Episode 4: From the invention of SAP HANA to Clean Core in 42 minutes - Emma Qian and Prof. Alexander Zeier

Guests: 

• Prof. Alexander Zeier - "The inventor of SAP HANA" formerly reporting to Hasso Plattner, CTO Accenture and Professor at MIT, now Chief Scientist and co-founder Novasoftware.AI
• Emma Qian - Formerly Google DeepMind and Meta AI, now CEO Novasoftware.AI

Host: 

• Alexander Greb - Consulting Director - cbs corporate business solutions

In this episode, join Alexander Greb in an exceptional conversation with two outstanding tech visionaries: Professor Alexander Zeier, inventor of SAP HANA, and Emma Qian, co-founder of Nova AI and alumna of Google DeepMind and Meta AI.

In this episode, Alexander Zeier shares the gripping story behind SAP's most transformative innovation—SAP HANA—from its inception as Hasso Plattner’s visionary idea to becoming the backbone of modern SAP technology. Discover fascinating insights from the pioneering days, unexpected challenges, and groundbreaking decisions that shaped enterprise technology as we know it today.

Prof. Zeier has 30+ years of experience in SAP, is the co-inventor of SAP HANA, the „New Architecture“ powering S/4HANA, and served as Global CTO for SAP at Accenture over a decade to deliver many of the most complex S/4HANA Transformations, before he joined as a Co-Founder Nova AI. 

Emma Qian then introduces us to Nova AI's revolutionary Agentic AI, designed to accelerate SAP optimization dramatically. Experience firsthand the astonishing power of this innovation, capable of condensing tasks that typically require years of consulting into mere minutes.

Explore how AI is fundamentally reshaping the role of SAP consultants and learn about the tremendous opportunities—and responsibilities—this brings for professionals and businesses alike.

All this and more, exclusively here on Transformation Every Day.

Follow Alexander Zeier: https://www.linkedin.com/in/azeier/
Follow Emma Qian: https://www.linkedin.com/in/eqian99/
https://www.linkedin.com/company/nova-ai-software?trk=blended-typeahead

Follow Alexander: https://www.linkedin.com/in/alexandergreb/
Follow Johannes: https://www.linkedin.com/in/johanneslangguth/

Thanks to Steven Spears for lending his voice for our podcast intro.
Theme music by Luis Álvarez a.k.a. Fourth Dogma

Transcript

Alexander Greb (00:01.952)
Okay, how would you summarize your two days here until now in one word? Exhausting?

Yeah, was gonna say busy or exhausting. sounds good.

Alexander, what about you?

very busy days but also very positive days because there was a lot of client feedback interesting and also after our keynote yesterday, directly approached several people so it is an exciting day.

until you're getting traction.

Prof Alexander Zeier (00:36.322)
And also lot of interest, even more interesting that the people from SAP, but also from the client and from partners or different activities, very interesting. Even more than I expected today, a few days.

and 10.

Alexander Greb (00:52.034)
And remember what's your plans and goals for tomorrow for the last day.

for tomorrow. Oh, unfortunately, I'm flying back early morning tomorrow because I have to be back in San Francisco for a

Which means you will miss my keynote.

That's very, I'll watch it virtually. I'll watch it all night.

We will take care that it's probably recorded.

Emma Qian (01:11.02)
Yeah, I think so. Yeah, I'll watch it on the play. You know, I'll buy Wi-Fi just to watch it.

Now you're here and now we are recording. Yes. Thank you very much. have in the room Professor Dr. Alexander Sire and Emma Chinn. I pronounced it right saying Chinn. Okay, okay. I always have to be a little bit careful about this. We are still here in Newtown Square at the SAP Experience Center. Great location. And we're talking about a very, very interesting topic spanning almost two decades.

I'm on

Yes, it's true.

Alexander Greb (01:45.324)
If I'm correct. 18 years roughly. And it says we have today a little bit weird title from the birth of SAP HANA to a clean core in 42 minutes. we will come to the point why this is not as weird as it sounds. Let's start with an introduction round now. Alexander, have your very, very strong prominent figure within the SAP. Eco space within the SAP.

Yep, 18 years.

Alexander Greb (02:12.654)
The footsteps you generated are huge, but would you introduce yourself with a few sentences?

Sure. Yeah, my name is Alexander Zeier. I'm doing SAP 33 years at the end. when SAP started with 400 employees, 1992. Yeah, since this time, 400. Yeah.

400. Almost startup like.

And after a few years of consulting, I gained my PhD on the SAP APO, already sponsored by the SAP board at the end. I've written a few books about SAP competence, SAP excellence. Then I was responsible at SAP as a product manager for the APO, which was also my PhD topic at this time.

After this, done some strategic projects for some other board members of SAP. And in April 2006, I got a call that Hasu Plattner would like to see me.

Alexander Greb (03:22.062)
Which is a call for everybody involved in SAP, may lead to nervousness. Yes.

I have never seen him in person before, so it was an interesting meeting. We had a conversation and he asked his question, what I've done before, and then at the end he asked me if I would like to work for him directly and leave FCP and work under his personal guidance as his deputy.

and starting a project about the new architecture of SAP.

And the new architecture has a certain name we will not say now, we will say it in a minute later on when we start. Emma, how about your story?

Yeah.

Emma Qian (04:11.256)
Very, very different field of technology that I come from. I was, ever since college, I studied computer science with a concentration of machine learning at Caltech. since then, I've always been more in this machine learning AI field. Previously, I worked at places like Google DeepMind and Facebook AI Research on more NLP research projects in the past, and then worked on a separate company earlier that was more in kind of foundation model and platform space.

So my background has generally been a lot more on just data science, ML, AI side.

So when you would be asked what was the compelling moment when you decided to go into that space? Is there something or was it just for you always clear that I want to do that tech stuff?

Yeah, I actually was more, I was always pretty interested in math growing up. So I actually didn't get into computer science until I would say like college, I would say. So before that I was quite interested in math and going to college, I thought I would, you know, at least like double major in math and actually I took, know, at Caltech, everyone has to take at least one intro computer science course and we built a bunch of projects there and I thought that was just like super.

how me as one person could build, create this software that is super, I can make all sorts of changes to it to my liking and other people can use it and find it fun and it just became clear to me that this is, yeah, I really enjoyed building new software. So that's when I actually switched over to more of a computer science major and.

Emma Qian (05:55.842)
But I was still very interested in math. I think math and computer science, the natural intersection there is more around data science and machine learning. So that's what I ended up specializing in in college.

Awesome, awesome. And if we go back to you, Alexander, for you then started a journey or that moment, that call concerning hustle, we have to work on something. Yeah, this kind of new architecture turned out to be something which would. think it's not wrong to say it would influence the world economy in a certain way.

that billions and billions of worth of orders of business is happening on what came out of that. It also is more or less a connection that we have because I was then about to implement what you developed back then. But tell us what was the name of the secret technology which came out there.

It is more or less the idea was that we start a new project and work for him. And there was a request more or less to think about new architecture. Many companies can generate a first generation, which was R2, 20 years in the market. And there was R3, 20 years. And Hasse told me it is very, very hard for company to build a third generation.

especially in the company itself. There was already two projects done. He told me Heidelberg and Vienna. And for some reasons, internal coordination. So it was...

Alexander Greb (07:43.542)
It was a deliberate decision of Hasso saying like we are now doing the third iteration, we are now going into the third generation and he was fully aware that this is for the company a highly important, it was also let's say a very risky moment because they said like what we are doing now and what we're deciding now will definitely be the beginning of a new phase of the company.

