Good practice

And the second pillar builds directly on that.

It's responsible AI, GXP, and cybersecurity.

This is the governance layer.

In a highly regulated environment, you can't just throw clinical data into a black box neural network.

And the governance metrics listed here are heavy.

We've got data integrity, cybersecurity, model drift, explainability, human oversight, and something called GxP validation

It's a lot to manage

For anyone outside the industry, what does it actually mean to validate an AI model to GxP standards?

So GxP is basically the pharmaceutical industry's overarching framework for good manufacturing and clinical practices

Okay

It is rigorous, heavily documented, and strictly enforced by agencies like the FDA.

Now, normally, GxP applies to physical things

But the real friction point, the phase that really separates pharma from Silicon Valley, is step six: validate and govern.

Underneath this step, the blueprint explicitly demands evidence, responsible AI practices, and adherence to GxP.

Okay, let's pause there.

Mm.

GxP is a massive acronym in the pharmaceutical world, for you listening who might not know.

It stands for Good Practice quality guidelines.

Mm-hmm.

So you have GMP for Good Manufacturing Practice, GCP for Good Clinical Practice.

It's amazing.

And crucially, because you followed the foundations and step five, that entire automated decision was documented and logged perfectly for GxP compliance.

That is right first time manufacturing in a nutshell.

It is a domino effect of integrated technology, standard processes, and validated data, saving a multi-million dollar batch of medicine.

So a lot of the processes that we put in place are making sure there's change control, there's audit trail, there's an ability to really abstract away what the operator does so that everything else in the background runs and can be compliant.

And that is the biggest, the biggest thing that we're finding now, where labs are trying to move into GXP environments, and they don't have that structure in place.

And we really need to engage to make sure that they understand.

They know what they don't know, right? And so we can go ahead and work together to make it compliant, basically.

That brings us right to the automate step.

The roadmap highlights two things here, automation and RPA, which stands for robotic process automation and GXP compliance automation.

Let's break that down.

Starting with GXP, what does that actually mean?

So GSP is basically an umbrella term in the industry.

The G stands for good, the P stands for practice, and the X is just a variable.

Yeah, it can mean good manufacturing practice, good clinical practice, good laboratory practice.

It is basically the overarching set of quality guidelines that keep drugs safe.

Otherwise you end up controlling everything, control tower at the center, even though you call it hybrid.

What role would you suggest senior leadership play in setting risk tolerance for innovation within GXP environments and how explicit should they be?

Well, Risk aversion is the typical problem with senior leaders.

And that's what they kind of put.

So once the structures are in place, that is where your AI really comes into picture.

And from the AI agents perspective, we actually launched a couple of agents in IBP, demand fulfillment agent, delta planning agent, there was also something on the jewel that we launched, which is looking at compliance monitoring there, which is exclusively for GXP systems and pharma systems, but the the demand fulfillment assessment agent uh it is actually traces all the fulfillment gap between the shortfall which batch which campaign what is the constraint around it so looks at all those parameters and helps you plan those so there has already been a lot of work that is launched and much more is upcoming and of course customers can develop their own ai models based on once you have these data structures in place all the building blocks in place

To be specific around how we were approaching AI, I think Aparna nailed it in that in the early days, so I would say like pre-2025, the AI that we were exploring and employing was more around primarily in the demand planning space.

So we were using a lot of AI and ML around demand forecast optimization, outlier detection, anomaly detection.

But it was a three in the pod, Illumina, SAP and the SA partner for Illumina who has actually worked on these things with ECC connection.

And IBP was, I would say, also still not yet matured when actually Leon was implementing IBP.

But the story here is that Illumina did not wait for Rai's perfect conditions to be on the cloud, for all the GXP master data to be on the cloud.

They just started it pretty early in their journey with ECC connections.