note on who we are.
Åsa Martinsson: So, Abundi are an AI agent platform and many companies that have been following us throughout the years um might think of us as a consultant company because we have done so many trainings on how to build agents. We have done workshops in the area and we still do all of those things. But today it is how we work with our customers that are using the platform. And the reason why we think that everyone should have a agent platform is that if you start to work with agents, they need somewhere to live.
00:20:00
Åsa Martinsson: They need a place where everyone in the organization can collaborate with them, not only one-on-one in some other space. And we have also seen that it's when the agents also start to collaborate with each other in the platform where companies have the biggest wins. And one of these companies that are working really successful with this is our client uh uh puzzle and…
Åsa Martinsson: they are also a business company which I know that many of you are in here. So instead of listening to me explaining how you can work with the platform, we will share a short video on them explaining how they are using the platform.
Leilei Tong: Yes. Let me just do this. All right. for us is a very powerful agent platform who enables everyone to build agents independent of role to solve their problems with their specific needs. Abund is also a very strategic AI advisor who helps us build our agents safely and compliant and also pushes us and coaches us through this change. It makes it much easier for us to move fast but still with control. Today we have hundreds of agents in our sandbox and we have over 20 agents that have already moved into production because a lot of internal AI projects they are actually stranding in the sandbox and Abundi has helped us push through and get it to production where the real value is delivered. We also have an agent running across all the product teams.
Leilei Tong:
Leilei Tong: He called Bion and he collaborate with the whole team and do all the admin when it comes to incident reporting. So he has actually already now become a pretty critical colleague. 3 months ago AI was a big part of puzzle but what we didn't have 3 months ago was a shared platform to build agents across teams and that's really what abund provides. It's not a centralized AI team that builds the agents. It's not consultants. It's built by the people who are closest to the problems. One thing that really surprised me is the engagement, how much fun the builders has building their agents and much less scary than we thought when we started this journey. What we see with AI agents is that human becomes more important, not less. That's probably the most important thing we've done is actually upscaling our puzzlers.
Leilei Tong: So by the end of this year, I do believe that we have scaled this even further. So then the team would be a combination of humans and AI agents because we actually see the AI agents not as tools but as AI colleagues. So I would for sure recommend Abundlay. We find the platform very easy to use with the security and…
Åsa Martinsson: Yes.
Leilei Tong: guardrails built in.
Leilei Tong: Maybe even more important, the support we've been able to get from Abundi has been quite a key component for us. Getting this far so fast and delivering this much value. Okay. So,…
Leilei Tong:
Leilei Tong: Um yeah that was a bit of uh shameless plugging what do you call it shameless marketing from our side but we like to show this video because I think uh signed the the main speaker in this video uh which works at the who she's the CTO of puzzle highlights a lot of the concepts and themes that we will actually be talking about today around like uh AI colleagues and some of the difficulties and speed and so on.
Leilei Tong: So uh yes I will get into that. So the topic for today um making AI agents work and then primarily focusing on how do you get this from a pilot stage where you might have a few pilots running to like the real transformation. Uh I wanted to kind of start with the with the end um and just visualize a bit what this looks like. So uh what it looks like in in this team uh this is one of our customers in their finance team is that you have the finance lead.
Leilei Tong: So the person who manages the team that person is there and that person manages in this case two people uh among some other people a billing specialist and a collection specialist and these two people each have their own agents working for them like one agent that works in checking billing completeness another agents that works with collections and they kind of work with and report to these two humans and then they also have a few other um agents in the team that works with one is like answering questions coming from other team another one is looking at contracts and so on that works with the whole team. And so you could say that in this finance team now uh it's a team that consists of humans and agents working side by side. Uh so this is like an orchart view. On the right side you have kind of the process view where you look at like the different parts of the things that they cover in this scene. Uh for each step they have you know a human and some agents working with it side by side.
00:25:00
Leilei Tong: And this is something that has really helped to free up the time of the people working in this team and actually also made their work more fun because they can spend less time on the kind of manual. Oh, I need to check this check this thing you know which is just like reading through a lot of text and whatnot. They can do less of that and more of the interesting stuff which is you know uh higher level work. Um so that is something we see a lot with the companies we work with actually that the EMPS the employee satisfaction increases. Okay. Um but taking a step back then um most of the companies or say almost all of the companies that we work with they start with some variant of this stage which is what we call the pilot stage.