Mm.

Prof Alexander Zeier (08:10.176)
Absolutely. I asked him, what is your concerns? Revenue is growing, more companies buying SAP. But he told me SAP is not ready for the future from the stack, from the architecture. It might will disappear in six or 10 years. The technology is not up to date anymore. And to develop a new architecture, you need to start now to be ready in five years plus.

And so he asked me to think out of the box to hire also the core team from outside of SCP. And he gave me some recommendations. So more or less you can hire a team where you can have more or less the same skills like the others, some of them, and then you get a result which is good, but it's not.

extremely or extraordinary from the performance or you can be so smart and hire people who are in every of his areas, the top of the top. And then you bring in in a small time frame, the skills and the capabilities to build something which is disruptive, which was the target. At the end, he wanted to build and stack with a new data platform. there was two key elements.

elements was it had to be multi-core because at this time there was a hit of the speed of a CPU and was everything multi-core and the second thing was the idea to have everything in

speed of thought, so in memory, is a fast excess, was always the idea below 700 milliseconds, this is a reaction of a person, and this was a two figures, multi-core and in memory, and...

Prof Alexander Zeier (10:07.342)
Then we analyzed 67 large SAP clients, many of the famous big ones. Apple was one of the partners. It's also now public. Colgate, Unilever, Daimler, and others. Many large companies. And I had a good contact also with them in the past. And I was many years with SAP. And it was an exciting story, more or to analyze their workload.

and we analyzed at the end that it's not so much different from the read and write events of the workloads of the companies, which was before expected that you have in a BW system much more.

read events only and very limited write events, but it was less than 10 % different at the end. And then we decided to develop a new data platform who can cover transactional and analytical workload.

the famous merge of OLAP and OLTP. Exactly. When we always had like insight and action or insight to action not happening, yeah, because you had at one, at first the static view on topics. And if you then wanted to change something that's happened completely somewhere else, this was for the beginning, definitely a goal of that.

Yeah, this was the idea, to build a new generation of systems who can outperform the market. This was the goal.

Alexander Greb (11:42.318)
When did it get the famous name with the four letters?

It was starting in April 2006 that we gave this as a working title.

So it was already a working title. Okay.

Yes, it was a working title. There was a HANA project in 2006, 2007, which was about a financial accelerator. It was a HANA 2 project, 2007, 2008. We had an ATP check. First was a financial application. Second one was a logistic, more or application.

the benchmark drivers.

Prof Alexander Zeier (12:25.27)
No, but I have done the APO before, so I know how complicated it was to build an available to promise with a global resource check and against all the different warehouses. And then there was HANA Project 3, HANA Project 4 for the different components. And then we realized...

that such a system can really work. But there was a lot of complaints and also many people thought it cannot fit in main memory, how you can handle this. Then we added to this lightweight compression, so dictionary compression, run lengths and coding and other components. There was also some parts were missing, but I built also a global

academic university network with some colleagues from SAP who helped here a lot. And then we had from Berkeley a famous professor, from MIT a famous professor, and from Stanford we worked together and then we were able to close the gaps of our end-to-end architecture what were needed. At the end it was 28 different features which were needed.

The compression was a key topic which was unable to handle how you get parallelization in such a system. when you have a business process, it's very hard to paralyze this because even when you have 100 billion data records, when it's not sorted, then you cannot really paralyze such an...

a table, but then we have sorted the main thing and then we are able to split it up to the different cores. yeah, these were all the different parts developed. And the interesting thing was that then we needed more or less also proof that it works and Hasso stated, OK, I have built by myself a finance system very early days, so take finance and rebuild this. That's what we have done.

Alexander Greb (14:41.09)
I remember very well back then two things, like what you mentioned first, the compression factor, which was very relevant, especially for customers then in the beginning who had...

tool.

had to think about, okay, what kind of hardware do I need? And the biggest fear of many was like, whoa, my whole data in memory, what will that cost? And the compression factor was then the relevant argument of toning down this kind of discussion and leveraging that fear that this would become something of a highly costly hardware battle to do that.

But it always, Hathor had always, we had the vision to take the next two, three steps ahead because everyone know memory become cheaper, CPU become cheaper. And to build a system for today, when it have to be built for the next 20 years, that's not the best idea. Exactly. can be smarter and build the system.

who will work in five years. is 26 started in 2011, launched to the market. And we got from Intel and the NDA the roadmap where the CPU was developed just in several people involved from Intel.

Prof Alexander Zeier (16:01.486)
who helped us here. And so we were able to develop a system over five years who really hit the market at this point with enough memory, with enough CPUs. And then it was more or less the starting point that SAP were able to build the third generation. Interestingly, it was the topic that

This was quite disruptive on the one thing. You know, it's a complete new database. On the other thing, there was a question. The client down on old ECC, will this not help SAP anymore? Will buy something the old solution? But then was the feedback from the client, oh, you realize that your old SAP system?

is not ready for the future and you will be now.

We know that you can do an upgrade of your old ECC system to an HANA-based ERP system now called S4. And so you will be, also in the future, our technology and enterprise ERP partners. Sales of SAP even increased, funny-wise, of the old ERP system because the client decided again to invest here. this worked very well.

Let me do a short sidestep to one thing you said earlier, which I think is very, interesting. Hasso gave you the advice or it wasn't really advice, it just was a statement. Like you could do it like your predecessors and hire people who are aligned and you get good results or you could hire the best and you get...

Alexander Greb (17:49.87)
quite a wild bunch, can probably call it like this. And you will get exceptional results. This is an advice. If you look at the publishing or on the social media, what you read, everybody always says like, if you do it like that, you will get a bunch of alpha men and women together, which will be an absolute pain to manage. try to...

keep it a little bit lower key and this will be better. He basically said the absolute opposite, like go for the best and it's your task to manage this wild bunch. What kind of experience in managing people that you have as a technologist at heart back then that's qualified you for this task to manage this extraordinary collection of high potentials.

Prof Alexander Zeier (18:51.298)
That's hard to say, but when you want to deliver, really the next generation of systems and also more or fulfill Hasso's requests and you have to do it. is no other choice. it was very hard to do. on the other side, it was also quite rewarding because you realize really that you can build something different. And it was also exciting to work on something like this.

You just do it.

Prof Alexander Zeier (19:20.43)
He gave his full support, but at the end he was there every few weeks. And then often he asked me, good progress, but can you run faster? I want to make more progress.

So there was pressure. He was highly involved.

Yeah, yeah, yeah. He was highly involved. He also organized that he got also the full support from SAP. So it was a core team of which I hired from outside to build just a high performance team. at the end, it was also very supported by the SAP board.

It was a core team, when I needed, for example, for some dedicated request, a few people more, or a few hundred people more, then was some call to SAP, and then it worked.

But I fully go along with what you said about his love for detail and for involvement, of course, too. Because I had the same experience that when I did the implementations back then, the first large enterprise customers, he called directly and asked for the status. Where are the projects? Sorry, where are the problems? Have you planned to get rid of them and so on? So what are the next steps? And this kind of extreme ownership

Alexander Greb (20:44.078)
I guess is something what you see very rare today on the one hand and also of course what separates him and like-minded people in his class of course from others who probably will not leave footsteps as big.

