Scaling enterprise innovation beyond pilots: 5 lessons from Bayer’s AI startup

Inside Bayer’s approach to building GenAI applications that deliver real value.

How do you go from sandbox experiments to fully deployed GenAI solutions that drive business value without getting lost in the hype?

That’s the question Bayer has been answering through its pioneering work in GenAI, particularly with its domain-specific model: E.L.Y. (Expert Learning for You)

In a recent episode of AI Insiders, Vincent Pirenne sat down with Dan Kurdys (Global Business Leader for GenAI at Bayer) and Sachi Desai (VP of AI Go-to-Market & Partnerships), and their story so far holds lessons for any organization looking to move past pilots into scalable innovation:

1. Start with the problem, not the tech.

Too many AI strategies start with the technology, not the problem. Bayer did the opposite:

Their GenAI journey began with a practical pain point: farmers were waiting too long for agronomic answers. Sales reps felt the strain of needing to show up as experts, and agronomists were overwhelmed trying to scale their knowledge.

That shaped the real problem: how to get timely, expert agronomic advice into the hands of frontline teams?

Everybody wants an answer 24 seven and there’s all kinds of questions at certain times of the year because decisions have to be made.

⸻ Dan Kurdys

In their experimentation, Bayer then found GenAI to be a strong force multiplier for teams that were overwhelmed, and by its adoption, they achieved 60% faster response time.

This clarity of purpose helped E.L.Y. find product-market fit early: GenAI wasn’t the solution from the outset; it became the best-fit solution once the problem was fully understood.

You start with a problem, and there’s a tool for about every problem. We happened to find that GenAI was a good tool for our problem

⸻ Dan Kurdys

2. Build fast, iterate faster.

The E.L.Y. team didn’t wait for a perfect strategy or enterprise-wide roadmap. They moved quickly by wrapping existing frontier models (like GPT-4) with retrieval-augmented generation (RAG) techniques and Bayer’s proprietary logic and documents.

All to get a working prototype into users’ hands.

It worked. Compared to base models, the tool delivered:

47% more accuracy on a 5,000-question benchmark
20% better breadth of knowledge recall
60% faster response time in field tests

We proved our data mattered … And once we saw that, we just kept running the next experiment.

⸻ Dan Kurdys

The team kept focusing on running experiments; each one designed to unlock a new constraint: accuracy, latency, or cost.

As new needs emerged, they adapted the stack. Eventually, they tested smaller language models to strike a better balance across performance and cost.

3. Make innovation scalable by design.

Instead of hardcoding use cases, the E.L.Y. team created an SDK-based architecture that lets different teams across Bayer spin up new applications fast on top of the existing capabilities.

One core insight: their frontline teams weren’t going to open laptops in the middle of a cornfield. A mobile app was essential:

You’re not opening a laptop in the middle of a cornfield, the mobile app drove the engagement.

⸻ Sachi Desai

Their SDK now powers both the core E.L.Y. app and other internal tools that serve curated agronomic knowledge whenever and wherever it’s needed. This approach also enables the team to continuously test new GenAI models behind the scenes: future-proofing their infrastructure.

4. Culture is the hidden infrastructure

Technology alone doesn’t drive adoption. People do.

Bayer’s Dynamic Shared Ownership (DSO) model enabled the E.L.Y. team to operate like an internal startup: lean, federated, and empowered. Experts from agronomy and knowledge transfer teams shaped onboarding flows and testing protocols from day one.

This cultural setup made a huge difference in rollout. One example: the ability for reps to auto-generate localized, agronomically accurate conversation scripts continues to impress internally:

It blows our people’s minds. They’re like: my gosh, I now have it in my power to create custom collateral that is localized, that has my expertise in positioning.

⸻ Dan Kurdys

In short, GenAI wasn’t imposed top-down. It was built with the people who needed it most.

5. From use case to movement

What began as a tool for agronomic Q&A has become a platform: a bridge between Bayer’s deep (often inaccessible) knowledge and frontline action.

Much of the company’s most valuable expertise (housed in internal PDFs, legacy systems, or the heads of experts) was out of reach for those who needed it most:

The best knowledge in ag often lives in portals no one reads. We’re making it accessible at the moment of need.

⸻ Dan Kurdys

The biggest breakthrough wasn’t flashy automation. It was simplicity. As Dan Kurdys put it: “Sometimes the real innovation is just replacing paper.” That focus on reducing friction (not reinventing the wheel) has helped E.L.Y. build momentum beyond its original scope and set a new standard for how Bayer approaches innovation at scale.

Advice to leaders: don’t wait to get it perfect

For companies trying to break out of AI pilot purgatory, Dan and Sachi offer clear advice:

Find the knowledge gap in your business. Almost every enterprise has one.
Start small with structured experiments, not 10-year business plans.
Use results to earn your next round of support.

And remember: you’re not building an LLM. You’re delivering an outcome.

⸻ Sachi Desai

What BOI sees here

Bayer’s E.L.Y. team is a successful blueprint for enterprises serious about GenAI: start with the business problem, move fast with modular infrastructure, and scale through cultural alignment; not just tooling.

This isn’t just a GenAI success story. It’s what happens when AI is embedded with intent, speed, and the mindset to solve real problems first.