Which digital technologies have delivered tangible operational value over the past 12–18 months and which remain over-hyped?

Digital twins have delivered genuine value in bioprocessing: improved process understanding, faster development cycles, and better scale transfer. But they are currently overhyped relative to what they can actually achieve. The limitation isn't computational; it's biological. Most process variables can be observed and modelled but not directly interrogated or controlled at the level where the biology is actually decided.

The modelling frameworks that will move the needle are those grounded in empirical, mechanistic understanding of the system, not just pattern-matching against historical process data. Correlative models describe what happened under a fixed set of conditions. Mechanistic understanding, combined with machine learning (ML), can tell you why it happened, and how to change it. That depth of understanding is what makes modelling meaningful, but insights are only as valuable as your ability to act on them. Optimization at this depth requires a lever precise and dynamic enough to translate model recommendations into real biological outcomes.

Prolific Machines is working exactly at that intersection: using optogenetic control of gene expression to both probe the biology at high throughput and build hybrid mechanistic ML models precise enough to act on. That combination is what the next generation of quality by design (QbD) looks like.

Though artificial intelligence (AI) is dominating conversations, it’s only delivering real operational value when applied to specific, practical use cases. Network management is one of the clearest examples.

In a healthcare environment, the network underpins everything, from patient monitoring systems and secure transmission of lab results to staff communications and in-room patient Wi-Fi. It has to stay up 24/7 with zero tolerance for downtime, even as upgrades, security checks, and hardware changes occur in the background. At the same time, IT teams are under constant pressure to resolve user issues while maintaining performance across an increasingly complex infrastructure.

This is where AI is proving its value. AI-driven network management enables automated optimization, real-time detection of anomalies or security risks, and faster root-cause analysis. In a critical environment like healthcare, many IT departments are moving to a 'human-in-the-loop' system, where IT simply has to review and approve AI actions. The result is a more resilient network, a better user experience, and significantly less manual work for IT teams. In a hospital environment, those seemingly incremental efficiency gains can make a huge impact over time for both medical staff and their patients.

In biologics CDMOs managing multiple client processes under GMP conditions, digital tools enhancing data integrity and process understanding have delivered the most value in recent years. One notable example is the manufacturing execution system (MES)-based electronic manufacturing batch record (eMBR). By digitizing batch documentation, eMBR improves traceability, reduces manual transcription errors, and streamlines deviation investigations.

Multivariate data analysis (MVDA) has become increasingly valuable for process monitoring. By contextualizing complex manufacturing data, MVDA enables earlier detection of abnormal process behavior and facilitates faster, more systematic root-cause analysis. Computational fluid dynamics (CFD) also delivers practical benefits during scale-up and technology transfer. Through the simulation of mixing performance and the evaluation of equipment differences across manufacturing sites, CFD helps mitigate technical risks and supports more efficient process transfer.

However, fully autonomous AI-driven manufacturing remains largely aspirational. Equipment interoperability gaps, analytical integration challenges, and regulatory validation constraints limit full automation. While digital tools improve process monitoring, translating insights into autonomous predictive process control is difficult due to process variability and regulatory requirements. Ultimately, the highest-value technologies today are those strengthening data reliability and process understanding rather than replacing human decision-making.

In the past 12–18 months, we have seen real-world value created from technologies that many of us would not necessarily consider new, but in recent months, advancements have accelerated rapidly — particularly for manufacturing organizations. Some examples include modern industrial data platforms with a focus on time-series data (a sequence of data points indexed, listed, or graphed in chronological order). Many industrial manufacturers already have years of historical data that can be mined to glean insights and create predictive models to enhance operations. In addition, a selective, applied use of generative AI can augment workflows, such as deviation and investigation management, when operations run into anomalies. The ability to assess root cause and take corrective action is improving the bottom line more quickly. These applications are worth keeping an eye on in 2026.