We had one brilliant skill. He more or less understands latest technology very fast and can transform this, how he can apply this to business. I've never seen a person like this. We usually have some technical guys from the background who are quite brilliant.

Then you have these MBA guys at the end more, which can talk and have this business knowledge. But when you go down to fulfill this with the latest technologies, this is...

very hard for them. Actually, I have an MBA degree and a technology, computer science degree, and now I'm computer science professor in addition to honor. So I always try to bridge this, and this helped me also to deliver what he asked for. And especially this was in this critical phase.

And he told me, you have to build the prototype for me and for SAP that it really works. And so there was this famous Colgate example, the European Productive System. It's now public, I think I can talk about. That we have 250 million invoices in the system. And then we...

Prof Alexander Zeier (22:28.558)
reprogrammed the system so that it's more or less the Dunning run for this overture and everything was on a classic database in 20 minutes and we parallelized this first only with the technology that it was one minute and we saw 20 minutes to one minute 20 times. That's win, yeah. We were quite satisfied. We showed it to Hathor and then...

Nobody cares about this 20 times. Please go back and come back when you are 1000 times faster or 800 times faster. Okay. That's a good... I think he told me, if not, then I go sailing only. And so we restarted and looked on the complete process.

statement.

Prof Alexander Zeier (23:22.742)
And then we realized it's a hard technology, what's is paralyzing something and we're quite good from the aggregation of the data, but...

It doesn't keep the intermediate results from the Dunning one. And then we realized, okay, when we do this in addition to some other tuning, then we can get from 20 minutes to 1.6 seconds. It was 800 times faster on a 24 core machine, better machine from Intel, which they gave us in front.

And so this was more or less a breakthrough that we applied business logic, know-how on the technology. And it made the difference. This made the difference.

But maybe that's exactly that area where the highly relevant people excel who innovate. Because innovation is not about just an invention, but innovation is application of an invention.

Applying.

Alexander Greb (24:27.54)
Exactly. That's where an idea becomes actually worth something. And I think we will talk, Emma, with you in short exactly about that thing, because that's what you heavily put an emphasis on. But when did it become apparent to you that what you are developing there in the Berlin area in that, I think it was a villa or a castle?

It was called the Reichsbahn Villa, old headquarter of the Eastern Germany railway.

Yeah, but it was bought by Hasso

sported and renovated and this was more or less.

When did you become aware that this would be something which even just by the name would be attached to basically anything that SAP is doing and would be globally such a big thing? Was it clear from the beginning?

Prof Alexander Zeier (25:27.404)
The goal was from the beginning. This was also the high pressure from him to more or less to guarantee the survival of ACP. Sometimes he wanted to this. So it was a high goal. the end it worked very well and it's quite nice to see when now more or less...

All larger companies in the world are running on the system you have built and created and designed and architect. I have eight core patents with HUSO on this. So it's also granted the patents. So the core patent which.

was probably a good idea.

Yes, this was an interesting piece. was at Accenture, not Accenture, I was late at Accenture, but at this time before I was with SAP, was a patent manager for all manufacturing industries. And so I already helped to do some important development to secure the IP for the company. And then I realized, with HANA we have...

We probably should think about it.

Prof Alexander Zeier (26:37.102)
We some interesting pieces and when Hussow won a Conquer the Market with it, then it makes sense to get lawyers, appeal lawyers, patent lawyers early involved. So I had set up my own programs, had my own lawyers, everything when it was developed more or less in the group before it was published or anything done.

It may have some relevance.

Prof Alexander Zeier (27:06.254)
I had a lawyer who built a provisional pattern over the weekend sometimes or something. And this is the reason that HANA is so highly patented. Many competitors try to build something similar or rebuild this or have done... But the core components, this core pattern about OLTP and OLAP on one was...

built in September 27, I think, I remember correctly. We filed the patent. And usually a patent that it accepted and not the competitors try to make it invalid. is what they try to do. It takes, you get the patent filed in two, three years.

But the HANA patent needed eight or nine years because there were so many tries to make it invalid. But at the end it was so unique and disruptive that it was granted.

Because an element is what probably nobody really talks about anymore and it's a part of SAP nostalgia probably. But we really have to think about the circumstances that were during that time. Because on one hand, I think the high time of your work back then was during the financial crisis, where everybody was under heavy strain. And I remember me being in the car and thinking about it like...

What's going to happen? How will we continue in the next years? It was also a time when, and I remember this quite vividly, I went to SAP in 2009. And this was a time when SAP was quite different than today. it has a face, it was probably not the best time for SAP. Because there were issues with the DSHE and the users, the focus strategically was not really on technology, but...

Alexander Greb (29:05.144)
growing by some maintenance topics and support topics which did not really resonate well with the customers. I remember people telling me like, hey, why do you go to SAP? And words were...

in connection with that, does SAP have a future? Because everybody was asking, what was the next big thing at that time? Because that's what the analysts and so on were expecting. We had R3, which was a huge step, and so on, and ECC, and so on. But that next thing, that one that you were referring to, was happening. And it was for us, who were not within this kind of inner circle. It was nowhere to be seen. So that everybody said, like, come on.

Why do you go there? Aren't there more attractive places, more future? And while this was happening, you were sitting in your villa there and thinking of the next steps. But in my opinion, it's now looking back, an absolutely fantastic story. To be honest.

I'm also proud about this. That's what you have developed.

You can. And we're not dropping names now, but that wild bunch you were managing in Potsdam back then, I think there were lots of now prominent names under that who made their way in academics, in technology management and so on. You had really like a dream team back then.

Prof Alexander Zeier (30:30.796)
Yeah, I was able to select really the top people from the different groups. was, there was also some team spirit at the end to bring them together. There was also from different culture, different groups, a core component for that really prototype finance system is working. There was one person who came

with another background but I was always convinced because he was a strong developer with a gaming hacker at the end who have developed C++ before which was not the main topic in Forhada.

would not expect a game hacker at a database development team at SAP.

Yeah, exactly, but he was...

really good in development and I thought it would be good to have him in the core team. He had also some other skills but at the end when we needed to program this additional intermediate business logic, when we realized that Hanna's system is not keeping these intermediate steps, results.

Prof Alexander Zeier (31:53.186)
to process further. He was able to program this in a few days. Probably nowadays it would take another environment much longer for sure. And then we were able to show that it really works.

You were not calling it a dead end, but at a very difficult moment. You had that problem, you had to overcome. And the solution to that, to the problem that may be decisive for the fate and the future was done by a game hacker. He solved that. To be honest, if I would view a movie...

Yep.

Alexander Greb (32:35.374)
Yeah, where this would happen and so on. I would say like, come on. This is thick Hollywood stuff. Too far for reality. But it is like that. That's fantastic. When you look back now, what are the big misconceptions about Hannah that annoy you most?

This was the case.

Prof Alexander Zeier (32:59.054)
At the end, originally when it was brought to the market, the people thought it's just an analytic accelerator, which was never the target to do. It was always a replacement of the database, the data platform under the SAP Core system. So it was a replacement of the transactional database.

as a side effect, also takes the competitor out of this, is an obvious and valid goal for a company to have a larger stack, what also Google or Apple has done, more or less they have also tried to control the database and the other layers.

And then we were able to build a platform who can now enable multi-core. It was in addition in memory another critical piece which we also needed to overcome. have not mentioned before a column store which is a HANA database.

Which was the big difference. Other database logics.

This is more or less a column store for transactional system was never created and invented before. It was not able to do it.