Leilei Tong: So they are in a space where they have maybe done some experimentation at the best I would say some have done haven't even done this but they might have done some experimentation in a few pockets of teams in their companies where where they have seen real wins um that they have talked about and showcased in the company but the problem is that this is fragmented right this is not something that has happened at scale it's just a few like proof of concepts or something like that. So, uh the problem right there's a few problems. One problem with this stage that they're in is that they have shown some real wins but not like true uh significant value because usually these pox or these pilots are picked chosen based on something like uh yes this is something that one person was really passionate about.
Leilei Tong: You know, every company has those people, right, that like love to explore new new tools and new toys as they see it and and they do something with that. Or it might be something that was chosen by convenience like, oh yes, this is an easy case, so let's do that. Not necessarily what brings the most value, right? Uh and usually when it happens at this level, it's not something where you like track and measure the value regularly. Um and it's not part of like the core work processes of the company where you see the real value. The other issue is that at this stage this happens a bit ad hoc meaning that these agents even if they are great they are only known by one person. So it's very fragile. Uh so whoever created the agent knows about the agent and in a lot of cases actually it's even just done on that person's computer. Uh if it's a company that has claude for example which a lot of our customers do uh it might be like a claude agent that just sits on that person's computer.
Leilei Tong: Uh so this creator knows how to do it but uh that's the the extent of it right so it means that it's extremely fragile because if that person leaves this agent is essentially orphaned and it also means that skills about building agents only sits with a few very enthusiastic uh people. Uh the third of course is a problem of shadow AI. So a lot going on uh under the surface right so the known pilots are like tip of the iceberg that little thing and then you have like the 90% happening below that where you don't really know what it is you know it's um a lack of like a common platform a usage of a lot of different tools and I think most importantly there's no quality assurance and there's no security review that happens systematically across all the agents being built in this company there's actually been some research done on this a very recent one
Leilei Tong: show that only among the companies who are running agents. So companies in the forefront you could say only about a quarter of them actually have like a central repository or list of who they are. So you know the vast majority 75% actually don't even know what agents they have in the company and that is actually quite scary if you think about a lot of the recent happenings with agents doing different things um in the last few months. So the question here is how do you go from pockets just having a few uh pilots to real transformation you know where you have this at scale where you have several agents in every team. So where you both have the kind of the breadth the comprehensiveness covering the whole company and the density and the depth of every team having several agents. So you get to this like human AI collaboration that we talked about in the beginning. How do you get to this transformation? Um this is the journey that a lot of our customers
Leilei Tong: go through. So they start in the in the you know the pilot stage right the bottom left of this this curve and then um um they they uh we come in and we say okay all right let's let's uh find a group of people so not every single person at the company at the same time but a select group of people that still covers the whole company in the sense of like every single function uh we give them some kind of designation like a champion AI champion type thing or or agent builder designation and then they start to experiment and they do a lot of enablement. We do a lot of enablement where they learn systematically how do you build agents? What should you think about? What are some of the key security risks you should consider and so on. So a lot of things happening and a lot of uh bubbling energy a lot of like uh trials and uh and so on. And then at the same time of course the companies uh work with governance.
00:30:00
Leilei Tong: So usually when you have done a lot of experimentation you scale back a little bit to catch up with governance because a lot of questions will arise at that point right even if you try to have your AI policy or AI agent policies ahead there will be a lot of practical questions that you need to resolve once it starts bubbling figuring out okay exactly how do we handle this system uh exactly how do we handle this level of data sensitivity and so on. So um once when you do that uh you get to the second climb which is kind of the real the real climb or the real value where you get to the true transformation where you start to do the kind of really value adding agents the core agents of the company. Uh now throughout this journey there's a lot of um points I guess where you could fail where you could flatline. One is the pilot flatline which we talked about in the beginning like just sticking to the pilots and not getting through and not doing more than that.
