AI in Procurement | The basics of AI readiness with Dr Klaus Iffländer

Fabian Heinrich (00:01)
Hello dear listeners, I am delighted to be back at Procurement Unplugged after a long time with a special season on the topic of artificial intelligence with Dr K.I. Klaus Iffländer. Together we will do 4-5 episodes where we will talk about artificial intelligence in procurement.

Yes, Klaus, I'm particularly pleased to be doing this podcast with you today, or rather recording this entire podcast series with you. Why don't you introduce yourself to our listeners, tell us what you've been up to over the last few years and how you came to the topic of artificial intelligence?

Dr. Klaus Iffländer (00:51)
Yes, thank you, I'd love to. In any case, I've always been interested in data and analysing it, especially large amounts of data. Even back then, when there were no large language models and no data science, I was always interested in it. My first data project was at Deutsche Bank, where I was allowed to analyse a property portfolio during my studies.

And have already created forecasting models that have shown which properties are developing in which direction, i.e. in terms of value, in order to improve credit risk management. I was always interested in these forecasts and kept doing data projects, then also worked as a data scientist and also in

During my doctorate, I always worked with data and carried out interesting experiments, so I founded Breezy Services in the last five years and now work for various clients, primarily supporting them with consulting and development in the areas of artificial intelligence and data projects. Exactly, and since the middle of the year I've been Head of Research and Development at Mercanis.

In this capacity, I am extremely pleased to be here today.

Fabian Heinrich (02:19)
Yes, Klaus, we're delighted too. With you, we have an absolute expert here when it comes to AI and data volumes. And I would like to take a closer look at a few topics today in our first episode together. Firstly, there's been a lot of hype about AI for the last year or two. What's behind it? The other topic is the history of artificial intelligence, so that you can perhaps shed some light on the whole thing. Where did we come from? Where is the journey going?

And the third topic would be how can I apply this as a person and how can I further the transformation as a company. Well, then I would say, of course, AI is on everyone's lips, you only have to open the newspapers. If we look at the hard figures, we somehow see more than 25 billion investments in private companies, which of course mostly includes start-ups.

That triggers a big response. If you look at the performance now, people used to say I don't trust AI, then all of a sudden you read headlines about Chat GPT passing the Bavarian Abitur with a 2.0. It seems to have a lot behind it now and maybe it's not yet at the level of a human, but it seems to work relatively well.

What is your opinion or how do you see the whole AI hype that has been going on for a year and a half?

Dr. Klaus Iffländer (03:54)
Yes, on the one hand, from my technical point of view it's not all that new; the technologies on which it is based have been around for some time. But what is new is how it is now being used. On the one hand, the accessibility, i.e. that it is accessible to many people because you don't have to be able to programme or you don't have to be familiar with statistics, but you can simply chat with the Large Language Model.

It's simply new and is therefore experiencing rapid adoption. And the other thing is that it really offers something completely new in terms of benefits. So it doesn't just offer a simple output like forecasting models, an estimate so to speak, or some other type of evaluation like other statistical models, but what you now get as a result from the large language models is now really an understanding.

It is now a human understanding, at least simulated, which is really helpful, so that it can work as a system alongside a person or support work processes, just like a real employee would. And these two aspects, I think, are very novel and that's what sets it apart from other technological developments in this field. And that's why I think this hype is justified in a way.

The fact that so many resources are being channelled into it shows that there is huge potential in this technology and its application. However, it has to be said that the big revenue generators are not yet on the market, but the big providers of artificial intelligence that we know now are of course still in the early stages.

Venture capital financed and they are not making any profits now. None of them are. And also in use with other customers, i.e. all these B2B connections where AI is now flowing in, these are the first features that are currently being developed. But it remains to be seen whether this will ultimately live up to the hype in terms of revenue.

Fabian Heinrich (06:20)
Yes, those are very exciting insights, that on the one hand you say that the technology still costs so much money that the providers are all loss-making and are still dependent on venture capital or private equity money. on the other hand, as I understand it from you now, the hype is of course also so great in the sense that AI has been given a face for the first time, if you like, with such a chat interface.

And I mean, the figures that you see there on the consumer side with the Chat GPT app or Chat GPT web interface are really bombastic. So the growth is really more than exponential, especially if you compare it with all the previous ones. But if you take a look at the corporate world, the B2B world, we don't yet see the same level of adoption there. So I mean, if you say now, okay.