Alexander Greb (34:23.63)
Can you maybe explain in a few sentences why this is so different? Because this was always the topic like, okay, columns, but why is this relevant? What is the big advantage and the big difference when I have a system working like that?

More or in a column store, data field is a column. And the column is like a fully indexed system. And this was the reason that you can do analytics on every data field. And in addition, just to add here,

To make it really in multi-core ready, the column store, was something from SAP before it developed, we have not everything new developed. We took pieces which were partly available, partly was also bought in addition to fulfill this. HANA is an

column store, who can be parallelized on multi-core, and there's also a normal store included. In addition, but the column store, who can be really parallelized and compressed, is the main thing.

And what I think we have to underwrite now is the relevance that, for example, in old ECC systems, if you look into them and if you look at the tables, you have different kinds of tables, you have aggregates, have these index tables which are relevant because the system itself is not fast enough to work.

Prof Alexander Zeier (35:53.624)
Yeah.

Alexander Greb (36:01.55)
with the line items. So you have these helpers with aggregates, with these index tables. Correct me if I'm wrong. Yeah. Because you're the expert. But the difference is that now that you have everything together, you have to get rid of the problems that you get with aggregates, where, be honest, the result, and maybe I'm wrong, but I ask once data scientists about the aggregate.

tables and say that to be honest when when I'm getting a KPI I'm looking not at the line items in this system I look at the aggregate tables which means that per definition this is not actual because it happened before they were prepared for me to extract the KPI but can you give me the guarantee that some elements so the a plus b

lot of patch ones exactly

Alexander Greb (36:52.18)
of the aggregate have actually the identical timestamp and he said no, it's not possible.

possible because an aggregate cannot be the latest.

Exactly. So it means that the result I get out of that aggregate table is per definition fuzzy. It's not precise, it's fuzzy. And you have of course then an aggregate of an aggregate and so on. So what you get out in your KPI and so on is always not 100 % up to the point which makes it not only difficult to steer, but it has of course huge consequences in

when you optimize KPIs, when you want to plan numbers. We are talking about millions here, what you can really save on a business case in that aspect. But when I have now a fully indexed column, I have this already done and I'm always able to work and I now can work on the mud-dog and not in different thousand tables. And without that, those things we are taking off for granted in cloud or in...

Yeah.

Alexander Greb (37:55.96)
We have the business data cloud now released a few weeks ago. This would be unthinkable because we could not think about having data together and in the next step, of course, providing this comprehensive data set to AI so that the answers that I get from AI actually make sense.

the latest correct data, not a few minutes or even longer, older data. is, you look on the process, I worked for some large insurance companies. had an ECC finance, usually general ledger and some other parts. Then moved to a BW system and the client took more than 30 hours to get this data. the data is 30 hours.

already old data. But also in addition in the core ECC system, per definition it's a layer.

the upper layer cannot be really parallelized because it doesn't have the structure. And so that really to speed the system up as needed and fulfill a larger amount of data, was this column store, the central piece. But there was some issue which we also then solved at the end. The column store is pretty weak on updates.

because the updates have a high workload compared to on a classical database to get up to 500,000 inserts for updates are possible in a column store it's 5 to 10,000 usually. This is not enough for it to fulfill for a large ERP system. So we more or less invented also insert only that you put

Prof Alexander Zeier (39:44.96)
set the old data record invalid and with a timestamp and have written the new data record in addition. this is now active so you can see the trends in the system. So the sales guys, but also auditors, love this because then they can go back in time. so a lot of advantages, part of the core architecture of which we invented and have implemented this.

So I would even say that the clients can apply AI as needed in the enterprise.

on current latest data and also on a larger data set which they need to feed AI to get the results. The HANA system is the central important basis and this new architecture to fulfill this.

Because then you really can apply once the full data set can access this in an old ECC system. There are so many cluster tables and others that it's really for many companies hard to get the data out as needed and build something on top of this. So this is a very important additional side effect or even important main effect when people move to the S4HANA platform.

And Emma, you are listening, laughing and thinking you guys had problems back then.

Emma Qian (41:22.766)
It's very interesting always to hear the past stories of how this comes about.

We jumped now a few years because you had a longer tenure as CTO of Accenture for the SAP business. And we're coming now to the present. You are highly visible again. And you are now also in the world of AI. How do you two meet, actually? Because basically, I think this is probably not the, let's say, common pairing. Yeah.

vice-periods.

Alexander Greb (42:02.624)
that you usually meet? Yeah. How did that happen?

Yeah, so a back story. We started this work on this AI company. And actually, originally, it was in AI for SAP, which happy to chat about how we got into that. But after we pivoted to SAP, we brought on some investors, including Paul Doherty, who's the former CTO of Accenture. So he's an angel investor and advisor for us. And I think, Alexander, at that time, you had been

like doing a lot of learning a lot about AI and Paul had known that you were very interesting AI so Paul was like you guys should totally meet and you know exchange ideas and that's how we initially met actually.

You're the company that you have to get in, which is called Nova Software. Did that exist already back then or was it something like a former version of it?

Yeah, it was very early stages back then. the company started working on generally using AI to solve software problems, so like not specialized to SAP. And then we had decided to pivot to applying this specifically for SAP for a variety of reasons. And sort of in the early stages of that, we had some early supporters like Paul Doherty. We had some initial results, but very, very early the team was just me and one other person at that time.

Emma Qian (43:30.28)
I think that was around the time that we met Alexander.

And you had a very deep AI background already at that time because you ex-Google DeepMind and were very deep in that world already. How then the move to SAP? Sounds a little bit counterintuitive.

Yeah. Yeah.

Emma Qian (43:45.854)
Yeah, so I Yeah for sure. Yeah, I still get this question all the time because I still live in San Francisco So all my friends are you know working on these fancy AI companies or you know building the models or building like these like, know consumer UI like sort of AI based experiences on top and then I'm like working on SAP and actually like it would be for last year I've never even seen an SAP system and I honestly did not really know what people use it for I just knew it was a big company and

I a lot, I still have to explain all the time to people. Like usually, when I tell people what I work on, I have to explain to them what SAP even is. So I get this question all the time, like why did I end up moving to SAP? So yeah, we started, so I started this company working on sort of building software, like AI agents for writing software. And that was a field that I was interested in, and sort of relevance for some stuff I had done in the past as well. But like I was...

very interesting in that because I do think that software is one of the, like probably the area that AI will have the biggest impact on. And I think lots of problems in this world could be sort of alleviated if we had better software where people were able to build more software. And I think AI is like how we're gonna get there.

Did you look back then just at SAP software from the perspective of a portfolio? Or did you also look into actual customer implementations? know, like, OK, how is this stuff actually used? Because the misconception is always like a running shoe. I always take the running shoe example. You may think that everybody who

buys a running shoe is now a perfect athlete. But it's not like this. And you have like maybe 2 % of the running shoe buyers who are actually going into athletics. Then you have like 40 % who are doing something for their health and so on. And then you still have 60 % who are just rudiment users or using it as a fashion item. And the same, from my experience, is happening with enterprise software.

Emma Qian (45:41.966)
Yeah.

Emma Qian (45:52.694)
Yeah.

Those who have really get the best out of it, but you have that huge trap value gap with those who bought it, implemented it, and then using just maybe like the same like we do in Excel, like doing just 5 % of what this actually is able to do. Because I think that the perspective then would be extremely interesting. Yeah. wow, how are you working with that? Yeah. You're not...

you

Alexander Greb (46:21.656)
doing the potential at all.