Leilei Tong: Another one is when you do a lot of experimentation and you start to have a lot of um bdding agents but they are not brought to the finish line. Um so there's a lot of areas here where it could flatline but if you do it well you can get to this curve uh that we have here. Uh okay so what are the ingredients then to success right? How do you get through this curve successfully? How do you land at the end there where you want to be? So one we already talked about the platforms you need one place for all your agents right a place where you can manage govern collaborate on them uh which solves kind of the shadow AI problem that is very common you know 75% of companies u the second one is enablement so finding these people throughout the company that can build that want to build and most importantly get the time to do so um they have they need to have like an allocated protected
Leilei Tong: the time uh to get to be able to do what they need to do. And the third one is uh to have uh a fleet plan. So by fleet I mean agent fleet having an idea of um what are the agents that this team actually should have, right? What does that fleet look like? What does that picture look like? that I showed in the beginning with the humans and the agents. What does that look like? And then having a road map, a plan for saying, okay, then we need to build this and this and that agent first and then only after those fundamental agents then we can build this and this and this agent. So having a plan for the fleet, which is typically actually the the hardest part uh for um for the customers that we work with. Um yes. So on the first one, the platform, you need to manage your agents as a central part of your business, right?
Leilei Tong: So you need to create uh ownership. You need to have ownership very clear that flows through the company. So if I'm an agent builder and I build an agent, then my boss, my team lead needs to know right that this agent exists and what it's supposed to do and it should flow up in the in the or chart you know accountability and responsibility wise. So you need visibility in order to do this. Um you need to work with of course systematic data access and also spend control. Um super importantly valuing uh measuring value uh as well right so it's not just becomes like a cost item but like measuring what are the output what is the output what's the productivity what's the value get getting we get from these agents uh and then a place where you can share and use and collaborate and improve on agents together and we actually see um that um among the companies that have like a strong positive return on the agent
Leilei Tong: that they have 50% of them had one approved platform whereas only 9% were like in experimental stage and having different platforms. So it's important to um to uh to actually have one central place for this to work well. It's a big driver. This is an example of how it looks in our platform. There are of course a lot of other platforms out there. This is just an example of uh this is how it actually looks in our team uh where you have like a bunch of agents. You can see how critical they are. You can see what teams they belong to and what output they deliver and also how much uh they spend as well. So you can like have an overview of all of this and you can manage all of them in one central place. So that's on the platform side.
00:35:00
Leilei Tong: Second part was about having uh people right uh people who can build across the company or throughout the company and we strongly believe that it should be the domain experts to build meaning that it's not a central uh IT team uh a central uh developer team or like a group of consultants or something like that because that is not scalable. We strongly strongly believe and we've seen this in practice in so many places that the people who work with something right they are the ones that should be building the agents for their work. Um because if you don't do that and someone else builds for you, you end up having a black box. You don't know exactly how it works. You don't know how to improve it. It's very slow, right? Having one central team developing all of this and it's very hard to actually get spot on uh because the people who build then aren't the users, right? So you should have the people to build agents for themselves or for their own teams.
Leilei Tong: Which is why you need to have these um AI champions or AI builders throughout the company because they know the bottlenecks of the work. They know the solutions often and they use these things, right? So they can constantly feedback loop and improve on them and then you can get to scale much faster because every team can work on this in parallel. Uh and then thirdly this fleet uh plan. So every team needs a road map. So um it's similar to product development actually but for agent development. So mapping the work uh the processes being done today figuring out what agents are needed in order to support those doing some prioritization to figure out which are your quick wins that you should do first and so on and then putting this all into a plan where there's like an owner and a timeline and all of that. Um and yeah, this is an example of what one of those fleets could look like. This is for a finance team.