If you trust the new surveys from Statista, we see 2% of companies in Germany as pioneers and I think less than 30% use AI at all. How did this come about or how do you see the topic?

Dr. Klaus Iffländer (07:33)
Yes, I also find it interesting to see this development, because on the one hand, or my impression, if you are involved in this technology or in the industry, my impression now is that adoption is happening very quickly. You also know the use cases that are really going through the roof, where entire departments of customer service agents are being dissolved and taken over by bots and sometimes even with better quality or all these content functions where content is really being created, where LLMs can obviously now also make a major contribution. These are of course the big success stories, but you're right that many companies are still hesitant.

And I think this is due to various things. In Europe and Germany in particular, there is also the fact that data protection is very strict. So there is clear regulation, which is also right. But it often leads to companies being very, very cautious because they don't want to make themselves legally vulnerable or make a mistake before the...

Before the benefit is even really there. We have to be very careful because the providers of LLMs do not currently disclose exactly which sources they have used to train the models to achieve this understanding. And this can also be seen from the fact that all these legal disputes are being conducted in the United States.

Texts were used as training material. So this really is an unresolved issue, so to speak. And in Europe, where we now have the General Data Protection Regulation, you can actually be prosecuted for this and no company wants to put itself at risk. So I think that's one point. And then, I think it's also a well-known fact that there are still many companies in Germany where the processes are not necessarily so formally known and digitalised.

Digitisation is still an issue and it shows that in many companies the work processes are not necessarily digitally recorded in the same way. Many efficiency gains come from digitising things so that they can then be mapped using automated processes. And if that hasn't happened, i.e. if the

If the processes are not recorded digitally, then of course nothing can be done with AI. In that case, of course, large language models can't work with it either. Instead, it would be additional work to type out the texts or whatever again. Only then to get any answers from AI models. I think that's another factor. And then it's just an organisational thing. But I think that's just my anecdotal knowledge.

Fabian Heinrich (10:32)
Hm.

Dr. Klaus Iffländer (10:46)
I believe that many German companies have very established processes. We are very strong in mechanical engineering or the automotive industry, for example. And that many organisational processes, if we take manufacturing as an example, are very fixed and this creates a kind of cultural problem in that these processes are not so easy to change.

Even if processes are now known and could be supported by AI, companies are still hesitant simply due to cultural conditions, because established processes are reluctant to be changed at the drop of a hat, even if there might be a potential benefit.

Fabian Heinrich (11:38)
So, there are many reasons why there are still problems with adoption in Germany. The regulatory and legal aspects, the respect for such solutions, the cultural aspects and ultimately the data basis. I find the data basis quite exciting. You mentioned earlier that artificial intelligence is nothing new. It's been around for 40 years, if you look at the beginnings of Machine learning is then further ahead than deep learning and NLP.

Can I say that I've missed the boat as a company if I haven't followed the whole thing so stringently over the last 10 or 20 years or, on the other hand, can I say that it doesn't really matter? I can, as they say, leapfrog, because these new LLM models actually take the work off my hands somewhere and the data no longer has to be so structured and I don't have to have this expertise in linear knowledge building.

Dr. Klaus Iffländer (12:41)
Interesting question. If, then you can leapfrog it now If, then you can get a good start now because all these applications are really coming onto the market and because it's so accessible now. Before, up until a few years ago, you really had to do the groundwork first. You really had to have a good data basis and then also really actually have experts in the company who can even can create and train such models and derive any benefit from them.

Forecast models, for example. Sure, that's perhaps a simple example. There are certainly established tools and solutions. But when it gets more complex, for example ... ... Customer segmentation cases. So imagine ... you somehow have a lot of customers and you know ... a certain proportion of them leave in the next three months and you would like to make them an offer to stay with your service, then of course it's important to find out which customers these are likely to be, so that you don't give gifts to customers who would have stayed with your product anyway. so that's a segmentation problem.

And for such things defining meaningful models that are reliable and that you also know, i.e. that you can estimate the precision of them. you need a lot of background knowledge or certain skills in the company to calculate such things. And of course it's incredibly expensive to build up these capabilities as a company. And it's not quite as strict with LLMs. So you don't have to.

You don't have to do quite so much preparatory work, you don't have to build up quite so much expert knowledge and access is much easier. First of all, many products are currently available free of charge, at least for testing. And then the data structure requirements are not quite as high. Of course, you have to have some kind of data, at least texts and documents, but you don't have to have a perfectly structured data warehouse, as is the case for other data centres.

