Tepsivo has built its company culture around optimizing pharmacovigilance processes with the aim to reduce workload and costs for clients and staff, allowing them to focus on critical tasks like risk assessment.

This blog post aims to highlight the significant changes in the automation area. It is the second entry in the “Tepsivo 5 Years” retrospective series; read the first post on digitalization here.

Automation has been around for centuries, with many now-common technologies once considered advanced automations. Automation is pervasive in daily life, from traffic lights and smart thermostats to common office tasks and remote work. 

Reflecting on predictions made in 2020, while some, like automated adverse event reporting via wearables, haven’t fully materialized, or even significantly taken steps forward, artificial intelligence has significantly simplified other specific tasks.

In this text, we explore the evolving impact of automation, particularly with the advent of AI tools, on both case processing and the broader scope of daily work.

Pharmacovigilance Automation at Tepsivo

From Tepsivo’s inception we have built our company culture around pharmacovigilance process optimization and automation. Our unique position and innovative approach, Tepsivo OnePV, to pharmacovigilance mean substantial workload reduction for our clients, our staff and partners.

As we combine our own software, streamlined processes and expert staff, we consequently bring down the cost of services and enable everyone involved in our processes to spend more time on the most important tasks, such as risk assessment, and a lot less or no time doing simple tasks like filling out and sending forms or tracking in multiple trackers and logging data in several locations as has been the process traditionally.

For Tepsivo to succeed, adopting automation on a daily basis has been absolutely crucial. We have been consistently sharing information and insights into how exactly we do this through our website, blog posts, and other channels, but we also invite anyone interested to contact us for a demo to see our systems and processes themselves.

During the first year of Tepsivo in 2020, I wrote two blog posts about automation. The first blog post was about the widely used everyday automation and the other about the commonly used automation to speed up routine work in business. Back then I wanted to show that automation is not just about safety reports coming automatically from emails to the safety database, but also about many other steps in the way we work in pharmacovigilance operations.

The main pharmacovigilance processes around the EU GVP modules, have not really changed in the past five years, but thanks to the rapid implementation of common technologies during the COVID times, and the external pressure of the AI powered technological leap of the past two years, there is a general sense of progress around the standard drug safety work as well. Cloud services and teleconferencing has reduced the actual physical paper and the need for face-to-face meetings.

And it is not just the rest of the world and the pharmacovigilance community: We can confidently say that automation at Tepsivo has also taken huge leaps since we started five years ago.

Automation is Everywhere

If you put on your automation-goggles, like I did when I wrote my first blog post about automation, and start thinking about automating things, there really are almost no parts of the regular office job that are not already somehow automated, intelligent, or somehow “smart”.

Just think about a standard day at the office: Keycard makes sure your credentials are checked when you enter the building. The elevator takes you to the correct floor. The coffee machine brews a cup of black magic potion for you. Computers replace all the office supplies you would ever need. Gmail suggests automatic email replies or even replies automatically if you don’t want to do that yourself. Google Meets creates a virtual meeting for you so that you do not have to walk to another room to meet with colleagues. Now most of these automations are enhanced with AI leaving even less of the thinking and doing to you.

In 2025 of course it is more common to work from home, or some other location that could be considered to be “remote”. It does not take a lot of imagination to start listing all the different technologies and automations that are involved in making it possible to work outside of the traditional office.

In Finland we used to use the abbreviation ATK instead of IT (information technology). In Finnish ATK stands for “automaattinen tietojenkäsittely” which in English translates to “automatic information processing”. So basically the meaning of ATK is that any of the work you are doing using a computer is automating information processing, work that used to be manual work. This is an interesting concept given how a large portion of the workday is spent in front of a computer.

Computers have been used to automate data processing now for about seven decades, but automation has been around for centuries and many of the automations have become such a norm that now they would be considered the manual and out-dated process, for example certain types of industrial machines that require human operators or writing text with a typewriter and making paper copies of the text. 

I am writing this blog post with my laptop using Google Docs. Just this process contains so many automated steps that I would probably lose count before I got to the point where the key strokes turn into ones and zeros in the operating system. 

If you really start thinking about it you will see automation everywhere. Traffic lights are automated systems that manage the flow of the traffic at intersections, smart thermostats keep the temperature at a certain level in the apartment and Netflix keeps billing your credit card automatically.

Right now when AI Agents are becoming the norm in everyday work, the work that is done with a computer is finally starting to feel like there is some “automated information processing” involved. But perhaps that will change in a few years as we get more used to the different AI tools.

World goes fast, PV needs to follow

When we talk about the use of AI Agents becoming the norm in everyday work, one key aspect to remember is that the healthcare industry moves slowly and cautiously a step or two behind the rest of the world.

