Agentic AI Use Cases with Real ROI

00:00:00: We are now really checking on their processes and hey, how can we automate it?

00:00:04: Sometimes the solution can even be not with an agent because you just build better IT systems that support your process.

00:00:16: It will have a completely different UX.

00:00:18: It's going to be combination of clicking next best action.

00:00:23: so there'll some kind of support agents.

00:00:27: people need to be involved and need understand the output of AI, take a right decision.

00:00:34: I disagree with you but we said that Google is not... Maybe in last one year or two years maybe they were not.

00:00:42: innovation at forefront for somethings.

00:00:44: The boardroom memo powered by Netconomy.

00:00:52: Welcome back to the bathroom memo The Netconomy podcast, where we cut through the noise on enterprise technology and focus ROI.

00:00:59: We'll move past the hype of two thousand twenty-three and two thousand forty four into their chanting era Of two thousand Twenty six joining me are to people who live this every day Boban, Joe David's our AI and data department lead And Maximilian Tornotage Our head of technical enablement.

00:01:24: Let's get in to it I would say.

00:01:27: when the trade deputy got released i think that The next year was a lot of these kind of up.

00:01:33: So everybody wanted to do a POC, so it doesn't really matter about which topic.

00:01:37: It's only like I need an AI use case and show this to somebody.

00:01:42: but this led to having POCs that didn't prove successful.

00:01:47: Some kind of percentages are extremely high like ninety something percent not being successful.

00:01:53: But over the time, it's that customers also in the market changed a bit.

00:01:58: So its more like do we really need this?

00:02:02: It was POC era then last year to the agentic era.

00:02:06: I still think that whole value is not yet proven as high already but yeah... This year i think expectations are very high and i think CEOs who are losing patience would be liked hard find their values for them.

00:02:23: I mean for me what where i see a lot of difference now is.

00:02:28: in the beginnings we had like this chat gpd wrapper which basically just used the api of chat GPD and doing some nice system prompt on top.

00:02:39: At the end, The value was quite limited to all those applications.

00:02:46: but Now We are also going A bit deeper with agent architectures Basically Where we really hook into into the core of the business or our e-commerce shops and try to solve real issues, not just having something that you can throw in a chat at the end.

00:03:06: And get to the same result.

00:03:09: for me also as you said The value is now lot more asked.

00:03:16: basically we are going alot deeper into what can we actually achieve with this technology and not just hype around, in my opinion.

00:03:29: I think part of the story is also about everybody was expecting like a few days you are done so traditional machine learning You need months maybe will get some value And In these I just create something for hours or two days.

00:03:44: i have it on production its working which is far from reality.

00:03:48: And I think like many clients really were disappointed because maybe you can achieve something very fast in a couple of days, but to productionalize this and have everything running as secure environment or at scale for the thousands of customers... In the end it just shifted to governance infrastructure and microservice layer.

00:04:14: building agencies Kinda easy, of course it still has challenges but the things around that are not less complex compared to traditional software development cycles.

00:04:27: Yeah I think also what we experienced in recent years is building AI agents at the end isn't far away from traditional software.

00:04:37: engineering started somewhere with machine learning engineers, building machine-learning models.

00:04:44: But it turned out that the LLMs are powerful enough in order to have more... a lot of use cases I would say covered without any touch off the Machine Learning model itself or other model itself which is quite interesting

00:05:03: yeah but also at some point i think in the beginnings we had like used LLM for everything So like, don't even write a deterministic logic.

00:05:13: Just let the LLN do it and initially if you just look at the results I mean LLn by nature are pretty confident what they're saying And you were just like oh this is perfect works perfectly.

00:05:22: i dont need to code anymore This stuff but then when you dig under the hood You see that its completely wrong the results that it's spit out.

00:05:31: So then we also had this kind of deterministic workflows, which is still I would say the most important stuff That you just really have a clear separation.

00:05:40: What is a workflow?

00:05:41: Where do i need determinism?

00:05:43: where Do I need some lm magic to actually get someday Which Is not possible?

00:05:48: yeah exactly I mean that's something where I Really discuss A lot with my engineers Currently Because My Engineers Tends To everything as deterministic as possible and then LLM somewhere in the middle of The workflow.

00:06:04: very tiny Very, very testable approach at the end.

00:06:09: But I think this is also something we need to learn over the next projects.

00:06:16: Where's the cut between?

00:06:18: Deterministic workflows And how free can be provide.

00:06:21: let the LLm run again in order to still get

00:06:25: a

00:06:25: good enough or the perfect output at the end.

00:06:29: I also think for the engineers it needs to be this kind of mindset shift because they're really used.

00:06:35: as you say, these deterministic workflows like coding in a way... This is maybe separate topic from vibe-coding and stuff but we recently had situation where we have build API services for prices and availability are not working.

00:06:50: We were now waiting for clients.

00:06:51: four weeks give us access.

