Plasma-based manufacturing is complex. Outcomes are influenced by dozens of variables, many of which hard to account for. Despite this resulting in significant challenges to achieving predictability and consistency, data driven process optimization presents significant opportunities to improve outcomes, such as delivering higher yields. In this case study, we examine the
Customer context: plasma-derived manufacturing at global scale
Grifols is a global healthcare company and a pioneer in the plasma industry, manufacturing plasma-derived medicines to treat chronic and rare conditions, as well as infectious diseases.
The Goal
- Improve yields and recoveries across product lines.
- Enable multivariate root cause analysis and efficient handling of time series profiles.
- Support real-time monitoring of equipment and process parameter trends.
- Move from simple, reactive evaluations to holistic, real-time analysis.
These areas of improvement reflect what senior decision-makers need: faster investigation and decision cycles, higher and more stable yields, and an approach that can scale across sites in a compliant manufacturing environment.
The Challenge
Process complexity
Plasma is a rare and heterogeneous starting material pooled from thousands of individual donations. Pools are fractionated and then purified to isolate therapeutic proteins, with tradeoffs and interdependencies across downstream products.
Manufacturing complexity
Plasma is collected from many donor centers with constraints. Fractionation and purification occur at multiple facilities, with intermediates transferring between sites and combining downstream—creating cross-site dependencies and complex batch genealogy.
Data complexity
Decision-critical data spans manual batch records, QC results, corporate systems (ERP), and industrial systems. Frequencies vary (weekly/daily ingestion, real-time feeds, and historical backfills), making it difficult to maintain a single, contextualized view of performance and drivers.
The approach: a pragmatic path to Intelligent GxP Manufacturing
Rather than a big-bang transformation, Grifols and Aizon followed a progressive “crawl–walk–run” approach.
Crawl
- Manually digitize data.
- Contextualize it in Aizon Unify.
- Build process understanding, then generate insights and monitor results.
Walk
- Ingest automation data into Unify for real-time visibility.
- Embed predictive analytics in Aizon Predict.
Run
- Digitize batch records with Aizon Execute to contextualize data around the batch and enable holistic, real-time predictive analytics (work in progress).
This maturity model was important for senior stakeholders: it supported early value capture, compounded the data foundation over time, and created a credible path from analysis to operationalization.
What we did: unify data, explain variability, operationalize AI
Unified and contextualized IT + OT data in a Data Fabric
With Aizon Unify, a GMP Data Fabric that unifies siloed IT and OT sources, Grifols was able to combine MBR relational data (for example, phase pH, reactor, and phase duration) with SCADA time series data, producing a batch-contextualized view that included derived phase metrics such as minimum temperature and average pressure.
This made complex manufacturing signals analysis-ready without losing the “batch story” needed for investigation and decision-making.
Built and industrialized AI models to uncover actionable CPPs
As data became unified and contextualized, AI models explained an increasing amount of variability—improving visibility into process dynamics and uncovering actionable CPPs related to yield and impurities. This work identified multiple key factors for subsequent evaluation.
Examples
AI-driven MVDA to optimize purification yield via coordinated pH adjustments
AI-driven MVDA was used to optimize yield in purification. Coordinated process adjustments can matter more than isolated changes.
Predictive monitoring to detect adverse trends early
By comparing predicted vs. actual yield results, we were able to identify issues earlier. Persistent gaps served as signals of adverse trends.
The Results: scalable value across sites and products
Over the course of this multi-year partnership, Grifols realized important gains such as:
- Significant yield increase across three sites and three products, achieved via 30+ process optimizations.
- Millions in cumulative added value thanks to said process optimizations.
- Enterprise-wide data integration and contextualization, plus a global suite of analytic dashboards with integrated AI for continuous process optimization.
Lessons Learned
The partnership showed that sustained, cross-site collaboration is what turns isolated wins into compounding gains. By grounding the work in a detailed process outline, the teams could consistently interpret the data in context and transfer knowledge more effectively across sites and functions.
Just as importantly, tight, transparent collaboration between subject matter experts and data scientists ensured insights translated into practical decisions. Automated data ingestion then shortened the loop—enabling rapid analysis cycles while supporting real-time tracking of equipment behavior—and, with unified data and cloud computing in place, the organization could maintain a holistic view of a complex, interdependent process end to end.
Multi-stage yield prediction then made it easier to spot outliers early and pinpoint the factors behind performance shifts, so optimization efforts could focus on the parameters that actually mattered. Critically, those adjustments were made with an explicit view of downstream consequences—balancing tradeoffs across steps and products—and cross-site visibility into upstream operations provided the context needed to interpret results correctly and make better decisions downstream.