Yeah, 100%. Yeah, that's one of, I think when I first learned about SAP, that was definitely one of the...

areas I was just really surprised by is how differently people use it. So how I kind of got into that is originally we built this general coding engine and we got really good results on that, like top of benchmark, like June of last year, but then looked at, we realized basically that in order to solve real world engineering problems, these are generally much larger and more complex than what a general coding engine can handle. So we were...

looking to specialize in a particular area so we can solve some of these high value real world engineering problems. And I did like a wide survey of the different types of engineering that's out there. And the thing that really caught my attention about things like SAP and Salesforce and all of those is I was surprised by like the fact that people not only, you know, people spend money on the software obviously, but then they have to spend generally a lot more money on consultants to sort of help them customize the software so that it works for.

them in a particular way.

Alexander Greb (47:26.555)
It's quite a blow for me as a consultant.

I was under here also, ex-consulted. But it's actually very, and I understand why it's necessary, I just wasn't used to this sort of dynamic before where the software you see as a startup, it's like a SaaS solution, you pay a little bit of money and you use it off the shelf, right?

Maybe to interrupt you, but I do not disagree with you because I think the thing which is so surprising also to me is maybe not that there are consultants necessarily, but what kind of activities are happening there? Are the consultants there to do some...

get that thing running stuff. Is that their main activity, which it is in most of the cases, unfortunately? Or is the job they are doing, they're actually there to deliver value, to make actually the customer better in what they're doing, to give him capabilities that he was not able to do before? And I think that's actually the real big problem, what those consultants are doing there.

Yeah, for sure. Yeah. Yeah, we were really like, had no idea what the consultants were doing with the customers generally, because it seems like I had just to maybe just need them to implement it and that's it. Right. But it seems like you always need them for maintenance and updating and like a lot of stuff. Right. So and I and after looking to SAP, I definitely understand why it's necessary because it's such a large software with so many ways to configure it. even fair point. Yeah. All the ways to configure is not enough. So now you have to write custom code. Right.

Emma Qian (49:03.376)
So I totally see why that is necessary, but that's also an area where we're like, okay, this is pretty interesting. Here's one area, especially with S4HANA, migration implementation, like transformation, where there's these very large projects that are taking multiple years, like army of consultants to do, and at the end, sometimes you don't get the results you really want. You don't actually build it in a way that makes the system run better and make it more upgradable in cloud.

Cloud ready, et cetera. So then we're like, OK, this seems like a pretty interesting problem to maybe see, is there something we can do to enable people to be able to do this faster and cheaper and better?

And maybe to add one perspective to it, because what I see as a problem is, is that what you say? Yes, that is necessary to do that. But the effect of it is not only that we have expensive projects, which could be probably more efficient and effective. The problem is also, in my opinion, the waste of human capital. Like, for example, when I need, let's say, in a huge project and

I see, for example, that a system integrator was using 100 people to solve a problem and to implement something. And at the same time, we see that the available capacity of consultants is one major restriction I have in the SAP user community moving forward to the cloud and so on, because the necessary resources are not available because they are caught.

I want to be more efficient in bringing customers on the new platforms. And SAP has its clear goals. We're in 2027 and 2030 and so on. And to be honest, if we take the speed we are having now, we are nowhere there. And if you extrapolate it, absolutely. And the problem is we have to use less consultants to achieve the same.

Emma Qian (50:55.128)
Yeah.

Alexander Greb (51:06.624)
in my opinion. Yeah, we have to be more efficient.

There's always a goal to have less consultants. I was at Accenture responsible for the 85,000 people who do the SAP work, the SAP business group of Accenture.

You have to deliver to the clients many cases in a fixed price project. So there is from the clients, send an IFP and request for proposal out with a concrete example, with a price tag. And it is just so complicated for many clients to do.

the upgrade of 2S4. I had many, many meetings over the last few years when CIO or CTO or sometimes CFOs, I had conversations and they told me, I like your HANA system, it's great, it's another technology but...

It is so complicated and expensive to upgrade to S4. SAP deliver now the standard code who works on S4. You can discuss how many value it brings to the clients. Many CFO told me it's or less the same system like in the past.

Prof Alexander Zeier (52:38.178)
But I have so many custom code in the system and it is very, very complex for the clients to upgrade this with manual activities with work. My HANA, my S4, my line of business for HANA and related services was going quite well because the clients moved to S4 HANA and so we had...

people on the projects on site, had offshore capabilities, but they needed often hundreds of people to do the custom code upgrade. And even when you do it from India or Manila, Philippines, it's still a lot of costs. And there was always the idea there should be.

some more structured, automated way to do an upgrade in the future. On the other side, was in the past not really possible. I have developed several tools by myself with the team at Accenture. looked on, I was also responsible for the ecosystems. So there was some partners who have developed something, but there was mainly some, I would say, standards, manual steps over

changed a little bit, there was no self-learning, self-improving system, which I can think this was all this exciting stuff. And Paul Doherty sent me an email or introduced me. have here a team who do...

claims they can do coding and want to do this for SAP. So they said, I had looked the last two, three years already very deeply in AI and have built even some AI systems by myself. And so I had already some good understanding what is possible, but I didn't find anyone who was able to get on this level of quality, have nothing seen in Europe or in

Prof Alexander Zeier (54:46.394)
also in the US, was also introduced to some other companies who are able to really do this deep change of a system and apply this. And this was an interesting case here that you can bring it to this absolute high end level.

But but too too.

another perspective what you said about the application or where we are and where the problem is that we are about to solve is that the move to S4HANA now happens since 10 years. It's not something which is really new, which is happening since 10 years and we've collected a lot of experience, be it inside SAP or being that consultants have collected experiences that and if you look what has happened within the last 10 years and we have seen of course,

a certain majority of companies choose to do a migration. So a typical what's called brownfield. They took what they had and they moved it over to as a technical conversion to the new platform with the backside that is just the first step that we should do. Because nothing is won by this. It's just, let's say, technical move. This is a technical exercise where the business

value is really debatable.

Prof Alexander Zeier (56:08.258)
The technical side, it's not even HANA used what it can.

You have a status in the end which is not even HANA optimized because you still have the old...

settings, the old, even custom codes, yeah, old tables you're working with, with the old building blocks, are for single core and for multi core and as such hindering your speed. So basically everything what you want to achieve with that is not really possible. But still so many people have done that, of course, and move that way without doing the necessary second and third step, which is the optimization phase. And they have not done it because

of what you told. It's costly, it's a little bit grayish what actually happens and it's just cleaning work and so on. And I experienced myself a lot this in the last 10 years. said, okay, we have now Go Live, we have made that technical migration, fine, let's go back to our daily business. Now we are on HANA, on S4. So let's continue. And nothing has changed. They worked like before and they are now on a setup which basically is blocking their future innovation ability.

Yeah.

Prof Alexander Zeier (57:17.134)
They have a system on HANA but using many of the old technology stack. When there is an old sub-gui still used, then you are not really on a HANA capability, what is able you can get out of the system. You are using a kind of intermediate step as you mentioned.

even make it probably more complicated for the future when they're not moving the next step to clean it and redevelop these major applications.

Mm-hmm.

And that's exactly where we return now to know our software. Yeah. Because we are talking about innovation is actually application. Yeah. And we're not talking about, I made a new JETGTP model and so on, but you made actually an application. Right. Which is where so many AI initiatives do not deliver. Yeah. And the problem we are talking about is exactly that. Can you...

