GSK IS A GLOBAL BIOPHARMA COMPANY with a purpose to unite science, technology and talent to get ahead of diseases. GSK aims to positively impact the health of 2.5 billion people over the next 10 years. Its bold ambitions for patients are reflected in new commitments to growth and a step-change in performance.
GSK prioritises innovation in vaccines and specialty medicines, maximising the increasing opportunities to prevent and treat disease. At the heart of this is its R&D focus on the science of the immune system, human genetics and advanced technologies, and its world-leading capabilities in vaccines and medicines development. It focuses on four therapeutic areas: infectious diseases, HIV, oncology, and immunology.
THE FEED TEAM’S ROLE
GSK’s Global Capital Projects organisation looks after the worldwide company capital investment in the pharma supply chain (new manufacturing equipment and new facilities). Within this organisation the Front-End Engineering & Design (FEED) team focuses on the initial stages of design and business case. Projects usually go through three key phases: business analysis, feasibility and concept selection. In each phase we assess the strategic business requirements and propose a selection of engineering solutions to address the business need. By the end of the concept selection process a single solution gets endorsed by the business and progresses into the delivery phase.
Modelling and analytics activities often play a big role in the front-end phases of capital projects, as they help the business to understand what information is critical. Knowing how big the demand for a certain product is going to be affects what size the facility should be, the kind of equipment that should be used for the production and what the production cycle is going to be like. This requires lots of different scenarios being modelled in a highly sophisticated network of dependencies and calculation of their impact on the investments the company must make. The budgets of the potential capital projects can range from £10m to £500m and the timeline for the plans needs to be made for the next 5 to 15 years. The FEED team uses modelling and simulation to understand how to optimise such investments by phasing them while still minimising the risks.
THE CHALLENGE
In 2020, the GSK FEED team was faced with planning for a new biopharma facility in Italy. To plan for such a large investment, the team had to carry out key work to support the business case, which involved developing the design concept and establishing the optimal level of resources to satisfy ten years of forecasted commercial demand. This meant that the team needed to calculate how many machines were needed and what the technology of the machines should be, which machines should be bought, and decide on their operational regime, because the type of the machine will dictate how they can be used operationally.
Product forecasting and demand fluctuations have always represented a big challenge for pharma companies, especially during the design of a new manufacturing line or plant. Most of the time the potential new products are still in development undergoing clinical trials and years ahead of the commercialisation launch. Getting the right product mix, coordination and optimisation of resources such as machines, operators, materials, etc. are some of the challenges that GSK face operationally. To address this, there was a need for a model that could be used to inform key aspects of the business case. It had to capture all the required complexity while being fast enough and flexible enough to answer the business question.
A SOLUTION THROUGH COLLABORATION
GSK asked Decision Lab to develop a decision-making tool to support the business case. Decision Lab is an award-winning tech company with a lot of experience in building highly complex models that aid decision-making in a range of sectors, including healthcare, aviation, infrastructure, logistics, and defence and security. GSK and Decision Lab have worked on projects previously and knew that they could work across the two organisations in one unified effort, which would be necessary for this complicated project.
Decision Lab’s solution was to develop a simulation model for the planned new production line. If able to accurately represent the operations of the facility, this model would allow GSK to test the future production line virtually and investigate its boundaries by evaluating several “what if” scenarios. It would provide an understanding of the factors that impact production capacity and operating costs. It would also give specific answers to key questions: how many machines they should buy, what technology these machines will use, and what their operational regime (number of operators, shift patterns, etc.) should be?
The FEED Team used the model to generate hundreds of plausible scenarios and test many different demand forecasts, equipment technologies and size, quantity of key equipment and allow them to understand the true operational bottleneck
Decision Lab and GSK technical experts worked closely on this project, in a highly collaborative way. It was a cross-functional effort involving about thirty GSK contributors across several countries and disciplines. We needed to involve different specialist teams to guide us on aspects of the data input, and this expertise was embedded into the model. The fact that the model condensed the knowledge and the requirements of about 20 active contributors from different disciplines says a lot about the value the development team gained in terms of engagement, knowledge transfer and decision support. The collaborative approach helped the design and production teams gain a holistic view of the capacity and operation as we uncovered the process logic and operational metrics. It also offered a high level of engagement within GSK, aided knowledge transfer and built confidence in the model and the insights it provides.