Machine learning use cases is the case.

Fabian Heinrich (15:07)
Yes, and I mean, you've already mentioned that the best time to start or the best leapfrogging would be now, so the time is now. How would I get started? Well, we always talk about AI readiness. So how do I know if I'm ready to get started? Am I always ready to get started? What do I need to do, what would be a guide to action if I am a medium-sized company in Germany in purchasing or a larger company, let's say, to get started now?

Dr. Klaus Iffländer (15:37)
Of course, it always depends on the use cases and it is often difficult to think about where AI can add value in the company. But if you make an effort or if you get the right people into a meeting and then put a room together, then you can come up with ideas where AI could really add value. And if we now talk about this.

Simple, my employees are now allowed to use Chat GPT. Anyone can do that somehow. Those are the copy-paste use cases. If you want to go a bit further than that as a company and really get into the topic and find the connection and really want to realise a bit more complex use cases, then I would actually base this on four things.

Firstly, the data, it just has to be there. Secondly, the infrastructure, i.e. where should what run, how do I want to set it up, simply from a technical point of view. Then the skills, i.e. what abilities and skills do my employees need, i.e. the team that will ultimately operate it.

And fourthly, the organisation, i.e. what I've just touched on, i.e. organisational processes or what is and isn't allowed. So simply these framework conditions that have to be in place organisationally so that the whole thing can work. yes, we can go through them briefly. On the subject of data, I would say yes.

Fabian Heinrich (17:22)
So maybe there again, if I may comply..

So as a kind of foundation, step zero, you're already saying that this would somehow be the definition of use cases. You shouldn't do anything AI-wise, but rather clearly define use cases where you can automate accordingly, or where it can help. That would be the very first step, wouldn't it? Did I understand that correctly?

Dr. Klaus Iffländer (17:47)
In any case

Otherwise, as a company, you run the risk of doing things or using resources on a technology without knowing what the benefits are. And I'm enough of a business economist to say that the benefit, the added value for the company, must of course take centre stage. You don't do this whole thing because of the technology or because of the hype, but because it can offer real added value in the company.

And there are certainly also professions where it doesn't play a role at all. Like, I don't know, in care or something. And it would be rubbish to force people to do something with AI anyway, just because it's a hype topic at the moment. On the contrary, it must always first be clear where added value can be created here, where it can create benefits and then orientate itself accordingly. And the technology must follow suit.

That's why it's always important to think about use cases first, where can the benefits be? And then you come up with the best technical solutions. So it's always like this.

Fabian Heinrich (18:59)
You just wanted to go into the subject of data in depth?

Dr. Klaus Iffländer (19:04)
Exactly, it's the same with the data, it basically represents the company, yes, digitally. And at least for the use cases or workflows that you want to support, you have to know at least some data, because if I have a workflow and there is no data or no texts, documents, anything that is digitally available, then it will also be difficult to tackle this with digital technology.

So that would be the first thing. And beyond that, the data has to be structured as well as possible. For example, if you really only have texts or documents, then you at least need to know what kind of texts or what kind of documents they are. In other words, what's in them, what you can draw from them.

Can or would like for the respective application. And they have to be suitable for this. Because, as with any IT system, garbage in, garbage out naturally also applies. So if you feed the models with information that is not well structured or is incorrect or unreliable, then the results will be exactly the same.

That's why you have to take care of how they are structured, i.e. where what is stored. Ideally, you have data in relational databases that can be queried and you can bring the entire company back together digitally so that the LLMs, i.e. large language models, really have a chance to understand connections and contexts. After all, that's a big difference here when it comes to large language models.

Can understand such things, at least as far as we as humans can judge. In the past, you could really only query individual data points and perhaps aggregate them or derive trends. But there was a lot of interpretation involved. And today, with the AI agents that we are now seeing coming onto the market, they can do this themselves. You can simply give them a lot of data and they then bring the understanding and interpretation directly with them.

So that's the big difference. Nevertheless, there is no getting around the fact that companies have to ensure data quality and the best way to do this is to assign responsibility for it. So you can't treat the data in such a way that it just accumulates somewhere and then lies around somewhere, but you have to get structure and quality into it.

It pays to have people responsible for it who take care of it, maintain it and at least know which data is reliable and of high quality and which is not, so that the right data can be used.

Fabian Heinrich (22:10)
Yes, exciting. I think we've heard a lot of content now. perhaps the listeners have once again broken down the five key steps to digital transformation in an exciting way. How can I get started now, to ask the whole topic with love?