Go deeper into detail Emma.

Emma Qian (58:23.052)
Yeah, sure. Like we are.

So at a higher level, we are building essentially this AI agent system to help people get from their legacy custom ABAP code to the more modern clean core compatible code where now you're much more upgradable, cloud ready. can take advantage of new things from SAP. You can build additional innovations on top. And this is really how the system is meant to be used. And this is one of the reasons why I

I found this area to be so interesting and why I decided to go into SAP because there's many different types of engineering in the world and I had no ties to SAP prior to deciding to pivot into this. But I saw that there was a huge opportunity here where I think clients don't want to spend a ton of money and just do this brownfield migration where you're theoretically compliant, you're on S4HANA but you waste all that money and all that time and the system doesn't work better than before.

basically mid-progress because you do not want to make that two years

Yeah, exactly. yeah. So I think the issue is like the second part was just like actually optimizing the system, become clean core, et cetera. So like when it's done manually, is how it's traditionally been done, it just takes so long and it's so expensive. So clients don't want to make that investment, right? So I think, you know, if there's a way that we could really significantly reduce the amount of effort and time and money and, you know, ensure higher quality, then

Emma Qian (59:54.466)
this would be something that could be, yeah, really help the clients get to a stage where they're not stuck with this legacy system.

But why do I need for that an agent? Because if I go to OpenAI, OpenAI can write me code also. Why do I need an agent for that? What does it make better?

Yeah, that's a great question. I think there's, know, obviously Anthropic, et cetera, has like some of the top AI models out there. And you could put something into this model and it generates you code and, probably looks roughly correct. Maybe it's like even 80 % the way correct and the last 20 % you have to debug, right? But a lot of times like debugging the last 20 % is actually the majority of the effort.

One kind of analogy that I think we like to use on what is the difference between just going to OpenAI and using the top model versus building this of really system of agents, which still take advantage of models underneath, but there's a lot of additional tooling orchestration on top. Imagine that I'm trying to build some software program. I have one person, and I give him just like,

pen and paper and just ask him, just write down one letter at a time. Like you cannot stop to think, you cannot go back and erase anything. Like I'm gonna tell you what to build and you just like write out the code on this paper and that's it.

Alexander Greb (01:01:12.886)
Which is also, by the way, the reality of many concepts.

And maybe this is why they need some AI tools. But yeah, so that's kind of, that's like one, let's say that's scenario A, right? And imagine the types of programs you can build with that. It's probably gonna be quite buggy. Like even a good programmer, right, you are not gonna remember exactly the syntax for everything. You might have a little bit of error here and there. The program might look pretty, like at first glance, it looks reasonable, but there's probably gonna be

and it would just not work and there's certainly no way you can build any complex software system with that. And the scenario B, is, look at, I guess, what we're doing today with any of the software companies. How do we build this rocket that can go to space or build SAP, which is powering the largest companies in the world, right? Clearly it's not using pen and paper, one person writing down one thing at a time. And how do we get to that? It's not like the people are 10 times smarter or 100 times smarter, right? It's like the same people.

except for we're giving these people much better tools. We're giving them this terminal to be able to, they can go execute code, they have debugger to be able to speak into the program to see what works and what doesn't. And when things doesn't work, which oftentimes it will not, even the best programmer cannot guarantee that everything they write will be 100 % correct the first time around. They're very good at debugging. So you take these tools to help you understand how the program works and where it doesn't work and how do you then go back and fix it. And then also,

you coordinate between teams of people. You don't have just one person build SAP. It's like a coordination program. There are people in charge of guiding the high level flow. And then they might assign sub-tasks to different people. And then collaborating together with more advanced tools, now you can build really robust, functional, much larger software. So I would say that's kind of a good analogy for what's the difference between

Emma Qian (01:03:09.134)
using just like TragiBitZ as a model. That's basically scenario A where you're asking like the model, you know, TragiBitZ is like generating one token at time, which is kind of, you can think of it as words. One token at a time. There's no ability to execute code. There's no ability to debug, to fix anything versus like, you know, a system like an AI agent system is really more like the second scenario we're talking about where we can have many different agents running around, which is kind of like people running around there assigned to do...

different tasks, we build out, you can build out like specific tools for it to help understand what is going on with the program, go execute and test and debug. And this is like the difference between one person writing on paper with a pen and what we can do today is the kind of difference that we can make from opening our model to really a agentic system tailored for a specific task. And I think it's not like, again, like the model doesn't need to be smarter, but there's clearly probably 100X, 1000X difference in the

quality of output there.

And we're coming out to the point which I experienced yesterday and I thought that... or it's probably not so easy to make me really...

stop and sit there in awe and think about it. What the hell did I see right now? But what happened yesterday was that first day like this, I get to be honest, daily two to three mails or messages on LinkedIn and so on about, hey, I figured out a new way, I stuff, agent model, whatever, and so on. I think about doing it. Ideas there, no shortage of that, definitely. But what you were showing me

Alexander Greb (01:04:50.624)
was the actual cleaning work, like what projects are in, for example, the optimization phase of a brown field when you move from ECC to S4HANA. And then you get the custom code done, you clean it up, you make a clean core, you get the APIs right and to the new ones and so on so on so on.

You got hand on an actual customer system. You got hand on the data and you performed this with your agent. something which takes maybe like one or two years and so on in an actual project, which is done by quite some people, which is costing a lot of money. And to such an extent that it's not done by many customers with the consequence we talked about.

I had the possibility yesterday to watch what happened, because you filmed it, in real time. And what took, or what takes, as it is, if humans do it, one to two years, was done in 42 minutes, which was shocking.

There was no stage, there was no fake. I could watch it, I could see what happens, what is done there and so on. We looked into the data, we looked into the structures and so on. And the cleaning work that nobody really wants to tackle and so on and what's hindering is done in 42 minutes.

We have today also, on the second day, we showed even the running system, not only the video recording. I am as an architect also very skeptical how does it work and so if it really works.

Alexander Greb (01:06:37.742)
It could be staged. So many stuff is staged in that area.

Over the last 10 years I was essentially responsible for the ecosystem of SAP plus tools. So I also have seen so many, I got too many calls and somebody contacted me. I have the next silver bullet and so I have proven everything and many of the silver bullets.

Go in Silicon Valley, go into a bar at night and at every table you meet people who are just that short thing before you open it.

Exactly.

specification makes a front and back end installation, front end installation and then you have at the end the theory with the results.

Alexander Greb (01:07:43.49)
Yeah.

Alexander Greb (01:07:52.664)
Because exactly, you even implement the Fiori apps and so on. And I think this is something where you're now at a point where you really have to be clear about that this is a singularity event within the.

Yeah.

Alexander Greb (01:08:08.684)
that we are now able to do such a thing within the SAP ecosystem. And the fun thing about that is I made, I also discussed this with Andreas Welch, I made four weeks ago a posting on LinkedIn, which was a bit provocative, I knew, because I wanted to generate a little bit discussion, but it was all about how AI will change the life of us white collar consultants. Because in that way, which kind of roles may have issues and will probably die.

which kind of roles may thrive because of AI and what kind of consequences they are. And the reaction was very interesting because the people who were of course addressed with this statement, so consultants which are technical, quite a few people, reacted in a way which I did not expect it to be that way because most, not most, but a large part actually said like, no, Manual testing,

Yeah.