00:06:55: why don't we just create a mock service and just white code this in like one hour.

00:07:00: And it doesn't really matter for the use case, they were like yeah sure we can do but please don't evaluate the code quality or judge me.

00:07:08: I don't care.

00:07:09: We will throw these away.

00:07:12: when you get to API we integrate them.

00:07:15: Do that?

00:07:18: That's for sure.

00:07:20: One way we need find ourselves How much test table is something at the end or how perfect it does to code need-to be and When is it okay?

00:07:33: To just write throwaway products.

00:07:35: Yeah,

00:07:35: touch ones are or never

00:07:39: True?

00:07:40: mean also I see me based on the Google next as we attended.

00:07:43: I mean also Google is investing a lot into this whole agentic game as everybody knows.

00:07:49: They now did this whole rebranding of Vertex AI to Agent Platform, so this is also like going in a direction which I was talking initially... Like have the surrounding things around the agents Plugable as possible in a way so that they can just focus on agent building.

00:08:07: and then you have agents registry.

00:08:09: You had all the monitoring in place All this stuff because This is what where most time was spent also with their Google ADK, which has really integrated into their ecosystem.

00:08:20: Really until we you can very fast deploy the agents And make them work at scale.

00:08:26: I like their approach.

00:08:28: Yeah especially What makes it Really interesting is this scaling from pro code agents where you can build everything, define everything.

00:08:38: It's registered in the agent registry but also the no-code agents which are on the Gemini Enterprise platform built by our business users at the end are registered there and you can provide them access into every kind of GCP service

00:08:54: etc.,

00:08:56: Which... Is the first time I personally see that we could probably bridge business user build and automations

00:09:06: with

00:09:07: actual engineering work in one platform.

00:09:11: How do you see this plate in the future?

00:09:13: In a way, how much would be pro code?

00:09:15: How much drag-and-drop and clicking click...

00:09:18: That will be interesting!

00:09:27: really interesting and business user tried to get into it but at the end you still need a lot technical know-how in order to properly build such workflows, understand them.

00:09:41: And since Google is now going into the direction of saying okay we have an agent builder and Gemina Enterprise but want keep as simple as possible for business users to use The pro code with Google ADK, the agent development kit where we can build nearly everything as engineers I think it will be even like seventy-thirty percent a split Where seventy percent are most probably even their business users because they know what in need.

00:10:16: They know What?

00:10:17: But they want to achieve and this somehow can built into our inner workflow with guidance.

00:10:22: Probably but they can't build it.

00:10:25: Thirty percent are really the difficult hard nuts we need to crack as engineers, where just it's not possible because the connectors are Hard-to-achieve.

00:10:37: We have still a lot of legacy software in our companies Where we need To attach them where no mcp server everything or anything is available Or we have workflows that are just going over multiple steps with many involvement of other systems where you need to prepare the context in a very, very refined way and that won't be possible I think for any business user.

00:11:05: To achieve.

00:11:05: Yeah true!

00:11:07: I'm also thinking there's more requests.

00:11:10: we see across the company about wide coded apps like small scale only for my department or only four few people which I think, we are talking about AI a lot but in the end.

00:11:23: We all just want to make our people more productive and save costs on increasing efficiency.

00:11:30: so-on with this wide coded apps there is also huge potential because not everything needs to be agent.

00:11:38: maybe somebody has just applied code an app two three days platform where they can just easily deploy this and make accessible to users in a secure way, I think would also bring these performance boosts that we're all expecting.

00:11:55: But don't you think then push the people away from their core work actually?

00:12:03: So there are now fumbling around with AI tooling not doing their actual work.

00:12:14: So what do you think?

00:12:16: Is there any...

00:12:17: I mean, how i see the people is a lot of motivation for me to try this out and many people are even experimenting outside because it's interesting.

00:12:27: There're really natural born builders so maybe they aren't coders but like build things.

00:12:33: Of course everybody should just do in free time But if now can invest One day, two days for one year to say from my department I don't know hours or days of work per person.

00:12:50: In the end that's a core benefit because in the end during the software development we had this kind of problems where developers didn't really understand clients and domain experts.

00:13:01: but now if you're bridging these gaps bringing capability for the domain expert overblowing it like they are now spending months on building this, but something really small scale which makes sense that they can just easily deploy.

00:13:19: It will provide a lot of

00:13:20: value.".

00:13:21: But for me exactly that is also the challenge we have because with technology where people who by far aren't used to building software That's fully true, and we saw it also in our company that this is now happening or we see actually.

00:13:49: But still the outcome of these tools are built not what I would call production-ready something you should scale on a whole company And this is something we are also currently working on with management and so on.

00:14:11: We need to define a way how to provide such tooling for the company, because I think it's fine.

00:14:21: if you build that tool then use your own.

00:14:24: If it works for you its perfect!