The model itself is a discrete event simulation where the user specifies the demand as a sequence of product batches and specifies a series of process steps that each batch may need to go through. For each process step the batch is assigned equipment that will perform the process. Equipment may have to be cleaned between batches. The model tries to process all the batches as quickly as possible and that the machines are utilised as equally as possible. shows the 2D view of the model, providing information on each machine’s current batch, activity and utilisation, as well as worker movement. The model is flexible and allows the user to control many different variables such as maximum length of queues, frequency and length of equipment failures, the number of available resources etc. The model produces a comprehensive set of outputs which the user can use to infer where the processing bottleneck is and if the number of machines/resources they selected is optimal. shows the 3D view of the model, providing batches on the conveyor belt as they go through different machines and workers moving around the environment. “The model has been fundamental in defining the right sizing of the facility as well as exploring the different product mix taking into account several variables (batch size, product type, shutdowns, change overs) by running multiple scenarios and sensitivities analyses. It also offers scheduling optimisation opportunities that can’t be achieved via traditional approach, leading to a more efficient use of the facility“: Site Strategy Lead (Italy).
FIGURE 1 2D VIEW OF THE “ITALIAN” MODEL
FIGURE 2 3D VIEW OF THE “ITALIAN” MODEL
The FEED Team used the model to generate hundreds of plausible scenarios and test many different demand forecasts, equipment technologies and size, quantity of key equipment and allow them to understand the true operational bottleneck. The model has been used in different stages of demand and this continues.
A SECOND APPLICATION: ORAL SOLID DOSE (OSD)
GSK has a well-established facility in the south of England and was considering whether to invest in updating to increase capacity at the site. The biopharma filling facility model had proved its utility for informing a business case, so GSK asked Decision Lab whether the model could be adapted. Decision Lab had intentionally developed the biopharma filling facility model in a flexible way, so that it could be applied more generally. This, in accordance with the FEED philosophy of reusing parts of design work on multiple projects (to accelerate the studies), meant that GSK could engage Decision Lab to adapt the model.
For the facility in England, we needed to carry out a study into how much new equipment was needed for the required increased capacity. This would involve running scenarios where we would consider different numbers of equipment and its performance to determine what the minimum number was to deliver the desired capacity. It's important to capture the key factors that influence the KPIs, including shift patterns, product mix, machine failures, etc. They will contribute to a more realistic solution – closing the simulation to real-world gap – and will enable the model to adapt to any future changes in business processes (which won’t require an entirely new model). In the facility, the product is moved between the machines in containers, so we had to be able to calculate their number, also the space for these containers, and how much resources are needed to run the facility. Another question of interest was about budgetary efficiencies, e.g. if we don’t buy a certain machine, how much capacity can we deliver? Basically, we want the model to allow us to identify the breaking points in the system.
The model provided a great starting point for us in terms of rapidly developing a model from a proven capability. We could add the required features and functionality to a working model that would allow us to run the range of scenarios that were needed. It also gave us greater confidence in the output. When you’re unsure about a model’s reliability, you might add a buffer to account for errors – for example buying extra machines which may or may not be needed – but of course using this safety buffer in the plan costs money. So, minimising this buffer by having a more accurate prediction can create a huge saving. shows the main runtime visualisation of this model, providing information about each machine, which batch to which they are currently assigned, their utilisation and queue size.
FIGURE 3 THE MAIN RUNTIME VISUALISATION OF THE “SOUTHERN ENGLAND” MODEL
We also had the benefit of a pre-existing Excel-based model for the site in England. We realised that this could not be used for the study because it could not handle the complexity involved with multiple SKUs, multiple ways to process the product between different machines, interdependencies between parameters, etc. However, it did provide useful insight into the current facility and processes, and a baseline for some model intercomparisons.