Alexander Greb (01:09:07.266)
That will be needed in years also for sure and so on. So you saw this kind of resistance thinking about what am I doing? Is that future proof? Isn't there maybe something digital which can be faster and cheaper than me? Because if that's faster and cheaper than me, I will definitely be replaced. There's no way around that. What's your take on that?

Oh yeah, I also think one, you know, I think my general thought is

you will probably not be replaced by an AI completely, but you'll probably be replaced by someone who understands how to use the AI. So I think, you know, I don't think there will be a world where there's like no human in the loop necessary. But it's about how do you understanding how to use these AI tools and sort of understanding maybe like, yeah, like the business process and what is necessary and maybe, yeah, like, know, sort of, you almost became like a manager of a bunch of AI agents in one sense.

So, but I think that's sort of maybe the world that we'll begin heading towards is like, like if people, if you can learn how to manage these AI agents very well, which is definitely a skill set, like how do these models behave and how do you actually get them to do the thing you want to do and what are the limitations? Like, I think those are important to understand, but if you could learn a skill set of like, how do I utilize these AI agents and models and how do I manage them all? Then I think you will be like very, very well positioned in this sort of new world that we're going towards.

I even would say, and I think about you, I think we have many people here, enterprise architects at this event or in other cases. So when you are able to apply AI to the SAP ecosystem, I would even expect that your salary will increase because your productivity will increase. So today you probably need 10 hours to do the job. Now with our Nova AI tools, we can do it less.

Prof Alexander Zeier (01:11:07.224)
probably today less than five hours. And in the future, we think we can do it that you only need one hour or something like this. So your productivity will increase of the result by today already 50%, in the future 80 plus percent, and then you will be better paid. But you need also to...

understand this and it's really a different, I call it a different piece. So it is not only some additional skills to think about, it's really think always about a self learning and self improving system. So when you're in the SAP world,

There was 50 years of great success with SAP systems. At the end, you have always taken manual business process, implemented this in software, and then it was done more efficient and you have scaled this. But there was not really a self-learning system, what you had in the past.

You had to improve the query or what you gave by yourself.

Exactly, the person has done the changes and now you have an agent to learn. This is really a different thing. This is also what I learned from Emma and from the team, which was a different beast.

Emma Qian (01:12:35.854)
you

Yeah, and I think this is typically human reaction because yes, always history repeats itself and to have factories automated, that was machines and lines and robots and so on, of course changed the life of many people working there.

.

Alexander Greb (01:12:56.674)
When you look at the big picture, it's not that we have a higher unemployment rate because of what happened with industrial automatization because of that. But we had new possibilities, had new chances and so on, combined with a higher rate of wealth within the whole society. And I think this is happened back then in an industrial way is now happening in the digital way.

Yeah. Yeah.

Alexander Greb (01:13:26.4)
Is it something we should be, in your opinion, am I be afraid of? Is it a friend or is it a foe? What's happening?

Yeah, think what you said is exactly right. Like if you look, I don't know, back to our ancestors whose like their work was, you know, sort of hunting and like gathering food, etc. Right? Like if you fast forward them to today. You're going back far. Yeah, that's a very long time ago. Or like farming.

even farther away than us with our Hanar.

You

Yeah, this is very far away. But even more recent with farming and stuff, if you like teleport them to today and then you see the work that we are doing where we're like sitting on the chair, we just like move our fingers a little bit on this little computer and then they'll be like, that's not work. What do you mean that's work? They could have never imagined that. So I think there may be like a similar shift in the future. There will just be jobs that we cannot even imagine today. then maybe 50 years, 100 years from now, if we can look that far in the future, people will be doing

Emma Qian (01:14:24.336)
doing things that we cannot even imagine those are actual jobs that people have. What do you mean, have to go out and hunt a deer or something? So I think there's definitely going to be a shift coming. that's where I think really trying to understand AI and what it's capable of. that will be important for people who looking to, who are in this, we are in the very rapid pace of like,

change of pace right now. So I think that will be important to understand. And in terms of whether it's friend or foe, I think, you know, I would say most likely it's going to be very, very positive outcome for society. Like imagine having like there's so many, like just one.

Example right so many rare diseases in the world where just the the number of people who have it does not justify the investment That it takes to really solve those problems imagine having like way higher sort of total amount intelligence cheap intelligence They could allocate towards solving all sorts of problems in the world including that including many others, right? So I think overall it's most likely gonna be a very positive outcome for society I would say especially in Silicon Valley There is a little bit of a hesitation as well because it's such a new thing, right?

So I think overall we should be sort of still like cautious and try to make sure we deploy these models responsibly and safely because it is like wrangling like I was gonna say the whole different piece, right? It's like gonna be capable of things that we cannot even imagine in the near future. So I think just making sure that it does not get into, yeah, like we sort of.

handle this in a way where people cannot use it for malicious intent because it can amplify a good person 1000x but also can amplify a bad actor 1000x right so that's sort of the downside of that we have to be cautious of a little bit as well

Prof Alexander Zeier (01:16:13.826)
But from this perspective, just to add here, think about when clients know from what we talked about and what the event here is, the clients can move in 10 % of the effort to S4HANA, this is an ancient multi-agent system, and can reduce the effort, the cost for this. They have then really a system who helps the company more, can adjust this.

really use HANA as a but we also use a side effect of the tool.

So at the end, everyone will win because there's not even enough people in the market with the right skills. you have, I always compared when somebody today asks, what are these old sub agents are doing here in the system? Well, you have a team of building your application in the past, I needed 10 or 12 people and it took four months or longer, often a year when it's more complex one and everyone do some job.

And at the end it was very cost intensive, so many clients have decided not to do it so far. And now S4HANA is redeveloped by SAP itself, more or after 12 years, so officially since end of 2024.

latest was now completely developed. is also no time to wait anymore. Like in the past, you mentioned it was 2018. But many clients told me, yeah, yeah, yeah, but it's missing the function ABC, direct store delivery for dedicated area. So I wait for this. So okay, I waited. But now it's no, it's no excuse anymore. Exactly.

Alexander Greb (01:18:12.846)
And you remember we had already shift considering the end of maintenance twice. Yeah, it was a completely different date Yeah, which was probably too optimistic back then. Yeah, but but it shifted backwards several times We talked about everything until now, but what is on your personal roadmap? What's what's coming up?

Yeah.

Emma Qian (01:18:34.766)
So yeah, so we have already this.

sort of AI agent system that can take your legacy code and document it thoroughly and simplify it by fitting standard and rebuilding it. And we are already at the point where we think we can save a significant amount, probably 50 % or more of the effort compared to traditional completely manual approaches. we have, there's so many things on our roadmap that we know can improve the agent system. it just, have only so many engineers, so we to prioritize a little bit, right?

many many times better. So I think we want to get to point where like 95 % plus automated in this interest, this effort to become clean core and to reduce your tech debt. So I think that's, that will be very exciting. There's a lot of like very low hanging fruits, immediate items that we're working on that I think will improve the system by quite a bit. So that's sort of the near term thing that we're excited about. And I think beyond that, right, like Alexander said, like I think by reducing amount of effort it takes to do things like this, like the transformation to reduce your

old code and technical debts, now it frees people up to actually build things that could actually make their system run better, like future innovations, how do you automate and optimize your system? And there's so much data, so much insights that's within SAP, and I think there's a lot of interesting things that could be done there on how you could.

better utilize all of the information that you have in the system. So I think there's a lot of exciting future directions for after our initial focus of clean core transformation as well.

