7 Ways to Predict Whether Your AI Initiative Will Fail

Are you confident that your initiative will succeed?

Most pharmaceutical companies we come in contact with are enthusiastically adopting AI for elements of their work – be it in discovery, R&D, clinical trials, regulatory, market access, or sales and marketing. Many pharma executives are aware that AI will disrupt pharma business but they are not sure how to ensure that their business is positively disrupted. In order to succeed, AI initiatives must be planned and executed effectively. Choosing a supplier is only the first step in achieving a successful scalable project. Assuming that a large company will achieve a good solution may not always be the case.

Most pilot projects run are designed to scale but the reality is that a recent survey showed only around 8% of AI projects implemented were scaled. Many pharmaceutical companies we come into contact with provide us with a project that has been started and failed – either from their internal team, or from an AI vendor or a consulting firm. When taking over projects that have previously failed, we first analyze why the previous attempt failed in order to ensure a success on our project.

It seems timely to write a quick guide to pharmaceutical executives planning, or implementing AI, to keep a few guidelines in mind to avoid having to redo the entire project, or having to abandon it part-way through. By keeping these things in mind, you can achieve goals and show the business value for your project.

1. Is there a clear vision and strategy for how AI will achieve specific business objectives in place prior to starting a project, and is this supported by the leadership team?

Most CEOs and C suites know that they need to be implementing AI to drive business value. The C suite task the strategy to various departments – be it IT, or Marketing or Innovation – but the resulting pilots fail to either succeed or scale, and much money gets wasted with one pilot after another without clear strategic objectives from top level management. Part of the challenge is that the C Suite themselves often do not understand what is possible and how to think about it. They need educating on the topic in order to be able to lead their companies successfully into the arena of AI. Many C suites think about AI like traditional business analytics but it is an entirely different ballgame.

Eularis have already implemented many trainings on understanding AI in Pharma so leadership teams can direct their teams to think about it appropriately. By leaving it to various departments, who themselves often do not fully understand the strategy side, one often ends up with numerous unsuccessful and un-transformative pilots that fail to achieve any real business objectives.

2. Do you have a clear strategic plan and roadmap behind the project prior to beginning?

We see so many AI projects that do not have a clear strategic purpose for the AI project itself nor clearly identified business value. Everything must start with strategy and the strategy must be planned by subject matter expert strategists who understand that area of the business– that is pharmaceutical strategists. Every project we start always begins with the strategic plan and roadmap. This includes identifying the strategic issue to be addressed, what big data is available to enable the project and the business value that will be gained from the project. AI projects are not plug and play and are very costly to implement. By simply asking AI companies to implement programs without first defining the right problem to solve and why often means the companies are not working on the right problem. Data scientists and big data engineers are expensive resources to throw at an ill-defined problem!

How do you know how to set a good strategic plan? Firstly, ensure everyone understands what is possible and the team identifies the right problem. Eularis have a process by which we take pharmaceutical teams to create an understanding of the situation, workshop the problems and identify the AI program options available and jointly decide on the optimal solution for that team. We then develop a strategic roadmap for the solution. This can be used to develop an in-house solution, or given to another vendor (or Eularis) to implement. A strategic plan must be in place or a lot of money can be wasted and the project doomed to fail.

3. Have you evaluated the business case and have a clear idea of the likely value and return that can be delivered in the first year?

So often no business case has been developed. We see many companies buying AI projects on the latest thing that sounds cool – be it a chatbot or Augmented Reality (which are both great but you need to think through your business case to see if they are the right tools for the job). This reminds me of a project we did 20 years ago for top 10 pharma with a blockbuster product. They had bought a very cool web automation tool and then came to us and said ‘Can you think of a strategy of how we should use it?’. We did, but it was like buying a jet and then saying ‘where can I fly?’. If the only place you want to fly to is close, a Cessna will do. By understanding your business case you can identify the best tools for the project rather than going in reverse and choosing the project because of the tool. When you have a hammer in your hand, everything looks like a nail! Although AI is costly, if you think through the business case appropriately, and plan the roadmap accordingly, the cost is well worth the investment from the significant gains made.

4. Do you fully understand what is involved in an AI project? AI is not plug and play. AI is time consuming, and costly if you want transformative business results. You may not be able to afford it.

AI projects start with strategy and then quickly go into data. Big data is the life blood of AI projects. Many companies expect AI from small data – a lot of small data – but small data nonetheless. Small data algorithms can be written with AI but this is not where AI really comes into its’ own. There are many phases of an AI project involving various teams, typically coordinated by (in our case) a scrum master who turns the chaos into order. The teams involved include the strategy team who are subject matter experts, the data science team who are AI experts, the data architects who set up the data flow across technology, big data engineering team who create the optimal tech stack for the project, the full stack programming team who get the data flowing through the tech stack and algorithms correctly. The user interface design team to get the visualizations created how they will be used by the end user. The front end programmers who program the visualizations and, if you are using language data rather than just numbers, the computational linguistics team.

You can get platforms that are plug and play but in all cases I have witnessed, if you just load data into these, without the entire data sorting, structuring, tagging, cleaning etc process, it is fairly limited in terms of the value that comes out. If you do the process properly, you will achieve strong business value from a properly planned and executed AI project.

5. Do you think all data needs cleaning? 

Data wrangling – i.e. data sorting, structuring, tagging, cleaning, feature creation, feature selection etc is a time consuming and costly business. Much data which is available is not relevant to many of the big data problems to be solved. There is usually no need to collect everything from everywhere in order to solve a business issue. Data tracking is a critical part of any big data project. A study by McKinsey showed 70% of the data cleaning being done was a waste of time. We recommend you are strategic. Start with strategy and then you will know what data is relevant and should be cleaned so that this effort can be much more streamlined.

6. Do you have a business translator?

No, not a language translator, a business translator. We learned early on that despite great communication skills of our data scientists, you need someone with a subject matter expert and strategy background, as well as an analytical mind who can understand what the client wants, take what the strategy team need, and translate that into what the data scientists need to know, and then translate the data scientists findings out of algorithms back into ‘business speak’. This is the type of person you need on the team in order to extract maximum value from the strategy and the data science.

7. Is your analytics team integrated into the business rather than isolated?

In so many pharma there is a structure in which analytics is a separate role from the business teams. The analytics team must be integrated into the business units to fully understand the business issues and be able to effectively analyze what is needed when it is needed. Some companies have done this well already, while others still have them siloed away. If your data scientist feels that their work has no impact and everyone still does things the way they always did, then that is a sign that the teams are not integrated well enough.