00:14:28: Then maybe two or three other users see hey someone built very nice tools.

00:14:34: I also want to have it.

00:14:35: Now you come into this deployment problem, i would call it like... You need to change your application that works for multiple users and you need to deploy on a platform that is accessible.

00:14:47: Probably you will have some database somewhere flying around because you want to store the data And then things start getting difficult especially for users who are not used to building such as software.

00:15:04: And at this point, also software engineers need to hook into the process or operations engineers.

00:15:10: At least to support and guide people.

00:15:13: This is something where we are currently thinking about multi-tier projects Where you can say that if there's a small application You want it for few users.

00:15:24: Feel free!

00:15:25: There's platform deployed with some readme.

00:15:30: you will figure it probably out, but if we want to scale it to a company level.

00:15:35: You most probably still need some engineering work in order to make it multi-tenant capable and scale at the end yeah

00:15:46: true I mean for the company scale that for sure cannot live within one department and on one person because there's also so many things to consider.

00:15:56: I mean with these smaller apps, all of them even like a lead level engineering leads have some alignments about the stuffing.

00:16:03: this is still used really in Excel filling this out, populating all of the stuff and for these five people if they can do things automatically just a nice UI where their click-and-clack thing.

00:16:16: And then you make it visible... Of course we think about agents doing that but I believe even at next level should try to see what am i trying to achieve before introducing an agent who is fully automated.

00:16:33: But I think this is also where the companies are quite stuck currently because we currently have to promise in a market like, hey just build an agent.

00:16:46: It will solve it for you.

00:16:48: but for building and agents We need to have data foundation And In addition we also need to understand The human process in order To build up those capabilities.

00:16:59: that This Is i Think Where most of the Companies Currently are stuck in the process of figuring out.

00:17:09: And also, we stuffing processes.

00:17:12: I think one of the best examples in our company because it's very human driven.

00:17:17: We have department leads that understand how what people we have and this whole Department and we Have a lot of discussions until one of our projects is stuffed to find The Best possible way for the project at yet.

00:17:35: And there is a lot, many attributes included in this process.

00:17:41: Like okay what are the skills of the employee?

00:17:45: What's the situation with project employees currently in and what will be shifted to... Do we have location-specific restrictions, do we have language specific?

00:18:01: Motivation

00:18:02: of the employee.

00:18:02: Exactly things like that.

00:18:04: and This makes it hard because in order to build an agent That can do It We need To Have this data somehow available.

00:18:15: so we Need to provide it.

00:18:16: And this is really hard Because its In The Brain Of People Not Somewhere In Our Systems Right Now.

00:18:24: True, true.

00:18:24: I mean also you know customers now are getting smarter of course and they would like to have a concrete KPIs behind everything.

00:18:32: okay how much of performance boost will i get if i get this agent?

00:18:37: but is that exactly what you said?

00:18:39: They're not even currently tracking.

00:18:42: How much time do they spend?

00:18:43: so we take as simple example Like i don't know recruiting And how much HR spending on the first screening phase we would like to maybe introduce an agent that is doing the CV screening and then giving advices to the HR, may be doing some kind of a not even grading but some kinda helper for the HR.

00:19:03: We don't know how much time you spend.

00:19:04: now many companies are not tracking.

00:19:08: they're expecting.

00:19:08: okay please you need to promise us we expect KPI or twenty percent performance boost which always chicken-and-egg problem.

00:19:19: stop the innovation before we get all of that data right, but still it needs to somehow make sense because in the end We are just ending up.

00:19:27: you know God feeling this will probably help me didn't really help.

00:19:32: You don't know what think.

00:19:34: it is some people thinking they doesn't.

00:19:36: so yeah

00:19:38: But for me It's also interesting too.

00:19:40: see if Because This Is I would say The first time at least That i saw that companies Are trying to push so much for return of investment.

00:19:53: Because

00:19:55: in earlier days or before AI basically started, there was a project.

00:20:00: we built some features and barely looked at KPIs or thought about what are the KPIs?

00:20:08: We will measure as soon as we implement this feature.

00:20:12: And now every agent use case is okay but how much more efficient can you get with that?

00:20:21: And that's

00:20:23: also what we saw with one of our customers, where... We started building up agents for internal processes.

00:20:36: The first thing was okay?

00:20:38: What is the efficiency gain we get?

00:20:40: and really needed to ask customer firstly how efficient are you currently?

00:20:49: We don't know.

00:20:51: Yeah, it's really the fun thing!

00:20:53: Exactly I mean also when you go into these kind of workshops and everything.

00:20:59: so when you are discussing about a process how much people you need to understand the process end-to-end even the people from the same department.

00:21:08: It is hard for them to drill this okay?

00:21:10: This is what the whole flow looks

00:21:12: like.

00:21:14: People often think that they get surprised on their workshop process driven design and drawn, they're confused that it looks like this.