Dr. Klaus Iffländer (22:26)
Exactly, so data as I said, then infrastructure, skills and organisation.

Fabian Heinrich (22:36)
No, I think that's a great guide to action and Klaus has just said that, of course you always have to have the business benefits in the background. are the use cases that make sense? Keyword use cases, one is on an organisational level, the other somewhere on an individual level. You know it yourself Klaus, here at Mercanis, I also tell the employees, they have to somehow scale themselves via these LLMs, via this artificial intelligence.

I mean, when the change came with the Office Suite at the beginning of the 90s, there were also opponents at the beginning who continued to use their slide rules because Excel was still so stepmotherly and semi-functional. At some point, of course, they were the kings who used Excel from the very beginning, because at some point, of course, you could... be 10x better.

Now, of course, the development is not just 10x, but probably 100x and much faster. So how do you see the whole thing, if I don't use these new tools and systems now or integrate these kinds of LLMs into my everyday life, will I be left behind or what, what do you advise, because they employees, especially now in purchasing, what can I do somehow? You also mentioned the topic of upskilling.

What is your recommendation?

Dr. Klaus Iffländer (24:04)
Exactly, so that goes right into the topic of skills. So in my opinion, you have to use these things now, just to learn how to deal with them. So I don't think you even need formal training or anything like that, as it might make more sense with Excel, but simply using these language models actually trains you so much that you learn above all to assess when I can trust it or when it's a help.

And when not, but to know until when the model has training data or what data it has access to. Because that's the only way I can ask questions or extract things, for example. And that's what people have to learn, i.e. to assess what it can and cannot do. After all, it's always stated that the language models sometimes hallucinate or make mistakes or can't do certain things and that's exactly what you have to learn to assess, because I can't just use it blindly.

And that's why I think it's a mistake to simply refuse to do so. Simply pretending it doesn't exist doesn't make you more precise or better at your job. Instead, I think it's a great help for the vast majority of office jobs. You just have to learn to assess the whole issue. So that's the thing

On an individual level, I would say, and on an organisational level, I think it is also a management challenge to encourage employees to try out and use these things. A lot will still need to be clarified: what about data protection, what data can I give out or not and where does it actually go? Will it be used to train the models or not? Because in the corporate context, you usually have to deal with proprietary data.

That's why you can't just blindly pass it back and forth with such an LLM. I would agree with that. Nevertheless, I think you have to favour this adoption by encouraging people and creating space for experimentation. Things like a hackathon in the development team so that they learn how to deal with it, or counterparts in normal working life.

Everyday life. So just letting people try it out, what can I do with it, how can I support existing processes with it or simply play around with it. You should simply give them this time, because I believe that it pays off in a very short time. Because you know it yourself from everyday life, you start using some tools and then you can't get away from them because they simply make you more efficient or because it's more fun.

This also favours many other things, because it often happens all by itself that the employees then realise that it would be great if this system now also had access to our internal data or to other data sources to link them and extract new insights or automate things that previously took a lot of time. And I also realise that it often comes with fears.

Fabian Heinrich (27:26)
Yes, here too.

Dr. Klaus Iffländer (27:26)
Because it is always a change and a change to existing processes that you are used to and that have worked. And yet it is unavoidable.

Fabian Heinrich (27:38)
Yes, this also shows that you just have to get started. So, I mean, what do we take away from today? Well, we also hear from you that AI is not a new topic, but the hype is real, to put it bluntly. And we've now heard clear instructions from you on how to get started in terms of, let's say, being AI-ready, becoming AI-ready, and there doesn't seem to be a better time to start loving the whole thing.

If I haven’t been a pioneer of AI until now, it’s quite fascinating what you just mentioned—that this isn’t just a one-way street where employees simply need to upskill and get started. That approach is probably not a bad thing, as I’ve just said, but the whole process also needs to be motivated and incentivised by management. Naturally, this is also connected to cultural transformation within companies.

Right, that’s it for our first episode, Klaus. It was very entertaining and a lot of fun talking with you. A quick preview of the next episode: we’ll be taking a closer look in detail at how to use Artificial Intelligence, especially the new technologies like generative AI, in procurement. We’ll also be talking about so-called racks or vector databases. So, I’m really looking forward to the next episode. See you then!

Dr. Klaus Iffländer (29:06)
Thank you, Fabian.