Leilei Tong: So kind of the first one I show the first slide I show there with the humans and the AI working side by side. These are this is the fleet. This is the AI fleet, the agent fleet of of that team uh that they have for like the different steps of the company. Okay. So how long does this take? Um for uh our customers I say it's about three months roughly. And what do you do there is this three months right? You want to get to the curve right? So what do you do in practice? first um select the champions, find the people who will do this and uh and uh enable them and so that they can kind of start to build and then have their first showcases and really learn right how do you build a good AI agent what is this at the same time do leadership training because um someone needs to be responsible for these agents right and that needs to flow through the orchard which means that leaders need to also be trained they need to understand what does it mean for me now that I'm not
Leilei Tong: just leading a team of humans. I'm leading a team of humans and AI agents. What does that mean? Right? Uh then once you've done this, you can do uh what we call the road mapping. So creating these kind of fleets and setting them into like a um a work plan. We call that a road map. Um so doing those often requires that you have first kind of done the training and understand what an agent is. Uh so that comes after and then you do these road mappings and you figure aha this is the fleet that I need for my team and we commit to building this and and this is our work plan. Uh uh uh usually in the agent building phase so because after you set the road maps you start to build a lot of agents right this is when you truly get to start to scale up on your value creating agents. You might need to be a you might need a bit of coaching to really get through so you don't get stuck somewhere. That is also important. And then lastly, and I think this is uh sometimes easily forgotten. I see this sometimes with our customers and we really try to encourage this.
Leilei Tong: creating networks, creating forums for these champions or builders to interact, to really like share their knowledge, share inspiration, you know, cross-pollinate and and collaborate. ate uh because you need the momentum to keep people going to get through this second climb. So you need to kind of also reward them, right? Or like um uh get them the opportunity to show what they have done uh so that they become like the heroes in a way um and are motivated to continue. Okay. Um what does this add up to? So some real numbers from uh this year on on the speed side of things.
Leilei Tong: Uh Singa mentioned some in the in the video in the beginning. Uh some other examples here, one customer had 113 not just agents that you know that are like blah and and obsolete and then and then not taken care of but like actually live production agents in in 16 weeks with 52 people uh onboarded and trained um throughout the company on on building these. So they're champions. Another company where they uh had a team of agents in every single team. So there 11 11 departments I guess you could say. Every single one of them had like a couple of production agents built by their they were also 50some um agent uh champions in that company in under three months. So you can get to this very fast uh if you uh if you get it right output. This is just some examples from you know agents.
00:40:00
Leilei Tong: This is yeah and this is it feels like um when I when I look at this it feels like I can mention two examples out of like thousands but you can get to a lot of output with your agents. This is a sales agent working in the commercial team with like hundreds of leads qualified in a month and like um a CRM agent that helps you clean up CRM. We see this so often with the companies that we work with that there's like data quality issues in their systems. uh and I don't think agents should replace systems but they can help to clean up they can help to make it so that new data is correct and clean and they can help to clean up the kind of historical data uh and you can get to a lot of value uh one case this is actually just from the last uh last week or so one agent that was um this company is also a B2B company when they have incoming customers they need to do a lot of analysis and for every single incoming customer uh
Leilei Tong: They saved a lot of time, six and a half hours to do this analysis because the agent could could help them with them. It was like three different agents that had to do things to do that. Another company um that works in the construction sector where they save 200 two million SE per year uh by creating this like valuation report that they need to do in the beginning of projects uh which is like fairly static manual work and doesn't require that much thinking and just a lot of clicking actually. So you can get a lot of value uh and if you think about getting like hundreds of agents that can create this value like in in under three months that is actually immense. Okay. So, uh to be a little bit practical, uh a checklist
Leilei Tong: for what we see uh is needed for companies to succeed. On the people's side, you need champions that have time and will uh an AI lead uh who has the mandate to really drive through this transformation and a leadership team that backs all of this, right? That gives their team the time to do it, that is on board with this, that are engaged, a follow-up and so on. uh on the kind of uh structural side uh we call the rails you need it to be on board of course you need a place for these agents to be right and it needs to approve that platform uh you need to set like an uh governance model for how you govern the agents you know how does it work right who is actually responsible how does budget flow for example uh is it central at or is it decentralized we typically recommend decentralized um and then uh to create momentum you need to set
Leilei Tong: set up certain things structures like regular follow-ups like any projects that you run. Make sure that you do show and share in your town halls not just within smaller groups but in like the all hands. Showcase the success uh and try to show value as early as possible to have momentum. You don't need to have all of this to get started. Okay? You only actually need a little bit to get started. need uh a strong AI lead uh and uh you know a leadership team that is positive and uh that is probably the the most minimal version uh and then what would be good to have then also right after that is champions and a willing IT. So these are the things that you need to get going. Uh okay. All right. Uh yeah some examples of what these like agent fleets could look like.
