Things to consider before starting an AI project

Introduction

AI has many use cases in pharma, from value pricing to faster reimbursement to identifying key opinion leaders, to understanding the customer journey to creating automated omnichannel marketing and more. However, 84% of companies implementing an AI solution have yet to take it out of the pilot or proof of concept (PoC) phase. 

Firms make several recurring mistakes that keep them stuck in PoC. Avoiding these and getting started with realistic expectations and effective practices is the only way to ensure your pharma company is part of the 16% that successfully implement a working AI solution, not the 84% stuck in perpetual development.

Before You Begin

Identify the best problems to solve

AI is a tool that we are using to solve business problems. So before starting AI, the first thing to do is to look at the business challenges that you have that are causing the biggest headaches for your team. Forget AI. Focus on finding problems that you cannot solve. These are typically the things that an AI solution is ideally suited for.  And I would take that one step further and ask businesses to first define a customer problem. From here, it becomes much easier to clearly define the problem.

Identify the optimal solution/s

Just as the problem statement should be customer-centric, so too should your solutions. Think carefully about each in light of the end customer and which serves them best. Take your best ideas and reimagine them from a customer perspective. Does your path from problem statement to problem solution provide the most benefit to the customer? If not, how can it be improved to do so? Does it also benefit the business? The perfect solution should do both.

Identify and Sort the Relevant Data

AI relies on data. Big data. You first need to clearly identify what data is relevant to the problem and solution at hand. Where is that data? Who owns it? Legally what can you do with it? Create your data inventory and start looking at what data you have and what data is missing, and where can you get the missing data.

One thing I have learned is that no data set is perfect and the dirtier and the more problematic the data, the more costly a project can become as it goes along. It is always good to get a sample (it doesn’t have to be a full data set – just a few hundred rows of each is sufficient) of every data set you will be using before you start and get the data scientists to evaluate the robustness of the data for the business questions you are trying to answer.

Plan Plan Plan

You may have a large ambitious project that will solve so many of your challenges. That is great! But make sure you plan it in a series of use cases that solve the problems one piece at a time, but that can then be linked together in your big ambitious project. This way you get quick wins along the way as you are building.

Secure Buy-In from Key Stakeholders

Depending on the size of the project, you may need significant organizational buy-in to support the implementation. If you are the team lead, you will need to explain the problem, the solution and why you are going down the AI route, what the benefits will be, and why it is legally and regulatory compliant. This last one is certainly one step that can take the longest in the project set-up – at least in our pharmaceutical setting for AI projects. 

Set the expectation that this will be a long-term project and even though there may be high costs (depending on what you are doing), and potentially mediocre short-term results, AI systems learn and get better and better over time so look at what kinds of long-term benefits can be expected as well. You can give examples from other companies if needs be.

For example, if you were planning on using a digital human for your call centre, one like the one who has taken over 50% of the Bank of America’s customer service jobs. The cost to set it up and train it on all your products and processes may have some higher upfront costs but over time the costs will be clear as in the case of Bank of America who reduced their human staff by 50%. This translates into massive savings year upon year as well as – interestingly – higher customer satisfaction as the digital human remembers every conversation so unlike a human who can only check the notes on the call, the digital human may remember that the person mentioned that it was their grandson’s birthday, and they were in a hurry and the digital human will remember that and ask about how it went for example.

Moreover, while firms are slow to adopt AI in many areas now, it will slowly become the norm as companies who are not implementing it will find it difficult to compete. AI sounds flashy and exciting, but in truth, it can sometimes be a long game with slow returns that snowball over time.

Think of AI not as an immediate game-changer, but as a potentially expensive but necessary long-term infrastructure investment that everyone will have to make to remain competitive in their industry over the next decade.  But do not just do it for that reason. You need to know what business benefits  you are aiming to solve. As long as the business cases are planned correctly and the right data is used, and the project planned well, you are in good shape.

Secure your budget

Once you have got buy-in, examine what the costs are likely to be and identify what the budget is likely to be, then add a contingency. Always add a contingency. Depending on the project, often there are challenges found along the way that were not foreseen and a contingency in your budget can be useful when that happens. In making your case for the budget, look at the business value you will get in 12 and 24 and 36 months time. Do include your total budget including internal headcount, contract resources, and IT infrastructure (including application and cloud licenses) to calculate your total investment on an AI project. Do not forget the cost of data acquisition (sourced internally or externally) as part of the total budget. This can, and often is, a costly element as well.

Create a Dedicated Team

You will need a dedicated team with proven experience taking AI to the operationalization stage, which means experience bringing the AI out of prototyping in the PoC stage into real-world use. Unless your company has implemented AI in the past, your existing IT team is probably not suited to this task, nor do they have the time to dedicate to such a significant project. If you do not have the expertise in house, you will need to bring in a team who have done this before (either consultants, contractors or new hires).

Proof of Concept Implementation

Once you have an AI team, a timeframe, and a budget, it is time to enter the Proof of Concept (PoC) stage. In this stage, the AI system will train in an entirely virtual space, using real historical data but not engaging in any real-world business activities. It will remain in this stage until you and the team are confident to deploy it into the real world, where its decisions can directly affect your business.

Clean Up Your Training Data

About 80% of the time spent in the PoC stage will not involve the AI itself. Instead, your AI team will be working with the training data to clean, optimize, and make it accessible to the AI platform. You can make this step go by a lot faster by ensuring beforehand that there is enough data available to train your AI effectively. You should also make sure that this data is stored in a format that the AI tool can easily access. Speak to your team about these concerns at the start of the project to avoid costly delays down the line.

Leverage Open Source Technology

You should make certain that your AI team is aware of the appropriate open source libraries available for their specific type of AI and that they have access to machine learning automation tools. Automation tools and libraries will make the AI training process go faster, ultimately shortening time spent in the PoC stage.

Operationalization

The next step after PoC is operationalization, which is the live deployment of AI into an environment with real business consequences.  If you followed our advice from the previous section and chose to work with a team that knows how to take AI beyond the PoC stage, you should be well equipped to transition to this stage.

Once you are satisfied with your AI’s performance in the operationalization stage, leverage your AI team to deploy additional AI projects that build on your wins from your first project.