00:21:26: But still what I really liked is we now somehow started to think about the process efficiency of how we can automate more processes with actually computer systems because every company, users computers and IT systems.

00:21:48: And many of our bigger customers or all of them actually have ERP systems where you think a bit about automation.

00:21:56: but now we are also heading into different other departments like marketing... checking on their processes and hey, how can we automate it?

00:22:09: Sometimes the solution can even be not with an agent because you just build better IT systems that support your process.

00:22:18: You see

00:22:22: more potential in these kind of internal use cases when it comes to agents and AI

00:22:29: Compared to compare

00:22:31: for the clients like four four in this industry For example, for the retail business for the end consumer or more.

00:22:36: on the internal efficiency gains.

00:22:41: I think Jumping into AI for retailers and our e-commerce shops is something we cannot And should not avoid.

00:22:52: This is something where you need to do?

00:22:54: I think Daniel mentioned it in there in a podcast recently.

00:22:59: He also said, this is something that will just happen and the companies need to jump on this train in order to survive.

00:23:07: And also continue selling products at the end for here.

00:23:14: The efficiency gain is most probably on the end customer side researching product things like or

00:23:20: experience maybe better experiences.

00:23:24: but there's a huge potential when it comes to internal process automation with AI.

00:23:31: And sometimes I tend to say, It's also easier to start with because if you have an e-commerce shop and you serve that two or few thousand clients every day building in the AI agent at this bulletproof...I think we have a lot of cases where these didn't work out well.

00:23:54: but Internally, you have your trusted user base basically.

00:23:59: You can also scale and try it out and find use cases that are small enough in order to start building gain experience... ...and then can scale bigger in order really having an end-to-end flow.

00:24:18: I've seen those from my natural role as automating our own company, I see a lot of potential and other use cases internally.

00:24:31: And i tend to say start internally gain experience because it's safe environment.

00:24:38: then we can always scale the end customers.

00:24:41: Yeah!

00:24:42: I also think that somehow customer organizations in general tends be too how to say it, like they're looking for the golden AI use case that will enable them days of increased efficiency across their organization.

00:25:01: And I think if you just and there's get lost into this kind of a search for these use cases or is Golden Use Case which doesn't really mean maybe exists but until you find it until implemented probably its something much more sophisticated.

00:25:18: If you are just focusing on small gains, maybe these departments say a few days this for some other use case.

00:25:24: Few days or smaller things?

00:25:26: These things add up in the end will achieve much more than just running around in circles having pool of fifty use cases and asking yourself okay what which one should I pick?

00:25:38: i don't know.

00:25:39: yeah The interesting part here is we're always looking at use cases from processes We currently doing And I think internally we have a very interesting project in agent areas, where... We actually built an agent for process which was before not possible to achieve and not working.

00:26:04: Because it took so much time that we didn't really fully do it!

00:26:13: So what did is have in our pre-sales area the problem that we are faced with a lot of company names basically and We have somehow an understanding what our ideal customer is.

00:26:32: So, basically we Know okay?

00:26:36: A customer is on certain range of revenue it has a certain structure And technology portfolio, then it's a good fit for us.

00:26:48: Now you're faced with one thousand and more customer names And you need to choose what customer might be the next best customer for us Something that is not achievable or not easily achievable in With just human work.

00:27:05: So why we basically Choose customers based on personal connections to certain customers, but did not really go into data-driven selection of the next best customer.

00:27:22: And this was... The first use case we actually did with an agent now where then I said okay We can do something like deep researching within agents in a matter of minutes instead someone researching over days for a single customer.

00:27:41: So what we really achieved is, We can provide the agent a list of names.

00:27:50: One hundred two-hundreds and The agent can come back in a few hours

00:27:55: with

00:27:56: completely scored ranked lists Of potential customers which are shaped for needs that we have on our customers.

00:28:09: We're not able to do before and even with AI help.

00:28:13: It was.

00:28:14: it took us a month in order to get two such lists Before without

00:28:18: building the yeah, exactly.

00:28:20: Exactly.

00:28:22: And now that With we are really able to say okay?

00:28:27: We have a new list.

00:28:28: one hundred customers throw it into into the system The agents go on a deep research journey in different areas put in a new scoring for compared to the existing list and we know, okay those are The one hundred next best customers for us.

00:28:47: And this was really an interesting journey For us because it was first of all the first journey We did on Google ADK and second the First Really big agent system we built and It turned out to be actually quite successful Because we now can do something in ours which was not possible to do in months now.

00:29:09: What is the feedback from salespeople?

00:29:13: I don't know how deeply are we evaluating these results, of course you can not really cross check everything but at least for customers that already knows it's not

00:29:34: perfect.

00:29:35: We're still talking about deep research and AI systems.

00:29:41: It's not perfect, but it is good enough to give us an indication where we could look into next And its also good enough for comparable results between the customers potential customers which have.