Alexander Greb (01:20:08.686)
So I guess then speed is not so much the target because if you do completely clean core within 42 minutes, there's no benefit in having it in 39 minutes, but it's definitely the scope.

Yeah, I think so. Yeah, so I think that's where we have sort of luxury of like, for a lot of other AI applications, I feel like speed and all of that is actually a huge factor, right? But we really are focusing on just the quality and sort of the functionality of the output. And we can afford to do sort of more, like, for example, tree-based search in planning and have tried different paths and stuff to really try to get to a system that just works really well. Yeah.

I think there are two aspects that Emma already mentioned that we are able to use the agents to help for the upgrade for clean core to get to S4. I think personally I have a second important area where I'm also excited is that there's so much

knowledge and differentiation for the large companies in their old processes. Some of them have good documentation, many very old things that they're not documented anymore, we have trained and also built our agents that more or less the knowledge of the past, which will be analyzed, will be built for every client in a knowledge core. So we store this knowledge.

And this can be then used for further advanced agents, which we can build for SAP, but it can be also used for BEYOND.

Alexander Greb (01:21:47.672)
like a best practice machine.

Yeah, you have the skills and knowledge what this company makes difference depending what is their market. They have usually 10, 20 % differentiation. Even when you say I will be go to clean core, there will be in the past was already a lot of differentiation what the leaders had in the market and they want to make sure that this will be for them directly more or less dedicated

Yeah, secured. this is for me even a more exciting factor. Okay, so upgrade is the demand and to do it, we will be sure we can do this for the client. And if it's 42 minutes or 30 minutes, it's another thing. I think we can do it even faster in the future.

when possible,

You have that ambition.

Prof Alexander Zeier (01:22:44.856)
Yeah, but I think this area is so exciting that you are more or less can build the business knowledge for the client and secure this in a dedicated AI system. This is for me a very interesting area where we can bring some additional value for the clients.

Yeah, so it's both like the you know sort of we mentioned maybe like best practices industry, etc both as also more importantly like sort of Really the preparatory like knowledge of the clients And what we do think about is as you onboard new human employees, right? There's gonna be some period we are onboarding them and they're getting used to like what does the company really do? How do you do certain things? How have the people done things in the past? As you're building out these like new AI agents essentially to automate workflows or help you

to take over certain processes, you can imagine there's maybe a similar process where you have to encode that knowledge into the AI agent. So the of knowledge core will be a way to automatically build to onboard these AI agents so you don't have to every time completely build it from scratch, which will take a very large investment. if you can really consolidate and codify this knowledge in a centralized system and built in a way that's easy for these.

like models to access then in the future it makes it much easier to build out different types of AI agents for various different tasks.

When I would ask you both, so please individual answers What are the question in your area? That nobody asks, but should be asked. What would it be?

Emma Qian (01:24:18.476)
Yeah.

Emma Qian (01:24:30.316)
like in AI and SAP, which is... The people ask that no one has asked. Too few people ask. Yeah, I do think the sort of...

Yeah, that's in that combination.

Alexander Greb (01:24:39.822)
or too few, which should be asked more often.

Emma Qian (01:24:46.83)
Well, I don't know if too few people, I think probably a decent number of people do think about it, but more people should think about it. It's what you mentioned about what it's really gonna look like in the future where these models are getting more more intelligent, right? And today it's able to automate X percent of the work and this number is only gonna go up in the future, right? So I think, I do think we will be like, you know, like the past, we have transitioned in many different types of jobs and new jobs get created. But I think the transitional period is something that's worth considering

because it might be, I think unless we're very thoughtful about how to approach it, you're seeing it with even the, maybe like, for example, self-driving cars and stuff, and a large number of people might lose their jobs there. So maybe really thinking about how do we support people to really be able to kind of take advantage of the technologies that coming out and help them into this transitional period. And I think maybe too many people are a little bit resistant and thinking that, this change is not gonna come.

My field is never going change in the future, which I just don't think is true for almost all fields in the world. So I'm kind thinking about that. But I do think we'll get to a world that will be better for everyone. if you are very resistant and not willing to learn and do these, like, use the new tools that are coming out, you might be at a big disadvantage compared to people who are willing to do that.

You're setting a very important point because the social responsibility that comes with that is huge. We should take it serious. support society, people to adapt to these kinds of revolutions that are happening. Otherwise we will have probably the things we will not like, number of people who did not...

Yeah.

Alexander Greb (01:26:34.294)
advanced the way that's probably necessary, did not keep track. And this is a very good point that you said. very grateful that you put it. Alexander.

Yeah, thank you.

Very good question. I would say it an exciting time now, which is a significant change. it is not just the next big thing, it's a different major thing. especially I'm...

with companies in San Francisco, but I'm sometimes also in Germany. so I really annoyed how the perspective, they don't see the opportunities, are all these regulations, which 900 pages of an EU AI act, so just that global companies...

not even make available the latest advancement for AI in Europe because they have the risk that they get huge fines of 6-7 % of the revenue that they think is too risky.

Prof Alexander Zeier (01:28:01.122)
When I take the German standpoint, I have the feeling that we do the wrong thing here. So we do not have the opportunity to take the advantages. We cannot apply them even. Besides, have not the core teams like here from Google DeepMind and other capabilities in the values that you build them. You not even can apply them. It's even more critical that you

Today I had many friends as consulting companies and you get 40 % productivity of AI projects now in average, sometimes more, sometimes less. Just think about every company get 40 % productivity gains. Yeah. You apply or not? Will you survive in a few years in a global competition? I don't think so. And so it is really an

I think to think about how you can use this and besides these changes.

Also think about today is so much shortage in the different regions. become older and there's, I have yesterday had somebody told me nice example. There's now a robot who helps the people who are elder, senior people who can more or less take care about them. Nobody will be able to have his personal assistant probably in the future. When you think about how the healthcare insurances increases every year.

in Germany and so on. So I think there is the opportunity to make it, when you use this phrase, better world, when you use AI in the different ways and think about, I think, the strategic thinking, what you can do with something when you are able to apply, learn, train, and leverage this. And this is, I think, very important also.

Prof Alexander Zeier (01:30:08.4)
that more or less everyone needs to think about. You can think about it personally, you can think about what it means for your company, how you can use this. This is my perspective here.

Fair point, definitely. this, in the meantime, 90 minutes long, very, very, very awesome and eye-opening discussion. People surely want to learn more about you, want to get in contact with you. Am I aware? Can they do this best?

Yeah, thank you. So Alexander and I, are co-founding this company, Nova AI. So you can find us at nova-software.ai online. You can get in touch with either of us at just Emma at nova-software.ai or Alexander at nova-software.ai. And yeah, we're very excited, especially if you're in the SAP industry. You have problems with your custom code. If you have a lot of legacy code that you want to move over and become clean

or just document, or you just want to build a lot of new applications, that's really where we can help. So, yeah, feel free to reach out there and we'll be very happy to chat. Social media, we're mostly just on LinkedIn and Twitter right now. So, LinkedIn is just Nova AI. And then Twitter, we're official Nova AI right now.

Any social media?

Alexander Greb (01:31:29.208)
Thank you very much. Emma Alexander, great to have you. It was really awesome. And I'm looking already forward to talk with you at the next stage.

Thank you for having us.

Yeah.