Leilei Tong: I will just quickly go through this since not everything is relevant for everyone and you will get this presentation afterwards. Um yeah so every company or every function sorry can have a fleet and so far I think all companies that we've worked with every single team actually has use cases for agents. These are just some examples across you know all all different teams. Um yeah this is a sales uh example. I think we looked at a similar one earlier. Uh but yeah so uh in within commercial you can have agents that work with pipeline right doing outbound research looking at um uh signups that come so looking at incoming and prioritizing that uh CRM agents uh yeah that's the one I talked about before agents that help in the actual sales itself so helping to brief in uh before meetings and helping to um summarize and log actions and so on afterwards and making sure that the CRM
Leilei Tong: is up to date and so on. Probably collaborating with the CRM experts. Uh agents that work more with reporting uh and like yeah um goals looking at how well that is going. And perhaps this is more so on the finance side, but like the handover to finance and working with payments or invoicing uh billing these kind of things and yeah this is from an implementation team so typically you know professional services or something like that at at B2B SAS companies so working with the scoping of what needs to be done right doing the statement of work or the scoping document or whatever it's called uh also working practically with quotations to see if there's anything there that needs to be flagged uh
00:45:00
Leilei Tong: and checking the readiness of what's being proposed. Uh the actual building, you can have agents that support with uh the building process of of doing the implementation. Uh agents to help with the go live stage. So like typically you need to maybe generate some testing, create some manuals for the company, do some training and so on. And at the end having some like oversight agents to help with like uh the portfolio of all the projects being done for example and highlighting things that are getting stuck and uh um yeah uh checking the profit or revenue and so on.
Leilei Tong: Uh and then in project and engineering um a lot of examples as well agents that work with backlogs or with uh feature requests uh turning you know requests into like structured tickets in the systems uh agents that actually help with building. We see this in our own company but also with a lot of customers that we work with where you have agents as like junior developers. So writing certain uh more I guess simple feature requests uh code for that uh writing the PRs and so on.
Leilei Tong: uh a lot of reviewing as well can be done and then releasing uh release notes for example writing release notes making sure that documentation is updated nobody likes writing documentation everybody hates that you can have an agent to do it instead um monitoring uh platform monitoring um proactively checking and seeing when there's like error spikes for example um and sending reports on those or like proposed fixes even um agents that keep track of uh security vulnerabilities that are out there, agents that do analytics um and so on. So yeah, a lot in product and engineering as well. These are some examples from a support team. So um triaging tickets, helping to prioritize them, uh spotting incidents and acting proactively on them, agents to help with the diagnostics uh stage. Uh agents that helps with fixing and knowledge agents.
Leilei Tong: I've seen this in so many times where it's like, oh yes, for each support ticket, we need to bring in information from three different systems and it takes us a lot of time to find those different things and bring that together. So having agents to help with that. Um agents that reply or if you don't want agents to um communicate with your customers, you can at least draft those replies. Uh, and having like a learning loop to make sure that you're constantly updating how you work with So you don't have to manually update everything. Uh and yeah, tracking KPIs once again. Yes, those were just some examples from uh fleets that we've seen in different teams. So I wanted to leave you with kind of like a practical Oh, five questions for Monday. It's uh I guess a lot of the people here today are from Sweden and I guess fall break is on Monday. So maybe you can see it as five questions for today or five questions for tomorrow. Um like checks that we think are healthy to do.
Leilei Tong: One is besides the people who build the agents, do other people actually know these agents in the company or is there a central list of where they are? Can each team lead see what the agents within their team cost and deliver? Um, does every team actually have a road map for which agents that they should build? Uh, do your builders actually have the time to finish, not just time to start. We see this so often in companies where you do training, try to enable people but don't really give them the time to finish. Then you're going to end up with a bunch of half finishedish agents and can your agents get access to new system in days or at most weeks but not months because if it takes months you lose momentum.
Leilei Tong: So these are some like check questions to help you think about maybe where you're uh depending on where you are in the journey, right? Uh where you might need to work on um yeah where you need to focus. Okay, I think that was it uh for the presentation.