00:29:59: so For Us Its a very valuable system already.

00:30:03: also the sales, it was in very near collaboration with the sales department.

00:30:11: And although they told us this is really valuable to continue and we also agreed together management that will invest more into it.

00:30:23: actually

00:30:24: I'm often asking myself like what means good enough?

00:30:31: You could stop in a way, because you can improve this indefinitely.

00:30:36: Until when do we say okay let's pause it for now and continue if technology gets better?

00:30:48: I think good enough is that what are going to be done with the list next?

00:31:02: Now we can introduce more third-party API integrations like providing linked in data, providing more technography data.

00:31:13: We really know that this customer is using SAP CRM products things like that.

00:31:22: but at some point you also figured out okay we know it from googling because that's what deep research does.

00:31:28: It goes to Google tries to find a match between the company and technologies that are relevant for us.

00:31:36: And most of the time it works at hand, this is also where we need to understand how make this cut because some point when you say okay now have these one hundred customers want approach them in some way And then the actual work of our sales department starts, because you need to understand the customer.

00:32:03: You have to find context and get in touch with them.

00:32:10: This is also where people start researching on their own as well Because this understanding isn't just getting AI output at the end It's really like investing time into... Into this customer or a potential customer.

00:32:30: Yeah,

00:32:31: I guess it's also like you know until the eighty percent You get pretty fast.

00:32:37: then The remaining next ten percent is kind of difficult and the next ten Is even more difficult?

00:32:42: And i guess also It's on the management and CEOs to decide okay when is the border?

00:32:48: and i think Also now it will go a bit into these kinds of what do we say new features but to evaluate Why do we need it and what do you expect to gain from?

00:32:58: It because maybe I think customers and organizations can really go into this kind of a trap.

00:33:04: Let's try to get something more.

00:33:07: But if you now need to invest additional, I don't know one month To get up fraction of a percent of increase does it make sense.

00:33:17: Maybe in this case skimming through LinkedIn If we don't have it could potentially makes sense.

00:33:22: but if we want not to go I don't know, social networks to see what the companies are discussing about all this stuff may be in there.

00:33:30: There is not a huge value compared with some other topics.

00:33:33: so it has to be prioritized.

00:33:35: Also we should go into that trap and say We want replace people here.

00:33:40: That's not our goal.

00:33:45: eighty percent with AI is most probably the right fit because that last twenty percent, it's a percentage basically where people need to be involved and needs understand output of the AI.

00:34:00: And take the right decisions.

00:34:01: Because overall if we would provide AI Decisions or decisions fully to AI I don't know If our company will survive at the end.

00:34:13: Also there are multiple studies around, could an AI lead a whole company and manage the whole company?

00:34:20: And we're not at this point I would say.

00:34:23: Yeah

00:34:24: i think it cannot even... At these stage really maybe some specific roles of course but in the end i think The biggest fear of people Realistic fear In my opinion Would be like My role will change and I'll do somethings differently.

00:34:40: yeah..and Maybe this differently I don't like it.

00:34:44: Because for example, maybe if i'm a designer and was drawing working in Photoshop doing everything manually all the layers or image changing And so on For now we still people are professional great image generation.

00:35:02: you need to use Photoshop but now is different.

00:35:04: Now you have AI.

00:35:05: You circle this say...I need change this.

00:35:08: Please move around.

00:35:10: This is a bit of difference work.

00:35:12: And I think some people, maybe it's not the best example but I think simple.

00:35:17: Some people are also reluctant to go in this direction because they like their job.

00:35:22: Also we had you know... People browsing through your website looking for sources and materials trying to get into these kind of creative mode.

00:35:32: But now AI already did this For you anyway Because it has access all over data.

00:35:38: Then if just ask a question You would probably ninety percent good response.

00:35:44: And it will be different, but some people maybe have a joy of going deep into different websites leading to the blogs investigating in topics and so on.

00:35:54: So this I think is big bit of mindset shift also for people that are needed.

00:35:58: But from me It's like just going deep onto topic.

00:36:05: This was sometimes at least the necessary part in order to gather real understanding of the matter.

00:36:13: So, I don't know...I'm into mountain biking.

00:36:17: if i would just throw things into AI deep research whatever for next best mountain bike will most probably not understand that.

00:36:31: geometry or frame what forks are very well currently things like that and this is something.

00:36:40: if you have a hobby or Something Like That You most probably want to jump into that.

00:36:45: And that's Where I think also Sometimes, you just need To go deep and do it on your own in order to fully understand It.

00:36:57: but sometimes If i'm not Into that like um...I don't know coffee machines I'm not into the coffee machine so much, So i just want to have a coffee machine that makes my coffee in the morning.

00:37:10: Then i would use deep research because hey what's the best price for the best coffee machine i can get?

00:37:17: Let us go with it!

00:37:18: Yeah...I mean My workflow is different.

