Why pharma needs smarter decision-making, not just more digital systems 

The pharmaceutical industry has invested heavily in digitisation over the past few years. What gaps in existing digital systems prompted the development of an operating system focused on decision-making rather than workflow automation? 

Most people in this industry looked at the same landscape and saw a digitisation problem. I saw a different problem entirely. The industry did not lack systems. It lacked reasoning. 

Every major digital investment in pharma over the past decade was built to answer one question: what happened? ERP, LIMS, CRM, extraordinary at recording. Built to tell you what occurred after it occurred. 

Nobody built something that answers a different question: what should we do next and why? 

That gap between recording and reasoning is what we saw. And that is what RIKO was built to close. Not because we wanted to build technology. Because we needed it to run the business. 

RIKO is introduced as an AI-native pharmaceutical operating system built on operational intelligence. How do you validate the reliability and consistency of its recommendations in regulated environments where human accountability remains critical? 

We designed RIKO around one principle from day one, it recommends, humans decide. 

Every output RIKO produces shows its work. Not just what it recommends but why, with the specific data points and their weights visible to the person making the decision. That person can agree, override, or challenge it. Every override is logged. Every reason is captured. 

That audit trail is not a compliance afterthought. It is the point. Over time the pattern of where humans override RIKO tells us exactly where the model needs to improve. The human judgment does not just check RIKO, it trains RIKO. 

On validation – we apply the same standard we apply to any analytical method in a GMP environment. You retrospectively test the system against historical outcomes. You document the accuracy. You set thresholds. You review it periodically. The industry has been doing this for computerised systems for decades under GAMP principles. An AI recommendation engine is not fundamentally different. It requires evidence that it does what it claims to do, consistently, within defined parameters. We treat that as a design requirement, not a burden. 

With regulators placing greater emphasis on data integrity, quality risk management and supply chain resilience, how do you see AI-driven decision support changing compliance practices across the pharmaceutical industry over the next five years? 

The shift that is coming is from reactive compliance to predictive compliance. And that is significant. 

Today compliance in pharma is almost entirely retrospective. An inspection finds something that went wrong months ago. A warning letter documents a pattern that existed for years. A CAPA closes a deviation that already happened. The entire quality management architecture is built around responding to events after they occur. 

What changes when you have continuous intelligence when a system is monitoring your supplier’s GMP certificate, inspection history, deviation patterns and regulatory alerts in real time is that you stop being surprised. You see the warning letter pattern before the warning letter arrives. You see the GMP risk before the audit. 

Over five years I think three specific things happen. Regulators begin recognising continuous monitoring as a quality indicator in its own right. A company that can demonstrate real-time supply chain surveillance will be treated differently in an inspection than one relying on periodic audits. Second, data standards will emerge that allow intelligence to flow across company boundaries without compromising confidentiality. Third and this is what most people are not thinking about yet, the definition of a qualified supplier will start to include the supplier’s data connectivity. Can they give you what you need to monitor them continuously? If not, they are a higher-risk supplier by definition. 

The companies investing in this infrastructure now will not just be better positioned. They will have set the standard everyone else has to catch up to. 

Many pharmaceutical companies struggle with fragmented data across manufacturing, quality, regulatory and commercial functions. What are the biggest barriers to creating an integrated decisionmaking framework, and how can companies overcome them? 

I will tell you what it is not. It is not the technology. The technology to integrate these systems has existed for years. The barriers are entirely human. 

The first is data ownership politics. In most pharmaceutical organisations quality owns quality data, commercial owns commercial data, regulatory owns regulatory data. Each function treats its data as institutional territory. An integrated decision framework requires that data to cross those boundaries. That is a culture and governance problem, and it can only be solved from the top. If the CEO does not personally own the integration agenda it does not happen. 

The second is what I call the single version of truth problem. When three different teams maintain three slightly different versions of the same product portfolio, different molecule names, different supplier codes, different filing statuses, no AI system can reason coherently across them. You have to fix the data before you can build the intelligence. That work is unglamorous and nobody wants to fund it, but it is the prerequisite for everything else. 

The third is specific to regulated industries. People in pharma are conditioned to be conservative about change. An AI system influencing a compliance decision feels riskier than a spreadsheet doing the same thing, even when the AI is more accurate. You overcome that by starting with lowstakes, high-frequency decisions where people can build trust in the system before applying it to anything consequential. Trust is earned incrementally. You cannot shortcut it. 

The platform claims to predict quality and compliance risks, including nitrosamine-related concerns, at an early stage. How should the industry balance predictive technologies with established regulatory and quality assurance processes to ensure confidence in such systems? 

Predictive tools should trigger the regulatory process, not replace it. That distinction matters enormously and the industry needs to hold it firmly. 

ICH M7 does not change because RIKO exists. What changes is when the risk is caught. Before the route is selected, not after the batch is manufactured. The regulatory process stays intact. It simply starts earlier. 

The confidence question is straightforward. A model whose outputs are reviewed by a qualified person before any action is taken is a defensible tool in a regulated environment. What is not defensible is using predictive confidence to skip steps that exist for good reason. Low probability in a model is not the same as tested and confirmed. That line does not move. 

As AI becomes more deeply embedded in pharmaceutical operations, what governance frameworks including transparency, validation, cybersecurity and data ownership do you believe will be essential for wider industry adoption? 

Four things. And I will be direct about why each one matters. 

Transparency first. Every AI output in a regulated context needs a visible reasoning trail: not a black box recommendation, but a documented chain from input to output with sources and weights visible. If someone asks why a decision was made, the answer cannot be the AI said so. It has to be the AI said so based on these specific inputs, reviewed and approved by this person, on this date. Design for that from day one. 

Validation second. The GAMP principles the industry already uses for computerised system validation apply here. Validation rigor should match the risk of the decision being influenced. A screening tool that identifies molecules for commercial consideration needs different validation than a system influencing batch release. Proportionate, documented, periodically reviewed. 

Cybersecurity third and this one is underestimated. Supply chain intelligence, regulatory filing data, commercial pricing: this data is valuable and it is increasingly targeted. The security architecture needs to match the commercial value of what the system knows. 

Data ownership fourth and this will be the most contested question in the industry over the next five years. The industry needs to establish those principles now before the systems are sophisticated enough that the answer has large commercial consequences. Write the rules before anyone has broken them. 

The principle underneath all four is simple. Governance is not there to restrict what AI can do. It is there to make what AI does trustworthy enough that the industry can rely on it. Restriction without trustworthiness is bureaucracy. Trustworthiness without governance is recklessness. You need both.

Kalyani.sharma@expressindia.com
journokalyani@gmail.com