Autonomous AI agents are vastly overhyped for application in industrial manufacturing environments. It’s the autonomy itself that requires more work before these agents’ true value can be realized in the industrial manufacturing world. When they focus on rote processes where decisions are clear and rule-based, they can add real value, for example, in accounts payable processing, discount optimization, and order fulfillment. From a capability perspective, we are still a few years away from being able to build the trust required for an autonomous agent to run critical manufacturing operations. However, many organizations are considering the ethical dimensions of AI, and in our industry, human intervention and oversight are critical when manufacturing medicines.

In almost every instance when talking about the tangible value of digital technologies, real value is driven by situations where digital tools augment human decision making rather than replacing it entirely; the latter is where the hype tends to lie.

In silico simulations and digital twins significantly accelerate research and reduce failure rates throughout the scientific lifecycle. For example, in the discovery stages, in silico tools, such as Lab Thread, can predict the outcome of restriction digests and ligations before a scientist even steps foot in a lab. In manufacturing, “process twins” have already demonstrably reduced cost of goods and development timelines for novel biologics. In silico models are even showing value in the clinic now, as the FDA has started to accept data from synthetic control arms to help benchmark efficacy in instances where recruitment of a control group would be either extremely difficult or unethical. But none of these technologies replace the need for humans to plan, validate, or execute the final experiments.

Similarly, while smart scientists are already leveraging AI to handle tasks such as large-scale literature reviews, documentation drafts, or data cleaning to make their workflows more efficient, the idea of fully autonomous AI labs or fully AI-driven drug discovery remains more closely in the realm of science fiction than science fact.

More closely aligned to our own area of interest, digital systems that can provide an automatic audit trail have already significantly reduced time spent on QC and preparation for regulatory filings. However, where first-generation digital lab management tools have so far failed to deliver is in connecting their “digital islands”; for example, a lab might use an ELN for record keeping and a LIMS system for sample management, but the two don’t speak to each other. Scientists still have to copy-paste between browsers, creating inefficiency and opportunity for error. Similarly, most ELNs currently on the market are great at storing PDFs, but terrible at making data "AI-Ready." If the data isn't structured at the point of capture, it's useless for the in silico models mentioned above. This is, of course, where we expect Lab Thread to do better.

When operational data is captured electronically and integrated across global sites and multiple digital systems, it enables the use of advanced tools, such as automated reporting, performance monitoring, and predictive supply planning. These capabilities provide practical operational insight and support faster, more informed decision making.

Enterprise resource planning (ERP) platforms continue to deliver significant value in regulated manufacturing environments. In our case, a global ERP system supports data-driven supply planning, lot traceability, and regulatory-compliant material management. These capabilities are particularly important for our GMP-compliant column packing services, enabling nimble supply planning to support industry-leading turnaround times of 3–5 weeks.

Equally important is the integration of quality systems within a harmonized global operating framework. Electronic quality management systems support change control activities, deviation management, and audit readiness, improving consistency and regulatory compliance across sites.

Emerging technologies, including artificial intelligence and advanced analytics, are becoming increasingly valuable. Their impact is most evident when supported by well-structured operational data. In many cases, the perception of over-hype reflects the gap between emerging technologies and the foundational operational data required to realize their full value.

Our continued investment in digital infrastructure ensures these systems can scale smartly as technologies and regulatory expectations evolve. With more than 30 years of bioprocessing experience, contributing to over 20 market approved drugs, we see digital technologies delivering the greatest value when they enable data-driven planning that support reliable, compliant manufacturing processes.

What's delivering real value:

• Protein structure prediction has become infrastructure. AlphaFold3 and RoseTTAFold All-Atom now handle protein–ligand and nucleic acid complexes, compressing early-stage structure determination from months to hours for drug target identification and antibody engineering.

• Genomic foundation models have demonstrated clear advantages for variant effect prediction and regulatory element classification, outperforming classical methods on well-benchmarked tasks.

• Biological image analysis, from digital pathology to cryo-EM reconstruction, has moved from research tool to production workflow.

• Scientific literature AI has become a real productivity multiplier for literature exploration, synthesis, and hypothesis generation.

What remains over-hyped:

• End-to-end autonomous drug discovery. Platforms generate interesting hits, but hard translational challenges persist. Biological judgment still drives the critical decisions and can bottleneck effective and efficient use of AI.

