Digitization and Process Analytical Technology (PAT) without a multimillion budget

Why most AI/AI initiatives do not reach pharma production

Executive Summary

Digitalization and Process Analytical Technology (PAT) offer you a clear competitive advantage by enabling precise control of production in real time. Instead of large batch sizes, they have flexible, small production slots that reduce costs and minimize risks such as quality problems, stock write-offs and complex production planning. Actively use existing machine data and quality controls for process optimization! With simple, customized tools and step-by-step implementations, you can reduce investments, GMP risks and implementation times. In this way, we increase quality, reduce costs and sustainably increase the effectiveness and efficiency of your production, without huge budgets and in small, clear sub-projects. No huge projects, but clear, rapid implementation with cost control!

Digitization and PAT here refer to technologies that can model processes. They are used to describe physical and chemical processes on a mathematical and statistical basis. LLM models such as ChatGPT etc. are not meant here.

The Digitization of pharmaceutical production and manufacturing is a decisive competitive advantage. Why?

Digitalization and PAT opened up the possibility of precisely monitoring and, even more importantly, controlling production during the manufacturing process! This opens up the possibility of flexible batch sizes, very fast product changes and continuous production.

The basic problem of our production in many companies is the optimization to maximum batch sizes. This is an attempt to keep administrative costs (purchasing of materials, internal processing of orders, release costs, etc.) as low as possible because these are all order-related. Fewer, but large orders minimize these costs.

This strategy has several danger points

• an increased risk if quality parameters are not in order: a very large product volume is immediately affected, which is then blocked and cannot be sold
• Production planning is complicated: if a batch is to run for five days, in many companies this production number is possible on very few weeks of the year. All weeks that include public holidays are automatically omitted, as the five days cannot be covered. This production must always start on a Monday and end on a Friday. If there are even small deviations from the start date, the entire production must be postponed. In reality, a buffer day is always created before this production - which was of course not included in the original calculations. Production during typical vacation periods (summer, winter, Easter, school vacations, flu season - there's not much left!) is also always associated with great risk or is completely impossible because the team is not complete.
• Large batches only work well if the sales planning or sales forecast is fantastically good! Is it always?

What do we mean by that?
The large production batch is usually packed directly into the respective packaging of the market/country. This works well for countries with high demand or high sales figures, but for countries with low call-off figures, these packs are then kept in stock. Relatively regularly, these stocks are written down considerably because sales are slower than planned or because regulatory changes to the packaging, package inserts or other small changes make the goods no longer saleable. In addition, there is of course the overstocking, i.e. if the goods fall below two years remaining shelf life, a discount usually has to be granted.

 

The large batch strategy is excellent if the sales figures can be planned very well and mainly large markets are supplied with a very short turnaround time: so almost never.

We see it as much more ideal to produce small production slots efficiently and thus be able to follow demand very closely. This minimizes possible write-offs of goods due to overstocking and numerous regulatory risks (e.g. packaging changes) or quality risks.

 

Discussing this idea is one thing, but driving it forward efficiently and cost-effectively is the real challenge.

If you read about digitalization and PAT, it is usually in connection with budgets worth millions and years of development time, followed by implementation problems. This is naturally a deterrent, as everyone with responsibility is under considerable cost and revenue pressure. As a result, such projects are usually not tackled at all.

We are fans of small and fast projects that are easy to keep within a manageable framework and are therefore easy to control.

What exactly do we mean by this?

 

Use of existing data

Instead of reinventing everything, we use the data that already exists! All systems and machines are already equipped with sensors. This data is usually only used for very simple control purposes, but very rarely to analyze the quality of the processes and to control not only the machines but also the process performance for product manufacture. What exactly can this be? Almost all plants have pumps or systems for API, excipients, solvents. Have you ever tried to compare your process fluctuation data with the process data of your machines and systems? In other words, how API content fluctuations in your tablets, capsules or vials relate to the delivery fluctuations of your pumps or other transport systems? It's worth taking a look! 

Are you really using your QC data?

The incoming inspections of the raw materials and the outgoing inspections of the manufactured products are rarely linked. Although they should actually be closely linked, right? Very specific measurements (content analysis, for example, or analysis of secondary constituents or impurities) are carried out in both incoming and outgoing inspections together with non-specific analyses such as color, pH value, bulk volume, etc.
All these analyses would be carried out day in, day out, but the data is only used for approval. It makes sense to actively link this data in order to improve process control. Quality trends of your suppliers become visible (here is a practical example: talcum powder according to the pharmacopoeia!) and often unforeseen production problems receive a solid explanation and can therefore be improved. You already have this data! It costs you nothing to generate it! You only need a few very simple tools and you can really make use of this data!

Simple Tools!

I have now mentioned the use of simple statistical tools several times. Of course, you can buy complex software packages with many updates and necessary training courses. You can certainly approach and carry out the analyses mentioned above. If you have specialized employees, of course. We do not advise you to buy complex tools for such analyses. They are very extensive, require intensive training and are therefore often not used or only used to a limited extent. We prefer to carry out such analyses with our customized software tools. This means you have no annual license fees and the applications do exactly what you need them to do. This is effective, efficient and easy on the budget.

Implementation in the company

We have very good experience with the gradual implementation of such measures in companies. So don't start a large program first, with complex planning, a large team and implementation in the relatively distant future. Instead, quickly implement very small sub-projects with a small team, test and check, adapt and continue to monitor. Only when it has been „running“ for a few months should it be extended to other areas or systems. This minimizes technical and GMP risks and stays within budget and on time.

The link

of new or existing sensors (IoT) is not rocket science! The first step is always to try! You usually learn a lot in the early stages and recognize where potential problems come from. We therefore proceed by linking and analyzing existing systems in an uncomplicated and inexpensive way. This can be a link via statistical analyses or by querying and analyzing sensor data directly via a network. It is very easy to determine whether the approach brings improvements or not. The data that has already been measured is analyzed together with the data from the machines and systems (practically all modern devices can be connected and queried). If statistical analyses reveal certain patterns that occur in good or bad processes (with higher variability), we can make targeted improvements. And the data links allow us to recognize whether we are going in the right direction or not. If this analysis and improvement works, you can introduce this system permanently. Sometimes it is the case that by linking the data and making a well-founded analysis, the reason for the problems is already found and the cause can already be tackled. This is the ideal case, because then you have raised the process to a better level. We therefore proceed step by step and do not try to establish a new „super process“. This facilitates internal coordination between different departments and also improves the persuasion work I have to do in any case.

What cannot be achieved with it

A production process that is set up incorrectly from the outset will not be stable even with digitalization. An example from practice: If the production speed of a tablet press is increased, this is only possible up to the limit set by the raw materials, pressing pressure and time. Beyond that, the quality depends heavily on the raw materials used (e.g. particle size distribution, exact composition, etc.). If you are already in the wrong range here, digitization and PAT will no longer save the process.

Conclusion

This article shows what is already realistically possible - the Implementation but requires experience in dealing with complex projects in a GMP environment.

PAT and digitalization bring considerable quality and cost benefits in production. With the introduction, you avoid data being measured but no longer being used, which represents a considerable waste.

Only by digitizing production and then linking all data to control the production process can you intervene very finely in process regulation. This also gives you the opportunity to speed up process changeovers when changing products and to cover a large part of the analytics through process measurements. This makes your production faster, better and cheaper!