AI Strategy and ChatGPT with Christopher Koenig
“ChatGPT is a complete game changer.”
Chatbots are pure passion for Christopher. He sees artificial intelligence as a must-have for companies in all industries, which should not be missing in any strategic planning. A conversation with the AI consultant about long-term success with AI.
Christopher, did you talk to Siri or Alexa today?
No, I don't like them at all and don't use them for data protection reasons.
Which AI technologies do you rely on in your everyday life?
ChatGPT I don't really have to work at all since the launch.
Like all of us here.
Jokes aside. Artificial intelligence is already involved in many things, Spotify, for example, is already well informed about me and is designing my playlist. First, we have to ask ourselves how we actually define AI. Machine learning is already AI, i.e. all self-learning systems.
Artificial intelligence has been a top topic since ChatGPT was introduced last year. How long
Are you already dealing with the topic?
I come from an economics background, completed my economics master's degree in quantitative economics and have worked on innovation economics, among other things. There was already a lot of AI involved and I tackled questions about what could be done with the new machine learning models. My semester abroad in Berkeley certainly also influenced me. At that time, there was no way around the topic in America.
When was that?
2019. I've heard a lot about the thinking of the researchers there, who, by the way, invented Apache Spark and later founded Databricks. When I was there, we investigated, among other things, the impact of technologies and AI on the labor market and automation opportunities.
Was ChatGPT already an issue there?
Language models were not as good at that time. I would say that about a year and a half before the ChatGPT launch, they were more solid to use. Nevertheless, topics such as text recognition, speech recognition, recommendation systems and automated decision-making systems have long been topics in America. The broad mass of German companies that do not naturally deal with digital issues did not have this on their mind yet. Data storage issues were also barely present.
“With ChatGPT, you can generate added value for yourself as a user.”
What exactly fascinates you about AI?
It is very challenging, both technically and in terms of content. As an AI consultant, I need every use case
Understand, think of a specialist department, a business model or a business problem. I also get to know technologies again and again and think about where I can solve which adjustment screw in order to achieve good quality in the end. Each case is new and promotes flexible thinking.
In which areas or industries do you currently see AI particularly popular?
In particular, the advent of ChatGPT, and thus Retrieval Augmented Generation, has made AI accessible to many areas. Companies combine their data with ChatGPT to inform. This is ideal for those who are not yet ready in their data initiative. RAG is useful for everything that has to do with language. If there are areas in the company where people talk a lot, ChatGPT can provide support and completely take on tasks.
Another area is knowledge management, in particular making knowledge available. RAG Systems are able to answer application-related questions relatively easily by accessing the available source systems. This is interesting for all industries. Every company has knowledge that is partly documented there and is sometimes treated very diffusely.
Are companies asking about AI because it's hip right now?
Yes AI is particularly tangible with ChatGPT. With ChatGPT, you can generate added value for yourself as a user. AI can also make predictions and automate decisions. This also includes machine learning models that are closer to traditional statistics. This gives companies access to various AI topics.
What ideas do they have?
That is very different. Some companies are working extensively on the technologies and come to us with a specific use case of exactly what they want to automate. But there are also those who say: “Hey, we notice that a lot is happening around us right now, what kind of potential do you see in us to use AI?”
For this purpose, you then develop a AI Strategy?
We look at the company's business model and processes. In doing so, we keep in mind where AI can be used to establish new processes or reduce costs. AI can also be a means of increasing sales, such as a simplified product search in the online shop or improved customer service in terms of availability and low threshold.
“AI can also be a revenue-boosting tool.”
To what extent does your company empower you to deal confidently with the topic of AI?
We introduce the company to the topic of AI, but not theoretically, but in a very striking way. We are happy to provide examples of this. And then we ask lots of questions: What are your challenges? Your goals? What do your resources look like? For example, it may also be that a company is already facing strong sales growth, so that self-learning AI can help make better predictions with a new model and new data. We have several use cases for the requirement engineering process.
Chatbots are currently particularly popular for use in customer service. How do companies implement them profitably?