Åsa Martinsson: Excellent. Thank you so much, Lily. So, we don't have that many questions in the chat,…
Åsa Martinsson: but if anyone have a question, feel free to raise your hand and we will let you out come out and speak. Yeah,…
Leilei Tong: Yeah. And you can also write it in the chat…
Leilei Tong: if you prefer that. Either way is fine.
Åsa Martinsson: of course.
Leilei Tong: Uh just share if you have any questions. No. I guess so.
00:50:00
Åsa Martinsson: All right, it must have been crystal clear. Then we have one question.
Leilei Tong: Uh you can of course ret risks.
Åsa Martinsson: They are coming. They are coming. Of course, it takes some time to formulate them in the chat. What are the top risks that you see lately?
Leilei Tong: Okay. If I think about this as a risks broadly. So not just for example security risks. Um, one risk I see I mean it actually comes down a lot to the the problems that I formulated in the beginning. So one one big risk is the risk of uh shadow AI. The other risk if I think about our customers and like where things don't uh go well or problems then uh one big risk is this flatlining as in X.
Leilei Tong: losing momentum and losing momentum can come from a lot of different places but most commonly I would say it is from lack of time so not having given this proper time uh so then you end up with yeah a bunch of half-made agents uh yeah that's a big one and uh let me think what else is a big risk um I think there's a risk that we see where companies have a tendency to be very cost focused first. So you think about okay how much does it cost to have these agents? What is the token consumption? Uh but then you forget to focus on the value side and if you don't showcase value early there's always also the risk of saying oh we don't know…
Åsa Martinsson: What?
Leilei Tong: if if we want to continue with this or we don't know uh how to invest in this. Okay. Uh let's see how are you placing your I think maybe orchestrating agents placing I'm not sure what placing means um if you want to uh I don't know if the person is still here I want to clarify the question a bit but orchestrating agents for me uh we do have a lot of orchestrating agents orchestrating agents sit on top of other agents and like manage that and that is something you can just do in the platform it's not a nothing too specific that I
Leilei Tong: that I have top of mind I guess on the topic. Um how can we get the slides to the course? Uh well if it's a site for this presentation we'll share it after. We also do an open uh training which you can find on our we can share that in an email afterwards for people uh if you're interested in learning more about building agents. Uh let's see. How should companies think about preparing their company context for the AI engines? Yeah, maybe you're thinking about data. Um because I'm saying that because I see that as a very common issue for companies that we work with. Um and um I think there's there's two things I want to mention on this topic. One is it's very easy to get stuck thinking that you need to have perfect context or perfect data.
Leilei Tong: I see that especially in larger companies that we work with where you have like it's like 10 steps that you need to work on before you have your kind of data models ready in order to be able to do things and that is so slow that I would advise to say uh try to figure out areas in your company where you could start know maybe not on the pilot stage but maybe pilots if that's where you can do it otherwise say like okay within this team like within finance we can do it at least. And then um don't spend that much time on figuring out and solving everything before but rather figuring out what are the agents you need and then figure out what data and what context do those agents need specifically and then you kind of solve it uh
Leilei Tong: through that way because then it actually is in our experience much faster. That is one thing I wanted to say. The other thing I wanted to say is we also see it quite commonly with our customers that they um do like foundational agents first. So agents that help with making sure that you have good data or good context that you can bring in. uh it doesn't have data doesn't have to be data in the sense of a database in MongoDB but it could be making sure that you have proper meeting notes right from all the meetings in the companies that you can do something with it so those kind of things uh how do you decide which agents to prioritize first what was the deciding factor join so sorry if I missed that part uh great question we didn't cover that so you didn't miss anything um it's uh kind of similar to how you do business case
Leilei Tong: business case prioritization I would say. So uh what we do when we do this kind of exercises with our customers we look at value first of all. So what is the value of the agents and then we do like a quick sort on the feasibility where we try to encourage to not think about uh the capability of AI because tech changes so fast but rather just take into account internal difficulties like data readiness that we just talked about or availability of time or data sensitivity questions and so on and then try to find high value quick wins and do those first and then high
00:55:00
Leilei Tong: value slightly bigger projects as the second um that's the yeah short version there are many uses AI how would you describe your definition of agents to other uses of AI ah yes uh great question I guess we didn't cover that and was something that was and I discussed should we talk about what an AI agent is um so yes there's a lot of different definitions of that first of all and our definition of agents is um an AI brain with a clear or with a job description with tools and with autonomy. If you don't have autonomy, you don't have an agent. So the most common use of AI for I guess most people I think is AI chat, right? You have a conversation with chat or with claude or Gemini and you get some uh ad hoc help on something that you're doing.