• Multi-omics AI integration. Impressive demos, sparse production deployments. The data harmonization problem is chronically underestimated.

• Large language models (LLMs) for bioinformatics pipelines without expert oversight. Errors in biological contexts are silent. They don't throw exceptions; they mislead.

There is a consistent pattern:

AI delivers where data is clean, versioned, and queryable. Where it disappoints, the root cause often lies upstream in fragmented, unharmonized, ungoverned data. Data readiness remains the real bottleneck. The most effective research approach today pairs AI-accelerated hypothesis generation with human-directed validation, and platforms that harmonize and integrate rich multi-omics data with computational workflows make that validation faster and more rigorous. The industry's most consequential AI investment right now may be in that data infrastructure layer.

Sterling’s data, IT, and scientific teams have successfully implemented secure, scalable AI tools that deliver tangible operational value while protecting sensitive customer data. Across global R&D, AI is now routinely used to accelerate data collection, streamline report writing, summarize literature, and guide hypothesis generation for technical problem solving.

Within process development, the chemistry services team has integrated AI tools into daily workflows to improve efficiency and ensure scientists can focus on the highest value activities. By automating routine but essential tasks, such as literature searches, preliminary safety assessments, and early report drafting, AI reduces time spent on work that does not require deep scientific/human judgment. The technology also helps minimize “gray work,” including time lost searching for information across multiple systems, reconciling data sources, or managing administrative overhead, thereby improving productivity and consistency.

Most importantly, these efficiencies allow chemists to concentrate on critical development challenges, where their expertise has the greatest impact. AI further supports this work by helping articulate hypotheses, design experimental approaches, and augment complex problem solving. This balanced deployment ensures that AI enhances scientific capability while maintaining the rigorous standards expected in pharmaceutical process development.

Generative AI tools like ChatGPT and Copilot are now integral to daily work, supporting text generation, summarization, ideation, and structured thinking. Their widespread use shapes employee expectations. Although less tangible, their value is impactful in building a culture of AI literacy across organizations.

AI integrated in workflows, like service desk for ticket response and triage, and software engineering for code, tests, and documentation is delivering value to organizations by reducing cycle times on knowledge work. Based on the same generative AI language as consumer tools, they show tangible returns when implemented intentionally.

Areas where AI is yet to realize value are in the fully autonomous “AI Agents.” They require a level of implementation and oversight that companies are not yet ready to invest in or fully trust. AI agents running repetitive tasks with clear boundaries will likely be the first to take off, but fully roaming, decision-making agents remain over-hyped. Additionally, “GenAI everywhere” concepts are more of a marketing concept than a reality of technology implementation, while the employee may have a GPT at their fingertips, AI in everything will lead to competing GenAI tools and decision fatigue on which one to choose for which job.

Regulatory constraints mean that the pharmaceutical industry tends to be slow to adopt new technologies, but AI is impacting day-to-day business operations, where it can change how people research, write, analyze information, and make decisions, rather than in the regulated GxP environments. The pace of development is extremely fast, meaning employees who are not learning how to work with AI tools increasingly risk falling behind.

Cloud data platforms are increasingly common and powerful, and modern technology allows scalable platforms to be set up in weeks rather than months, where the limiting factors are only data governance and data quality, which must be owned by the business.

Equipment connectivity and overall equipment effectiveness (OEE) solutions allow the collection of production data and improve visibility of performance and the identification of optimization opportunities. However, the full potential of such systems often only comes after the data foundation has been established.

Finally, modern cybersecurity platforms have also delivered clear value. Their primary impact is risk reduction and improved visibility, helping organizations reach a state-of-the-art security posture.

Conversely, one area that is often oversold is low-code development. While it promises democratized application development, valuable and maintainable applications still require strong technical expertise. Without proper governance and engineering capabilities, low-code environments can quickly become difficult to manage, and in practice, the technology works best as an acceleration layer for experienced developers, not as a replacement for engineering teams.