When it comes to relative slowness of implementing innovative drug safety solutions, people often say that the issue lies somewhere in the risk averse nature of pharmacovigilance and pharmaceutical industry as a whole meaning that the pharmacovigilance roles are often performed by people who truly care about the patients and want to make sure nothing bad happens to those who use the products, and rightly so. 

However, I think there is a disconnect with real risk assessment and the traditional approach to drug safety processes taken by the pharmaceutical industry. For example, what is the amount of truly new safety signals coming from scientific literature that leads to validated safety signals? What is the risk of missing one or more of those signals in the literature monitoring? How should this risk affect the implementation of automated literature monitoring and safety assessment using AI? If this signal detection is not a familiar process to you, I recommend checking my earlier blog post about wasted efforts around the activity.

There are good reasons to be conservative when it comes to implementing new solutions to automate pharmacovigilance processes, but even the European Medicines Agency (EMA) has recognized the fact that there are challenges related to the lack of regulatory standards, guidance and validation for the use of patient-level healthcare data, AI and that the regulatory process is ill adapted to an increasingly dynamic environment in which technology and science, particularly in areas such as use of device data, real world data, adaptive algorithms etc., are developing faster than current regulations and guidelines.

What Shouldn’t We Automate?

Missing standards, guidelines and legislation is one thing, but I think it is clear to everyone that there are things that we should not use AI, or automation, for, whether it’s related to pharmacovigilance work or any other work.

In 2020 I wrote that probably it is better not to automate all parts of the workday, for example lunch and exercise, as there are clear benefits for both having lunch with colleagues and exercising before or after work. 

Five years ago we were living through a phase of lonely lunches and a boom in sports and today you can still see many trails of that era. Lunch at the office with colleagues seems like a distant memory at workplaces that do not have strict at-office rules. Now it’s a meme in Finland, that an employee needs a free lunch as bait from the employer to come to the office.

This should not be a surprise. There are things that you can do and enjoy at home that you cannot imagine doing at an office.

Thanks to the increased automation, but maybe even more thanks to remote working, there’s more time for sports, wellness and mental health. These are still not automated in the TV Shop way (Abtronic and the like) even if there are a lot of supportive apps in the space. And still, maybe that’s for the best. Let’s see what I think in another 5 years.

Coming back to the pharmacovigilance processes, it is currently clear that there are certain types of decisions that should be made by humans.

Today, I would personally be confident to let AI to decide what are the adverse event reporting rules (timelines, format of reporting, method of reporting etc.) for a specific country because if these are well defined, any of today’s AI tools would get the information approximately correct and the biggest risk would be that maybe something was reported within 90 days instead of annual report or the report would be initially sent via email instead of gateway.

But at the same time, I would want a human to be overseeing all the reporting and reviewing that the reports that day or week have been sent, that they have been sent to the right place and at the right time.

What’s next in PV Automation?

Five years is a long time but at the same time it can also be a short time. While automation has taken huge leaps thanks to AI tools, the predictions about the future I made five years ago are still mainly predictions, but on the other hand the new consumer AI tools have made automation of specific types of tasks much easier than I could have ever imagined in 2020.

Back in 2020 I predicted that automation would increase in pharmacovigilance space due to smart watches and other wearable devices that would be registered as medical devices collecting information about the user enabling automated reporting of potential adverse events in much larger numbers than what is the current situation with manual reporting. 

While wearables have become much more common during the past few years and they are used by many for different types of health tracking and monitoring, the mainstream devices have not broken through fully to the healthcare space and none of the mainstream devices are registered as medical devices for drug safety reporting purposes. Furthermore, there have not been significant steps towards automated reporting of adverse events or increasing the amount of safety data collected.

In broader perspective, the main improvements in pharmacovigilance automation during the past five years have been around general updates to the authority systems and recent AI features in vendor software. 

The regulatory authorities such as European Medicines Agency and US FDA have been improving the general system structures to reduce the manual work related to product information submissions and management of product information. This includes improvements at the US FDA around adoption of E2B R3 format in adverse event reporting and the updated submission portal. This type of updates actually reduce the need for automation, as they reduce the amount of manual duplication of efforts, that is common in both regulatory and pharmacovigilance processes.

In the first phase these improvements, at least at EMA, reduce the manual workload around marketing authorization applications and variations, but hopefully later these can be utilized to remove the manual steps around periodic reporting and signal detection. These tasks, which are basically data analysis performed regularly, should be able to benefit from the use of AI in data analysis and the use of centralized data management systems, where all relevant information about the products and the related safety reports are stored. 

So what will pharmacovigilance automation look like in 2030? I believe that we will use a lot more AI tools for everything in pharmacovigilance: generating PSMFs, preparing RMPs, PSURs and signal detection reports. At the same time, I’m afraid that the outdated systems, like EMA’s XEVMPD, and processes, like PSURs, will still be there. I also don’t think that the way we collect post-marketing adverse event reports from patients will change, even if the technology would be there to get a truly relevant amount of reports from them directly through smart devices.