Florian Zirnstein is the CFO of Bayer Indonesia, and he’s also driving the use of generative AI within Bayer.
"At the beginning of the year, I asked myself, ‘where can I add value beyond where I'm currently working?’ And then I thought about AI,” Florian said.
Since then, he’s implemented 3 AI initiatives across the company. His success criteria (and how he thinks about ROI) may surprise you, coming from a CFO.
How Bayer is using AI internally
Bayer has made multiple LLMs available to its employees, including ChatGPT, LLaMA from Meta, and Mistral – all in a secure environment where they can confidently share confidential data with the LLM. Funding comes from the company’s global IT budget.
With these tools available, Florian was inspired to apply what he learned in Section’s AI for Business Mini-MBA. He identified 3 AI use cases, both based on his personal takeaways from the course, and by asking his teams what their biggest pain points are.
Use Case #1: Sales team training for new products
In 2024, Bayer Indonesia launched a new product line (hybrid corn seeds). At the time, the sales team had little experience selling hybrid corn seeds and limited context about the new product. This made it challenging for salespeople to confidently sell the product and answer farmers’ questions.
“I asked one business lead in our country, ‘What’s your biggest problem?’ She said, ‘We launched a new product line, but we don’t have the resources and know-how yet in the salesforce to sell that product.’"
Florian saw this as an opportunity for AI. He built a custom GPT trained on company information about the product as well as research and data that went into the product’s development.
After feeding his GPT all the relevant information he gathered from Bayer’s R&D department, Florian developed the prompt that guided the custom GPT on how to interact with sales people and which resources to refer to when answering different questions. He fine-tuned this prompt over two months by pressure testing the accuracy and clarity of responses to expected questions and adjusting the custom GPT’s instructions.
His system prompt also ensured that answers to entered questions could be provided in Indonesian as well as English, based on feedback he heard around a language barrier between farmers and salespeople. And to further streamline what is otherwise a very jargon-heavy product, he instructed the GPT to provide answers in terms a 12-year-old could understand.
After some training on the capabilities of the custom GPT, salespeople can now direct the complex questions they get in the field to the custom GPT and get an immediate, easy to understand answer.
Use Case #2: AI as problem solver and decision-making thought partner
Bayer is currently restructuring its organization to be flatter and more decentralized, encouraging more ownership and less micromanagement among employees.
"The idea is that people become more empowered to make decisions where the work gets done versus micromanagement and alignment through very hierarchical structures."
Based on this goal, Florian used his experience using AI as a thought partner (and a 1:1 coaching session with Section’s AI Mini-MBA lecturer, Geoff Woods) to spearhead a training program for Bayer employees on how to use AI as their problem solving and decision-making thought partner.
Florian’s overarching goal is to change behavior and help make generative AI usage a habit within the organization. He notes that currently, only a few employees consider themselves AI-powered.
He started by using AI as a thought partner himself, to find the best available problem solving and decision making framework focused on customer needs and found the concept of ‘design thinking’. He designed a custom GPT around the principles of design thinking, to help his colleagues build the muscle of working with AI.
He supplements that by providing training sessions on the ‘why’ and ‘what’ of design thinking, linked with stage-specific prompts that each employee can learn and apply for their individual role and team challenges.
Florian hopes that this embedded approach will help employees understand how AI can foster collaboration and help them get to solutions faster.
He also leads monthly live training sessions to inspire people to make using AI a daily habit. The live sessions mainly focus on teaching the art of prompting based on six relevant building blocks (he credits Section’s Ed Ortega as inspiration).
Before each session, he collects up to three real use cases from attendees for problems they need to solve. During the session, he prompts ChatGPT live to demonstrate not only how to use AI to come up with solutions, but also how to fine tune prompts to get to better results.
Florian hopes this supports Bayer’s wider goal of higher ownership and faster decisions among Bayer employees around their challenges (and increases wider AI adoption).
Use Case #3: AI as a consultant
Like many large organizations, Bayer relies on accounting and consulting firms for auditing, taxes, and consulting. Florian sees AI as an opportunity to reduce Bayer’s dependency on external consultants, challenge tax authorities' opinions in audits, enhance risk management, and embed AI into finance-related workflows.