Leilei Tong: Oh, analysis is sending some data or whatever it is that is not uh um autonomous uh typically and what we try to do with Abundantly specifically uh is that we work on the agents that are not personal but rather for the company. So personal personal agents uh you can do that quite well in places like claude for example but team agents you know the things that work with billing or that work with outreach or whatever it is work with the core processes of the company you want to have visibility and so on and so those team agents are the ones that we are I guess more so referring to here
Leilei Tong: Uh and I guess that maybe kind of half answers the next question. In which ways does Abundy's agent platform stand out compared to using only Microsoft solution for agent management? Uh the biggest actual difference to using Microsoft if I uh think about this from the perspective of the customers that we have worked with we actually are not a Microsoft company ourselves. We don't use Microsoft but we see it a lot with our customers because often they sit with Copilot for example. Um, and the most common feedback we hear from that is that Microsoft just doesn't work well. So that's yeah, I know that's a very broad answer, but it just doesn't work well. Um, yeah, if we was to pilot with a specific department, where would you recommend to start? I would recommend to start where you have the people in place.
Leilei Tong: So where you have a leader that is very in like engaged, passionate about this and willing to give people time to do this because cases exist in every single function. Uh so I think it's more about the motivation and the capacity. Does abundantly save any work that AI agents perform? Main concern is GDPR compliance. Uh yes, there's so many layers to this question. Uh there's of course the question of where if you save something on our platform and this is to some extent something you decide right a lot of the work that is done if that is taking information from one system that you already have to another system that you already have then we don't actually store the data per se right but we handle the data and how do we handle the data well um we have our hosting in our own platform is hosted in Europe and then we use LLMs
Leilei Tong: And depending on which LLM you choose or which other sub processors you choose, they are in various places. But there are ways of doing this only in EU for example if that is a requirement. Um yeah, so we work with companies that have that kind of requirements and that is uh fine. Oh much that sounds like a follow-up question to something. Uh not sure exactly what that was referring to. Uh how do you find the right role task for AI agents? What may what way do you use to choose this task is good for an agent? That's a great question. We do we cover that a lot in our trainings actually. Uh so we actually have a whole process of teaching people how you think about AI agent use cases, but it's um it's easier than you might think.
Leilei Tong: It was a lot harder because a year ago it was much more narrow the kind of tasks that agents could do and today I would say it's much broad much broader. So a lot of the things that humans can do that humans don't want to do I think that's the baseline of where you start. That's the easiest place to start. Anything that you do today that you don't want to do, um, is where we start and then once you start to do it uh you get a pretty good sense but I would say limitations are not that many the technical limitations of what an AI agent can do uh as in like from an AI agent capability point of view is not that much anymore but there can be a lot of limitations in reality around more like data sensitivity and data readiness and so on.
01:00:00
Leilei Tong: Uh but when it comes to roles and tasks, I think um uh um I almost want to say almost anything where you need to think is more so…
Leilei Tong: how you divide the work. Uh is this one agent? Is it several? How should they collaborate? That is actually the harder part these days. Uh a lot of the work they can do.
Åsa Martinsson: Okay, great.
Åsa Martinsson: I think that that sums it up.
Leilei Tong: Yes. Yes.
Åsa Martinsson: No more question and we are right on time. It's excellent. So if you have any further questions or if you might want to see a demo of the platform, please reach out to us. Don't be a stranger. We love to talk about how the uh platform can help you and also showcase how different clients are working with us today.
Leilei Tong: Yes. So, don't be a stranger. Please reach out. And thank you all so much for the nice emojis. We appreciate the positive positive feedback.
Åsa Martinsson: Thank you.
Leilei Tong: Uh yeah, thank you all very much. Thanks for attending. Cool. See you all. I wish you all a great day, great afternoon and evening for the one in Australia. Bye.
Meeting ended after 01:54:47 👋
This editable transcript was computer generated and might contain errors. People can also change the text after it was created.