00:37:20: For example Because i am into Coffee machines and i like To use Deep Research But as an initial trigger And then i go deeper and deeper Into these topics.

00:37:32: Maybe it can happen that I don't even open a website for this.

00:37:36: Maybe now see what tools do we need to make a perfect espresso, so i'll continue the interaction with AI and ask things about that in my way of understanding what is relevant also maybe trigger another deep research to look at it.

00:37:58: I mean, i'm losing.

00:37:59: maybe you know all the nice images and different stuff but in a way...I also say a bit of time not going through some content which is there's always like boilerplate content in all these blog posts and information etc.

00:38:12: so..i just take it as a different medium for example.

00:38:16: So your trusting AI?

00:38:18: Deep enough that answers are correct enough, let's say like that because it won't be one hundred.

00:38:24: I

00:38:24: mean if something really important for me which i don't know is about safety or something like that... I for example tend to also trigger a deep research focused only on scientific publications and things around that And based on that you get also feeling That this is much more correct Because If You're Just Focused On The Blog Post anybody can write a blog post.

00:38:49: I mean, also for the research.

00:38:52: that also can be true because there are many researchers which don't really make sense.

00:38:56: they're maybe some like tested on one hundred people.

00:39:00: but if it's very important then i go and open publications scrim to the abstract.

00:39:10: not every scientific paper is equal in its references.

00:39:16: that's already maybe too deep.

00:39:18: But I think we went a bit off the topic in general, but what do you see like for clients?

00:39:26: What are their main use cases they're interested in?

00:39:33: Since coming from traditional e-commerce most of our customers until now approached us with e-commerce features and an agentic commerce at the end, which is very relevant.

00:39:51: And we also need to be experts and gain a lot more knowledge in that area.

00:39:59: but now I have the feeling you're involved with many of all projects.

00:40:05: actually The clients are really coming into internal automation Workflows.

00:40:17: where can we hook and also there I have the feeling Clients really ask for guidance.

00:40:25: Yeah, I don't know what's your.

00:40:27: you're a feeling on that?

00:40:30: For sure We all had recently a situation were client head like really nicely identified use cases.

00:40:40: there is even some kind of a indication, you know costs or expected effort saving and stuff like that.

00:40:47: but it's like fifty use cases.

00:40:50: And I mean this also done between the departments.

00:40:53: so in theory we can have some kind for ranking list.

00:40:57: But to really properly evaluate impact i think these are where they need our support.

00:41:02: most often We are not maybe that deep into organization, but just because of our experience we're looking at the use cases and like challenging this from out perspective.

00:41:14: You really see something.

00:41:16: what will work?

00:41:17: What would not work?

00:41:18: Maybe it's where technology is not ready yet or may be something they can just use a tool to build custom use case.

00:41:25: so... And in lot of cases I think there also these kind of struggle.

00:41:30: Do i need chatbots Or do i need workflow?

00:41:34: Do I really need this chat experience?

00:41:37: Exactly.

00:41:38: And especially, the chat experience is something... Yeah okay we have large language models which are language models so that the chat experiences quite a natural thing to start with but also Here, I think we now need to learn on how to get a bit away from this whole chat experience and providing better more human-friendly interface because everyone of us is bored reading through all these AIA messages.

00:42:17: There's even a trend called AI fatigue where you really... tired actually of reading through it, and there are many AI vendors that now looking into different approaches on integrating AI outputs or providing AI output.

00:42:45: I think the number one use case they see is that clients in our industry interested is chatbots maybe sometimes not even a vision behind what these chatbots will provide to me, because there is still not a proven value.

00:43:02: I mean they are numbers you know twenty percent of conversion rates and all this stuff but you can never... You need put it into context.

00:43:10: really see.

00:43:11: does make sense for me as an organization or not?

00:43:15: And somehow its bit natural also the competition.

00:43:18: maybe my competitor now has a chatbot.

00:43:22: do I really need it?

00:43:23: Is my search, for example regular search working?

00:43:28: is my product data ready for this?

00:43:31: because if you have ten attributes for each product and no descriptions the chatbot cannot help.

00:43:39: It can guide your bits but if you ask any kind of a specific questions hey i'm looking for address which is good for wedding has some kind of luxury vibe.

00:43:51: if there is nothing mentioned in the product data chatbot, it's not magical to just you know do this stuff.

00:43:57: of course we can now discuss about.

00:43:59: okay we're gonna trigger a google search.

00:44:02: We find that they are but this is like over complicated

00:44:05: yeah sure especially more bad data because googling can also go horribly wrong at the end.

00:44:17: When it comes to Googling and bad data, we had in the beginnings really few cool use cases which are around product data enrichment.

00:44:26: you have for one client like implementing production product description generator based on the attributes, but also this is where they have a flexibility to do some kind of configurations.

00:44:37: The tone of voice style.

00:44:39: so it's not just as simple block-of-text.