The most tangible operational value in pharma over the past 12 to 18 months has come from strengthening core digital foundations as part of a broader digital journey. From an end-to-end delivery perspective, investments in new SaaS/Cloud-based systems like enterprise resource planning (ERP), quality management systems (QMS), MES, and laboratory information management systems (LIMS) are improving efficiency, enabling stronger customer collaboration, and ensuring operations remain compliant and quality-driven. The shift toward a harmonized digital strategy has been key in delivering a seamless customer experience across all regions for global organizations like Minaris (which is new and has gone through M&As). Smart integrated systems provide real-time visibility into manufacturing and testing activities, like batch status, quality events, and analytical results / certificates of analysis (COAs), helping teams make faster, more informed decisions. Together, this connected intelligent IT ecosystem of tomorrow is helping us deliver therapies to patients faster!

Although promising, AI is still evolving in its maturity for the regulated pharma industry. While it supports use cases like content generation, reporting, and summarization, broader applications that meaningfully impact manufacturing, testing, and quality workflows are still emerging. The industry is exploring how to apply AI in more scalable and compliant ways, and its true operational impact will depend on how effectively those use cases are embedded into everyday workflows.

Digital technology has become a major catalyst for operational value in the pharma supply chain. The technologies that have delivered the most value are the ones that are successfully integrated in everyday operations. Targeted automation and robotics have paid off in the right settings, where volumes are there and processes are mature. Alongside improved data, reporting has made a real difference by improving inventory accuracy, speeding up exception management, and giving teams clearer cost-to-serve insight. Those benefits are practical, measurable and scalable.

We’re seeing AI add exceptional operational value, especially in network studies and optimization work. Capabilities like AI-powered network design (digital twins), autonomous logistics, and smart routing are helping streamline deliveries and inventory decisions. When paired with solid data foundations (e.g., data lakes), AI can enable near real-time insights, better performance, and profitability, as well as potential sustainability gains through optimized resource use and less waste.

At the same time, automation isn’t a universal fit. In lower-volume or more variable operations, the upfront investment, integration lift, and added complexity can outweigh the benefits. What we’ve found is that the winners are the technologies that fit naturally into core warehouse management system (WMS) / transportation management system (TMS) workflows and support a stable operating model. The tools that promise to “transform everything' without that foundation often end up being the ones that are still over-hyped.

Over the past year, sponsors have been increasingly adopting risk-based quality management (RBQM) for their trials amid new regulatory frameworks, such as ICH (E6) R3, which emphasizes risk-proportionate approaches according to criticality of data and proactive quality by design. When the FDA recently shifted toward a single trial paradigm for drug approvals, this increased pressure on sponsors to strengthen evidence before regulatory submission. These changes are encouraging sponsors to rethink trial design, accelerating the deployment of data intelligence platforms that support integrated quality risk management (IQRM) approaches to RBQM and centralized monitoring, augmented by AI, to provide risk-based data review and quality management for oversight of the entire portfolio, mitigating risk across the trial life cycle. Tangible operational value has included measurable return on investment (ROI) through efficiency gains, quicker issue resolution, quality, and regulatory compliance.

Embedded AI capabilities add a layer of support by reducing manual labor, analyzing large data sets, and proactively flagging risks so that trial teams can take necessary action to resolve issues. It’s critical that AI be seamlessly integrated into existing workflows with specific use cases and demonstrable success, to ensure the technology doesn’t get over-hyped, and to drive breakthrough treatments, getting lifesaving drugs to patients safely, faster.

In the past 12–18 months, AI has begun to move from experimentation to real operational impact across clinical research. In ophthalmology in particular, AI-powered site selection and patient recruitment tools help teams analyze imaging data sets, identify eligible patients earlier, and make more data-driven decisions about where trials should run. By combining imaging analysis, real-world data, and historical site performance, these platforms are beginning to accelerate enrollment strategies and strengthen trial planning. More broadly, AI is enabling research teams to synthesize large data sets and translate operational data into faster, more informed decisions.