With the increasing pressures, our timelines often become shorter. In the past we could have had six months to run a study but now we only have two months. Therefore, time becomes a critical factor that often drives decisions about the tools that we put into action. Having a baseline model we can quickly adapt was crucial to this project, and we were well placed that we had developed the biopharma filling facility tool as a generic model that we can use for different projects and requirements.
The model for the OSD facility allows us to see how the change of different inputs influences how long it takes to process all the demand and how efficient the equipment and resource usage is. For example, with a particular configuration, what percentage of the demand for this drug are we going to meet or even if we have some spare capacity. Our focus was very much on capacity and the equipment required to deliver it, as this has major cost implications. However, in the future we can expand it to consider aspects such as figuring out the best space layout that would allow for the best operations.
The model for the OSD facility allows us to see how the change of different inputs influences how long it takes to process all the demand and how efficient the equipment and resource usage is
There are also other ways in which we can develop it, and these reflect the stages of the site itself. To build the facility, we need to plan it first. This was why we developed the model to enable us to conduct the study to work out the design that allows the site to meet the demand as cost effectively as possible. While the site is being built, the model can be further adapted so that it can be used to refine the details of the plan or address detailed issues that come up, and this could be especially important if unexpected changes are required or there are demand fluctuations that could impact the solution.
Once the site is built, the model, with minor enhancements, can become an operational support tool. It could be used to help make decisions by providing modelling results to complex questions for specific situations that arise, and some may even require optimisation solutions. Some examples are optimising batch production, predicting capacity over the next six months.
WHAT THE FUTURE MIGHT HOLD
Decision Lab and the FEED team have developed a modelling capability to support the business case for large scale investment in pharma production facilities, one in the facility in Italy, and one in the UK. As part of this process, GSK has assessed the top benefits of the modelling activity and identified that it:
Results in a faster approval timeline – the trust in the modelling outputs has smoothed the stage gate approval of the concept and allowed faster moves into the next project phase.
Provided depth and breadth in its analysis, with its ability to evaluate over 200 different business scenarios in a short amount of time.
Provided a high level of confidence in the amount of equipment required for the new production line that ultimately led to capital expenditure reduction of 20%.
The collaboration made a material difference to the business and both the Italian facility and UK facility went to the construction phase.
Our tool has the capability to deliver continued value to the project so that in the future it will be possible to further evaluate capacity requirements as function of the demand forecast variability. The model also has the capability to be extended to people and material flow for the final operational layout.
Together, the GSK FEED project team and Decision Lab have laid the foundation for the future. Over the years we developed a new methodology that now can be applied and scaled to most of the pharma processes and investments. This will deliver speed and efficiency in the FEED and a greater confidence for the decision makers in the business. The next challenge will be to further develop such models so that they can be used by the operational team once the facility has been commissioned and handed over. That could lead to even greater benefit to the business, reducing cost of goods, inventory, and many other KPIs.
Over the years we developed a new methodology that now can be applied and scaled to most of the pharma processes and investments
Additional information
Notes on contributors
Giovanni Giorgio
Giovanni Giorgio is a Senior Digital Engineer at GSK. Giovanni is a Chemical Engineer with 15+ years of experience in the pharmaceutical industry (R&D and Global Manufacturing), with a strong background in API chemical & process engineering and with an intensive and diversified experience acquired in different business units. Giovanni is currently leading the modelling and analytic strategy in the Front-End Engineering and Design team in Global Capital Projects. He has developed an interest in applying advanced modelling techniques to solve complex business problems.
Natasha Zheltovskaya
Natasha Zheltovskaya is Decision Lab’s head of marketing. She has worked in marketing for more than 15 years, leading the brand development work for some of the world top brands in retail, fast moving consumer goods (FMCG) and technology.
Peter Riley
Peter Riley is a Principal Consultant at Decision Lab where he is leading the simulation team. Peter has built agent-based, discrete event and system dynamics simulations for a wide range of industries, including pharmaceutical, retail and defence.
Jacob Whyte
Jacob Whyte is a Simulation Consultant at Decision Lab. His background is in computer science. Jacob develops digital twins and simulation models for clients and is working on a major simulation model for creating synthetic data for AI model training.