00:44:41: It really quite complicated and This already something which will give them additional boost.

00:44:47: But also when comes like really researching the product attributes because many customers are struggling I don't know, twenty-thirty attributes per product.

00:44:58: They didn't even know where to start.

00:45:00: Some companies simply don't want to invest time because it's definitely time consuming But for some other companies It takes a massive amount of time just because...it is simple workflow.

00:45:11: You get an new iPhone.

00:45:13: If your supplier doesn't provide you exactly the attribute.

00:45:17: they are googling this or reading through materials and saying okay This makes sense?

00:45:22: Doesn't make sense.

00:45:24: It's an error, prone and time consuming process.

00:45:26: And we got some use cases.

00:45:29: in the end on a POC stage where you can do deep research about product attributes maybe sometimes there will be a clash for one attribute that you need to select as human.

00:45:41: which of them makes sense?

00:45:42: You'll get references probably believe more official iPhone website than local dealer And this is something which I think provides even better value.

00:45:54: The problem with this use case, for example when we build it the connection will be actual enterprise system because currently how we built was working in isolation.

00:46:04: you can really do it nicely and have export maybe in CSV but since whole human-in-the loop thing to validate see check some stuff cause its cannot fully automated.

00:46:17: yet This is where then the real investment comes in because building this kind of an agent Is easier than having it integrated.

00:46:26: and to end when I accept these that It's already my PIM.

00:46:29: They're all ready on my website.

00:46:31: So, this is where the complexity also somehow its hidden but it huge

00:46:35: especially this human in a loop Problem i would even call at is still something we need To figure out.

00:46:42: what went to pull this human into the AI process.

00:46:47: But for me, that's a very interesting use case because if first of all you have work.

00:46:56: That is very manual and requires a lot of time in order to get this attributes right?

00:47:04: And the second thing Is If You Have Your Attributes Right Most Probably your search will also Work Better.

00:47:10: You Have Already In Your Shop Also Your SEO Ranking As Well And that's, I think for me the right way to think about AI features.

00:47:24: Not hey everyone else has this chatbot?

00:47:27: I also need a chatbot more like where can we really make a difference for us or our customers?

00:47:35: at the end Also had one customer That was kind of bored off there of the Chat interface and then be Also here in the PUC.

00:47:51: build up a system that recommends order improvements, order efficiency improvement.

00:47:57: Like next

00:47:58: best action or something?

00:48:00: Yeah exactly and this was also quite interesting because we hooked it completely different into the interface.

00:48:06: We did not provide any chat-like interface for now And just said okay you're on the card!

00:48:12: We know your buying other stuff from time to time.

00:48:17: maybe you also want to add that to the card now or You are ordering in different clusters, same product.

00:48:27: Maybe you want to combine it in order to save delivery costs and handling costs.

00:48:35: I think this is a smart way of thinking about AI features.

00:48:41: Where can we provide the feature for customers?

00:48:45: get the customer into the AI idea.

00:48:47: True, true but there are also customers on the opposite side of a spectrum where they're like.

00:48:53: I don't like the chats as it is.

00:48:56: in a way There's a chat bubble.

00:48:57: by clicking here and getting small thing.

00:49:00: i would like to have half of page now in a chat interface or full-page.

00:49:07: i think probably the way how do we interact with our web shops will be quite different.

00:49:15: You know, maybe you'll get into some kind of a. this is just my expectation potentially or in the next two years that it would have complete different UXs.

00:49:25: It's not for sure.

00:49:26: It will be combination of clicking Next best action.

00:49:30: so there Will Be Some Kind Of Support agent or somebody from the AI observing what we do, providing next best action.

00:49:40: I can ask questions.

00:49:42: it can recommend products for me.

00:49:43: so i think this whole dynamic UI changing and shifting is somehow future.

00:49:49: let's see if you get there how soon that will come but could be quite interesting?

00:49:57: building a prototype, just providing a prompt box and nothing else.

00:50:00: Let's see

00:50:01: how well this will work!

00:50:03: But now we saw the whole training of models with Corsair so that it can do actions for you in a way... With a Corsar voice not even text maybe... Maybe Neuralink?

00:50:16: Exactly!

00:50:17: Browsing

00:50:18: shop our stuff.

00:50:22: For sure this is going to be much stronger talking to Siri on your iPhone, recommending products things like that.

00:50:30: This is for sure a much stronger say this potential

00:50:34: in the future.

00:50:35: How do you see generally Google position?

00:50:38: In all AI game and are we under right track?

00:50:41: betting on Google?

00:50:43: It's always back-and-forth I would say when it comes to selecting a vendor because sometimes vendors much upfront in this game, like OpenAI was at the very beginning.

00:50:57: Now people or companies are a lot nearer than even before OpenAI.

00:51:06: but what I also saw on Google Next is that Google has solid foundation and we We are also early adopters of this whole platform.