Decentralized trials and remote data monitoring continue to evolve as important digital capabilities. They hold real promise for improving patient access and trial flexibility, particularly through tools such as telemedicine visits, remote data capture, wearable digital technologies that enable real-time physiologic monitoring, and more continuous patient engagement outside the clinic. However, the greatest operational value has often come from hybrid models that integrate these technologies with traditional site-based care. Fully decentralized approaches can introduce complexity around procedures, data reliability, endpoint validation, and patient adherence, making thoughtful implementation essential to realizing their potential.

Over the last year or so, we’ve seen wearables and implantable sensors really prove their worth. These isn't just about cool gadgets; they provide a steady stream of real-time data that helps doctors make better calls and keeps patients healthier. Moving away from "once-in-a-while" checkups to continuous monitoring means teams can spot trends early, step in before things get serious, and manage their time much better. It’s been a total gamechanger for chronic care and remote clinical trials.

On the flip side, plenty of digital health tech is still mostly hype. Generic wellness apps and "AI diagnostics" that haven't been properly vetted tend to grab a lot of headlines, but they often don't do much to actually improve patient health or make a doctor's life easier.

The bottom line is that if a technology fits naturally into a treatment plan and gives doctors data they can actually trust, it’s a win. If it’s just a shiny new toy without the clinical proof to back it up, it’s probably going to underdeliver.

In today’s small molecule API landscape, time‑to‑clinic is critical. However, increasing molecular complexity, characterized by longer synthetic routes, intensifies pressure on costs, supply chain security, safety, and environmental performance. Digital technologies can deliver the greatest operational value when integrated into workflows, as they accelerate development, reduce experimental lead time, and de-risk scale-up in response to these pressures.

At Lonza, we’ve seen significant impact in reducing timelines from the integrated application of AI‑enabled route scouting, high‑throughput experimentation (HTE) coupled with ML and Bayesian optimization (BO). AI‑enabled route scouting broadens synthetic design spaces and ensures true industrial viability of the proposed innovative pathways early by identifying robust supply chain options and highlighting cost-driving steps. HTE then compresses development cycles through rapid exploration of reaction conditions, thereby validating and optimizing the routes faster than traditional sequential experimentation. These rapidly generated dense high-quality experimental data sets enable us to build a digital twin, a mathematical model which can be refined with only a handful of focused experiments, unlike the required full experimental matrices for traditional design of experiments (DoE). The digital twin provides rapid insights into process robustness and potential deviations, supporting confident decision-making.

By contrast, “AI‑for‑everything” platforms, immersive metaverse factory concepts, and fully autonomous plant visions are still in the early stages of practical chemistry integration. Therefore, the real value we see comes from focused, experimentally anchored, domain‑specific digital tools that are consistently accelerating drug development and manufacturing.

The most significant operational value from digital technologies has come from improvements in automation, monitoring, and control systems that enhance operational excellence, rather than from the newest AI trends. Organizations have seen real gains from technologies that strengthen data quality, automate workflows, improve process visibility, and enhance risk controls, particularly in regulated environments where reliability and auditability are essential. While large language models (LLMs) and agentic AI systems have generated lots of excitement, the practicality of their impact in highly regulated, deterministic applications remains limited today due to concerns around reproducibility, accuracy, explainability, and governance. In contrast, investments in better operational controls, system integration, and process automation continue to deliver measurable value by improving efficiency, reducing errors, and strengthening compliance.

Over the past 12–18 months, technologies that have delivered the most tangible value in development have been embedded directly into scientific workflows, rather than layered on top as generic tools. In particular, AI‑enabled formulation design, modeling, and decision support have moved from promise to practical impact when integrated with real‑world data and execution.

It's clear that AI creates value when it accelerates early, high‑risk decisions, like formulation selection. By reducing experimental burden, shortening timelines, and enabling faster iteration that is informed by human data, molecules can reach the clinic faster.

This is exemplified by the integration of Intrepid Labs’ AI‑driven formulation platforms into Quotient Sciences Translational Pharmaceutics^® platform. Machine‑learning predictions can deliver measurable gains in speed to clinic, API savings, and confidence in development decisions. We are excited for the future as we expand our work with Intrepid Labs.