Business

Building the smartest pharma company in the world

We are on the verge of widespread adoption of artificial intelligence (AI) in business as well as society at large. A growing number of people are publishing creative and cutting edge solutions and state-of-the-art models are freely available as open source software. Regardless of the field, everyone seems to be “using AI” or “developing algorithms” to boost their business or to make the world a better place. Then why is it that, besides some of the silicon valley giants, so few companies are really changing their business with AI?

The answer is that transforming into a truly AI-capable organization, and fully leveraging AI and data driven practices, takes more than a proof-of-concept. Depending on your organization’s level of maturity in operating with data and AI, and what you want to achieve, it might even take a complete revamping of your practices, processes and culture – just like digitalization once did.

The Finnish pharmaceutical company Orion Pharma, with help from Reaktor AI, has been on a journey to introduce AI into company operations across the board, to fully utilize all available data, and to build the capability to quickly adapt in an ever- changing world. In other words, their goal is to become the smartest pharma company in the world.

In this post, we discuss what it takes to transform into an “AI company”, and what we have learned along the way.

The problem with the problem

AI design is challenging. For starters, most of the time you won’t know if something is going to work until you build it. Second, when designing AI solutions, we’re often starting from a problem about the problem, not from a solution to a problem. In the words of the end user, instead of “we have this problem, can you fix it with software”, the starting point is something like “our work is hard, but we don’t really know what could make it easier”. It often takes the data scientist a considerable effort to understand the domain well enough to realize what the actual challenges are and what could be done to tackle those. And even then, it’s a long way to go from the latest state-of-the-art algorithm to a value-creating solution running in an industry production environment.

If designing AI is challenging, designing pharmaceuticals is on another level. The drug discovery process operates at the cutting edge of multiple areas of science, including medicine, biology, and chemistry. The pharmaceutical industry is also one of the most regulated fields in the world. There are feedback loops that span years or even decades. The amount of scientific information to deal with is immense, mostly unstructured, often hidden in masses of text data, scattered all over the digital realm, and the complexity of the domain of human physiology is unparalleled. On top of that, as a pharmaceutical company, you also have manufacturing, marketing, and regulatory operations to add to the mix. The field is dynamic, as seen in the current shift in products and business model changing from blockbusters towards more personalised solutions.

How to pull off an AI transformation

To even start to tackle such an enormous challenge as bringing AI into such a large organization, on such a challenging field, a few things are essential (and as a matter of fact, many of them apply to any industry).

1) Support from the top is key

One crucial element is to have forward thinking leadership – in other words, executives that passionately strive to stay at the forefront of development and identify the emerging and disruptive technologies early on. The kind of executives that are curious and willing to educate themselves on the technology’s possibilities, potential and shortcomings.

Support from the leadership is important for any meaningful change to happen. In transforming to an AI capable company, it is crucial. For many, AI is still a relatively new thing where there is lots of hype and little concrete knowledge. This can lead to faulty expectations like: “can’t we just have AI design a perfect drug”. On the other hand, this can also lead to scepticism, fear and apprehension: “will AI take my job?”, “my work is too complex, AI will never be of any use in this”, etc. That’s why it’s of the utmost importance that the leadership makes the transformation a priority, emphasises the importance to the whole organization, educates and spreads knowledge and creates the conditions for success.

2) Focus on specific solutions and fostering a collaborative culture

Success in AI transformation does not mean one AI solution that somehow mysteriously solves all problems and turns debt to profit. Success is built on two things: specific solutions and a shift in culture.

The actual solutions consist of many small components. Common to all of these components is that they develop at interfaces of all relevant expertise. Whether you have a dedicated and multidisciplinary team, a part-time working group or a loose collection of people, the important thing is that all necessary knowledge is working together. The AI experts need to learn vast amounts of information about the domain and have the right subject matter knowledge to get an understanding of the challenges. The subject matter experts need to learn what AI can and can not do and what the available data sources are. These interfaces can be found in various ways and anyone with an idea or interest can initiate one. This is the shift in culture towards collaboration, empowerment and initiative. The leadership’s responsibility is to support these connections and interfaces, spread the knowledge and make the AI expertise available to anyone in the organization to connect with.

3) Make sure it’s an iterative process

Another important element is to ensure the whole process is iterative. The development of a single AI solution can start in many ways. Sometimes data scientists start with their “best educated guess” and build something to show and create an interface to assess the solutions value and develop it further. Other times it might be a subject matter expert coming to the data scientists and presenting an idea or posing a question that they together try to create a solution for. It is important to understand that it’s all about constant improvement.

The solutions are continuously developed in an iterative manner. When the problem domain is complex, the only way to know how something will perform and the impact it has, is to use it, learn and make it better, continuously. Regardless of how the solutions came to be, the ownership of the tools, and the direction they are being developed should end up with the end users/business unit to provide them with ever better tools for the job at hand.

4) Design processes and solutions together

When designing the AI solutions, you can either plan to do the same thing but with fewer resources (automation), or do something in a new way – maybe even something that has never been done before. Therefore it’s not enough to develop solutions in isolation, separated from the process.

The processes themselves and the ways of working must also be evaluated together, at the interfaces where development happens. Only working within the existing processes is not truly transforming. The AI solutions and business processes should be developed together and continuously improved through experimentation and adaption.

5) Start small and keep building

If all this sounds like a big thing, it’s because it is. Transformation takes commitment and honest self evaluation. What it should not be is a big upfront investment. It’s not necessary to start by building a massive distributed data lake and locking down the company data strategy with all the bells and whistles on the market. It’s not about purchasing the latest turnkey solution, platform or framework.

It’s much better to use your different projects, the “move fast and break things” parts of agile development, to prototype and test the most important systems first and use that knowledge to align, for example, the data management strategy for the whole organization. Likewise, it’s not necessary to build a state-of-the-art cloud infrastructure right off the bat. Let the first solutions teach you how things should be, and let the spirit of “move slow and fix things” guide you to adapt, adjust and maintain.

Orion’s continuing transformation

Today, Orion is still a rare company as it has not only accepted the challenge to really transform into a pharma company of the AI age, but also taken the plunge and put the transformation in motion on a technological as well as a cultural level across the organization.

Following the principles outlined above, in the past year Orion and Reaktor AI have built and launched machine learning based tools to support Orion R&D and the regulatory processes, built a scalable cloud infrastructure to support current and future work from experiments and proof-of-concepts all the way to production, evaluated products, tested and experimented with ideas and approaches, taken part to the Elements of AI challenge with great success and much more. And we continue to forge ahead on all fronts to make sure Orion is the smartest pharma company in the world.

Never miss a post