00:51:22: We jumped on the platform right at the beginning and we also saw their evolution over the years now, And Google was maybe or is very often not the first one to provide something Also entropic and so on in the market.

00:51:43: but Google always providing capabilities on a solid level, where we can also count on them.

00:51:52: Also with Google ADK, Gemina Enterprise and so on And the CX stack.

00:51:58: now We have really a platform Where we can fulfill I would say Ninety percent of customers' needs Either by just configuring their products or building it on the platform.

00:52:18: Yeah, I mean i disagree with you but you said that Google is not... Maybe in recent one year two years maybe they were not with innovation at a forefront for somethings.

00:52:30: if we take look at nano-banana this was again changer to everybody.

00:52:34: This like sooner With LLMs over course of last fifteen years.

00:52:41: they were always, in my opinion the innovators.

00:52:43: If you compare to some other companies like Microsoft or AWS I mean all AI innovation came from Google Transformers.

00:52:52: it came from them.

00:52:53: so all the topics...I think They made a mistake.

00:52:56: maybe that took this whole LLM game again too likely in the beginning.

00:53:02: I think that whole size of their organization was something which is blocking them to be there or at a forefront, so OpenAR recognized the pursuanty.

00:53:12: they had smaller team much cleaner as smart on Google but didn't have this whole anchor for big organizations.

00:53:21: stop innovation

00:53:25: fully agree.

00:53:27: Also, we always need to understand that Google is not a small scale startup.

00:53:34: And also people at Google told us they have never seen such a dramatic shift as it now happened with AI basically in their organization because every part of the organizations there are going into AI which is really interesting and They always proved us that they can deliver in form of LLMs with Gemini III, three dot one.

00:54:02: The models are a lot better and comparable to any competitor on the market like OpenAI and Tropic models.

00:54:12: I mean they also have it end-to-end.

00:54:15: In a way they close the loop because their GPUs deeply integrated into ecosystems across the world.

00:54:23: They have tons of people and they had money.

00:54:25: So in a way, I think...I really think that on good track we are their partners.

00:54:32: but maybe compared to SAP and this whole collaboration because we're also SAP partner how do you see the collaboration?

00:54:42: When is Google going to use SAP?

00:54:44: Exactly!

00:54:46: I think SAP will always be needed strong in their product area like ERP, CX and so on.

00:54:56: And for me it's not we do one or we do another.

00:55:03: It will be a good synergy at the end.

00:55:06: So We can build very good working and customer tailored agents on Google, connected hopefully with SAP at the end.

00:55:16: And SAP knows how to interact their systems.

00:55:20: so with protocols like A to A things might have a very well-working synergy in future.

00:55:28: Yeah exactly because I also think that in the future building agents will not be there difficult In the end as SAP let's be honest came into AI game a bit late But they're catching up.

00:55:40: I mean, last year in the SAP Techhead They announced all of this openness, agentic platform also on their end.

00:55:47: So as you said through A to A and collaboration between agents In the end we'll probably have some kind of an agent pool Of agents across different mediators that We don't even need to do All these big data pipelines Across systems.

00:56:05: And so maybe Let's see, maybe the agents will interact with each other and just exchange information because in the end you need something for your agent to still have a textual response.

00:56:18: If you instruct an agent who can direct it into a database or ask another agent nobody cares.

00:56:24: as long as they get their final outcome there?

00:56:27: That is right.

00:56:28: but at the end we need find a good way of talking.

00:56:35: our clients provide them a way of getting into this authentic game.

00:56:47: We have in general, like the whole approach depending on where the customer starts.

00:56:53: so are there already some kind of use cases that were defined or they're completely clueless to start?

00:57:01: Or maybe you already had an apartment area of business where actually they would like to start with the AI use cases and we have this kind of a structured workshops, but really drilling into asking a lot of why's.

00:57:13: Really trying to get these business value out of it first off course focusing on the business problems because I think that is often the case for mail.

00:57:22: maybe there isn't any tool or some idea about if you don't know what problem your are trying solve And if you don't track the data, how much it is a big of problem.

00:57:35: If can be improved by AI You are already like starting without vision and there's no guarantee that will actually succeed.

00:57:43: on these workshops we have lot interesting discussions How people see this things differently.

00:57:49: Also challenging Is the data available?

00:57:53: Challenging also if technology ready.

00:57:55: sometimes maybe they're over creative, this is where you know.

00:57:59: maybe the technology is not ready there or because they still see it as a magic thing and magic box that can do everything what they like.

00:58:09: I think this also bit of hype has been produced around AI And This is our task to do at client to ground them basically guide them through but be realistic Like Not Everything Will Be Possible But a lot more will be possible now.

00:58:29: Exactly, but it's also funny because sometimes we end up in the thing that for this specific problem... We don't need AI at all!

00:58:38: It's like simple deterministic flow programmatically that can be done integration within systems and yeah That's it.