Turning pilots into production: How to create, scale, and succeed with GenAI

Data analysis

Pharmaceutical companies are eagerly pursuing GenAI products and use cases to generate value within their organisations. Many are heavily investing in consultants to help develop strategies and transition from pilots to full production.

A recent report indicates that US consulting firms will earn over $392 billion this year due to the surge in GenAI work.1 Despite lacking a clear-cut strategy, many companies recognise the potential of GenAI, though 90% of pilots may not reach production.2

This is where the right partner can add significant value and impact. Leveraging a partner with domain expertise in pharma and experience in AI and GenAI is essential for transitioning from data to actionable insights.

The pilot-to-production pathway

A typical pilot-to-production pathway in life sciences includes exploration, experimentation, industrialisation/scaling, adoption, and monetisation. There are seven critical components for scaling and maximising adoption and monetisation:

1. Alignment with strategic objectives and governance

Let’s start with the basics: How does GenAI fit into your company's goals? What’s your big picture for GenAI? To figure this out, get your C-level executives talking with your data science leaders. Those conversations are key to setting the stage for success.

Now, governance is just as important. Big pharma companies often have dozens or even hundreds of pilots going on at once. How do you pick the winners? You need a clear set of criteria agreed upon by all key players.

Having a structured stage gate process helps ensure you’re not just throwing darts at a board. Start from brainstorming, then narrow it down to ideas that are worth solving with GenAI. Assess if they’re technically feasible and check how they meet customer needs.

Also, make it a practice to check in monthly with business unit leaders. These meetings should lead to clear decisions about your pilots: Do you keep going, expand, pivot, or stop?

2. Data strategy

Next, let’s talk about data strategy. You need to think about every step – from acquiring data to making it useful. This includes integrating and curating data, turning it into insights and recommendations, creating data products, and ensuring it’s accessible and usable.

Traditionally, data stores were all about structured data. However, with GenAI, you must also handle unstructured data and metadata. Think about those valuable nuggets of information like doctors’ notes or social media interactions: they can be a goldmine when turned into “generative AI-ready datasets” (GRDs) that train your large language models (LLMs).

3. Operating model

Now, let’s dive into how your operating model should look to leverage GenAI successfully. First off, you need to secure executive sponsorship to ensure your initiative aligns with the company’s strategic goals. Creating a dedicated GenAI task force with cross-functional experts is crucial. This team will oversee the integration and application of GenAI across various departments.

Clearly define what’s handled in-house and what’s outsourced. This helps streamline operations. Also, foster a culture of continuous learning. Upskilling your team ensures they’re ready to generate and utilise GenAI insights effectively. Establish governance mechanisms, like cross-functional councils, to ensure coordination and oversight. This helps align with regulatory standards and best practices.

4. Skills and experience

To fully leverage GenAI, focus on developing key skills like prompt engineering, GenAI development, and LLMOps (LLM Operations).

Prompt engineering increases the accuracy and usability of GenAI responses, a vital skill for data scientists and developers. GenAI development involves selecting the right LLM and integrating it into business workflows. And LLMOps is crucial for managing the GenAI lifecycle, including CI/CD, performance monitoring, and scalability.

Additionally, front-end design skills are essential for creating user-friendly interfaces that drive engagement and adoption. Invest in training across these areas to harness GenAI’s potential, foster innovation, and gain a competitive edge.

5. Technology

Scaling GenAI for pharma companies involves key technology components tailored to enhance commercial and clinical operations. Most pharma firms leverage GPT models like GPT-3.5 and GPT-4 to develop GenAI applications that boost the productivity of medical representatives and reduce marketing spend.

Effective prompt engineering is crucial. It ensures these models give accurate responses tailored to specific needs. Retrieval-augmented generation (RAG) is another key component. It combines LLMs with a knowledge database to ground responses in factual data.

Evaluating LLM output is essential to drive accuracy, which in turn engenders adoption and digital trust. Researchers and cloud and tool providers such as Microsoft and Databricks have proposed a wide range of metrics such as BLEU for comparing generated output to reference text, perplexity for evaluating internal consistency, and BERTScore for assessing semantic similarity to human-like responses.3 Assess these metrics and evaluate which are most suited for your specific use cases Operationalising these models, or LLMOps, involves CI/CD pipelines for seamless updates, robust monitoring tools, and scalable infrastructure managed via cloud platforms.

6. Adoption

To drive adoption of GenAI applications, focus on clear communication, user engagement, and showing value. Establish a strategic roadmap and get executive sponsorship to ensure commitment at all levels.

Identifying and empowering champions from the target user base is also crucial. These champions advocate for the technology, facilitate feedback, and ensure development teams collaborate with users, rather than imposing changes.

Provide comprehensive training and ongoing education on GenAI’s benefits and functionalities. This will build user confidence and competence. Showcase early pilot projects that deliver tangible improvements to generate momentum and encourage broader adoption.

Fostering a supportive environment and engaging users throughout the process will accelerate the acceptance and integration of GenAI applications.

7. Responsible AI

Using GenAI tools responsibly is critical, especially in pharma, where data security, data privacy, and regulatory compliance are paramount. Mistakes can damage reputations and lead to legal issues. Documenting oversight, frequently verifying models, and monitoring security and compliance are essential.

Transparency is key. Models must be traceable and explainable. Ethical data usage without biases or intellectual property violations is critical to maintaining trust and integrity.

Success drives scalability

Following these steps lays the groundwork for scalability. Scaling at an enterprise level requires making the right set of strategic data and technology decisions for your organisation and executing with speed and simplicity with appropriate levels of governance. Choosing the right data and technology partner for this is critical.


Driving success with GenAI in the pharmaceutical industry hinges on strategic alignment, robust governance, a comprehensive data strategy, and the right choice and application of technology.

Align GenAI initiatives with your company's goals and establish strong governance to prioritise and evaluate pilots. Develop a detailed data strategy that includes handling unstructured data and creating generative AI-ready datasets. Build a meticulous operating model supported by executive sponsorship, continuous learning, and cross-functional collaboration. Choose the right set of models, patterns (such as RAG), effective prompt engineering, and ensure measurement and monitoring of LLM output accuracy, using LLMOps to deploy applications in a seamless and reliable manner.

By focusing on these key areas, pharma companies can transform pilots into scalable, impactful GenAI solutions.


  1. Mickle T. The A.I. boom has an unlikely early winner: wonky consultants. New York Times. June 26, 2024. Accessed June 26, 2024. https://www.nytimes.com/2024/06/26/technology/ai-consultants.html
  2. https://www.forbes.com/sites/peterbendorsamuel/2024/01/08/reasons-why-generative-ai-pilots-fail-to-move-into-production/
  3. Evaluation metrics | Microsoft Learn
Sudip Chakraborty
profile mask
Sudip Chakraborty