Automating processes is an essential process that requires a lot of resources. It makes implementation easier when the technological and content setting is available. You should also consider whether it must be a fully automated system, or a supporting one, such as co-pilot from Microsoft, might be enough. It is difficult when a company settles on a use case and then realizes during implementation that it doesn't need it at all. Investing in AI is always worthwhile when it makes processes better or more efficient — which, of course, must be reflected in costs or turnover. AI can make sense, particularly for resource-intensive processes.
What are your experiences with chatbots as a user?
I've never really been a fan of chatbots, I have to say honestly. Before ChatGPT, I had never been helped by a chatbot as a customer. It has now rotated 180 degrees. I would love to only work with chatbots, rather than with prediction methods.
“I would love to only work with chatbots.”
Whether it's a chatbot or other automation: How do I choose the right use case?
It is better to choose a use case that also has relevant benefits if successful, even if the outcome is uncertain. But you can't say that in general terms, it always depends on the company.
Is there a typical project process?
That is also different again. Companies rarely come to us with perfect data. Data preparation in machine learning projects is often a large part of the work. In ChatGPT projects, this in turn does not take up as much space. The way we interact with customers is very different. We generally work according to the CRISP-DM framework, a standard for data mining projects. Otherwise, we act in an insight-driven manner and then adjust the requirements again and again.
How long does an AI project usually take?
We complete an MVP within around three months. It takes us three to four days for an initial mockup, on the basis of which the decision on a further course of action can be made. After the MVP, there is the lengthy part. In production, we protect ourselves against all contingencies and take ethical and legal measures. This could take another four to five months. In this usage phase, we collect a lot of feedback from users and continue refining at the same time.
User feedback is elementary here, isn't it?
Totally. Here you can also see the difference between prediction and chatbot projects. With machine learning algorithms, we can optimize the model specifically for a KPI or metric. This is not possible with chatbots because ChatGPT cannot always provide the same answers. Measuring how good each answer is much more difficult. We always rely on user feedback to know how good the answer is.
How do you document this feedback?
Since we use LLM GPT-4 and not the OpenAI web app, we develop a frontend ourselves and install a feedback mechanism there. It's similar to the feature that ChatGPT already has: thumbs up or thumbs down. Our development cycles are longer as we collect feedback. Faster testing would be possible, for example by generating synthetic test questions and answers, keyword groundedness evaluation. But through our approach, we directly address the human factor and the business case. In this way, we avoid optimizing past the problem.
“We are no longer just building the model logic, but the model itself learns what the best results are.”
What else is needed to be successful with AI?
Certainly a certain degree of openness to technology. And that across the entire company. Change management is an important issue, and the relevant departments should be involved, i.e. to alleviate fears. Integrating AI can cause friction in the short term, but persevering is worth it. In order to get employees involved, it makes sense to find solutions and processes for abrupt changes.
What data basis should be available?
Data resources and data quality are important topics. However, the data doesn't even have to be structured on a large scale; unstructured data from SharePoint, PDF files or website content, for example, can also be processed. But some data must be there. They must then be cleaned.
How should companies be strategically and technologically positioned in the next few years to move into a competitive future driven by data and AI?
What I would really advise companies is to pick out a use case and say: 'Let's just do it now! 'You should start the whole thing as a secured test balloon, which can also have a certain radiance internally. So that the entire workforce is willing to work with such technologies. ChatGPT is a huge thing, really, and you should use that now.
Just start that way? After all, these are investments that many have to plan for.
Sure, but that's a strategic question. Most companies are so involved in their day-to-day business that they pay too little attention to such technology and do not see the potential. It is certainly a topic for the C-level. This is where you have to sit down and think in a structured way how AI can be integrated. ChatGPT is a complete game changer, comparable to the introduction of the PC. How do we deal with this in the long term? Not one or two projects are enough, but you have to think long-term. It is a technology that can turn entire industries upside down.
Can small and medium-sized companies benefit from AI as large companies?
You can also Build a solution that quickly delivers added value with relatively little effort. AI is an exciting topic, especially for companies that cannot afford many employees. The development costs pay off quickly.
Which AI technology would you like to further develop for yourself?
I'd like to build my next chatbot using LangChain, an open source framework designed to build applications using LLMs. That can be exciting.
This article appeared in a similar form for the first time in Edition 01/24 by data! You can find all issues and articles of our biannual magazine here:
Data! Magazine: Cloud Services, Data Analytics & AI | taod