He mentions this is still an active work in progress. The top use cases he’s testing now are leveraging AI for:
Transfer pricing documentation, a standardized report that multinational groups are required to provide to prove that their pricing complies with local tax law and international guidelines.
Expense report analytics to identify potential patterns of spend misuse.
Tax audit advice to assess tax authorities’ requests and opinions, as well as to help draft defense memos.
“We get 100 pages of opinions on how to handle these things. I let them run through generative AI to understand our financial risk and necessary profit adjustments.”
He also used AI to pressure test an external valuation for a sale of shares between affiliate companies. He hopes AI will eventually help Bayer move more of this kind of work in-house, rather than heavily relying on (multiple) pricey consulting firms.
“I produce more self-confidence in teams by challenging external opinions and not just accepting what they propose to us."
This has significant potential for cost savings and productivity gain, but much of the “ROI” Florian sees comes from the second opinion or added “insurance” in high stakes, financial decisions.
“I see AI going beyond data analytics and using it as a thought partner or financial advisor."
Measuring the ROI of Bayer’s internal AI initiatives
Of all the roles in an organization, CFOs seem most likely to demand quantitative measurement of AI ROI. But when it comes to evaluating the roll out of general purpose AI like ChatGPT, Florian believes that qualitative metrics such as employee confidence and improved decision-making are the right success metrics in the short term.
For example, he’s counting more confident customer conversations and higher quality customer engagement as success for the custom GPT he built for salespeople as he continues to gather evidence on its effectiveness on conversions.
“If they get back and say, 'Hey, the tool helped me to better engage with the farmer and really enter into a dialogue, I feel confident to provide an answer even if I need to use the tool during the conversation,’ I’m happy with that as better quality advice should result in higher sales."
In the long term, Florian will look at other metrics to understand the custom GPT’s impact on the sales process:
- Florian plans to look at CSAT (customer satisfaction) scores to understand if customers are happier with the interactions they had with salespeople who used the GPT assistant during their conversations.
- He is also asking the salespeople to log how often they use the custom GPT when interacting with customers. He plans to review sales performance vs. GPT usage to understand if higher usage correlates to more sales.
He’s taking the same approach (for now) in measuring AI as a thought partner and problem solver. So far, confidence is the largest success metric for those two use cases.
“Here in Indonesia, people typically hesitate to speak up. So against that background, I think AI can be very powerful in giving people more confidence in their own thoughts and to challenge other opinions,” he says. “It's a soft one, but if people come to me and say, 'Now I dare to speak up', that is the number one key thing I track in this program."
Long term, he is aiming to prove a 25% productivity gain across GPT users – a benchmark from the Harvard Business Review/BCG study that found that consultants using AI worked 25% faster. He’s planning to survey attendees prior to his training sessions to get a sense of their current AI usage and confidence in their work. Then he’ll survey them again 90 days after their training to see if there has been an increase in confidence and productivity.
Florian stresses that a behavioral change (confidence) has to come first, especially considering few of his colleagues are currently AI-driven. That's why qualitative metrics that demonstrate a mindshift are more important to him in the short term than getting hard numbers.
Parting thoughts
Your CFO will always ask how you’ll measure success when launching AI initiatives, but as Florian demonstrates, the initial stages of success can be measured by qualitative factors rather than quantitative.
"As a CFO, I know it should be more quantifiable, but I'd be happy if these people come back and say, 'Hey, it really adds value, and I can feel that I am more productive’. That would be good enough."
Florian’s emphasis on seeing growth in qualitative metrics, like confidence, gets at a real issue with AI: figuring out how to get enough meaningful adoption to start seeing productivity gains. And that hinges on strong change management and training practices.
His hypothesis is that these softer results will eventually lead to tangible financial returns. More confident salespeople will sell more, more strategic thinkers will come up with better solutions, and happier employees will add more value to the company.
So he is thinking like a CFO – he’s just also thinking like a good people leader.
