When working with Bayesian optimization there’s a common misconception amongst first-time users:

When working with Bayesian optimization there’s a common misconception amongst first-time users:

The algorithm will immediately pinpoint your ideal process parameters and deliver your highest yield right from the start.

In reality - and when starting without prior data - before any objective optimization can happen, the model must first map the chemical space.

That means learning not only where success lies, but equally importantly where it doesn’t.

These early “non-optimal” experiments often frustrate scientists because they appear to give poor results. 

But in truth, they’re the foundation for everything that follows. They fill in the underrepresented parts of the data landscape - regions where the chemistry fails, where the yield drops, or where the process simply doesn’t behave as expected.

For Bayesian optimization to work effectively, it must understand both the peaks and the valleys of performance. 

This balance between exploration and exploitation is what allows the method to confidently steer you toward high-performing conditions.

In batch optimization, some of your experiments will intentionally be explorative. They may produce lower objective values - but they are just as crucial as the ones that look successful. 

Each point contributes to the global understanding of your reaction space.

Optimization is iterative, data-hungry, and scientific at its core. It’s not just about chasing the maximum, it’s about understanding why it exists, where it lies, and how to reach it with confidence.

If you’re curious to learn more about how we use Bayesian optimization in real-world chemical development, feel free to drop me a message - I’m always happy to jump on a quick call and walk you through how the method works in practice.

‍