How a leading pharmaceutical company uses GenAI to boost its marketing efficiency

May 13, 2024

sales, marketing and service
chemicals
Microsoft Azure
artificial intelligence

share on

One of delaware’s customers, a leading pharmaceutical company, is exploring how GenAI can support its regional marketing and sales teams. With help from delaware, a proof-of-concept was set up to train the company’s very own GPT-model for creating accurate and compliant marketing collateral.

Client: A global pharmaceutical company operating across more than 170 countries.
Challenge: Reducing the time marketeers must spend on creating accurate and compliant marketing materials.
Solution: A secure, proof-of-concept GPT-model running on Azure OpenAI and trained on the company’s vast repository of existing marketing materials, PIM, and DAM.

The challenge: Improving marketing efficiency

As a leading pharmaceutical company, the organization has to comply with a host of regulations, on multiple national, international and sector-specific levels. Its marketing team is no exception. When communicating about any of its products, the information needs to be accurate and up to date – especially with regards to ingredients and potential risks and side effects. However, as the company’s R&D teams are constantly working to improve products and formulations, that information can change quickly.

To create commercial presentations, folders, websites, etc., the sales and marketing teams extract information and details from the company’s continuously updated product information management (PIM) and digital asset management (DAM) systems. This information is then processed manually into marketing materials. The whole endeavor is very time-consuming and adds significantly to employees’ workloads. In many cases, this also means there isn’t a lot of time left for creativity or a human touch – which are the areas where sales and marketing teams can generate most value.

The solution: Automating with GenAI

Inspired by the rapid rise of GenAI technology, the corporation’s regional marketing team decided to embark on an experiment: what if it was possible to automate the creation of accurate and up-to-date marketing materials based on PIM and DAM data?

With expert guidance of delaware’s GenAI team, a proof-of-concept (PoC) was set up to create the company’s very own, private GPT-model. To emulate its specific tone of voice, the model was trained on the company’s enormous database of historical marketing collateral, e.g. webpages, brochures, and other documents. To ground the model and ensure factual accuracy, it is connected to the organization’s continuously updated PIM and DAM systems.

Of course, security plays an essential role, the pharma industry is no exception. Therefore, the GPT-model was set up on a highly secured Azure OpenAI environment, which is also where all data processing is taking place.

Balancing power and cost efficiency

In its PoC, the organization opted to implement OpenAI’s GPT-3.5 model, instead of the newer GPT-4.0 Turbo model. “A very deliberate choice,” explains Sven Arnauts, global AI lead at delaware.

“OpenAI’s most recent models cost 15 times more than older versions. And while they are generally more powerful, this doesn’t necessarily translate into better outcomes for all use cases. So when embarking on a project, it’s worth considering what you want to get out of it before deciding to go for the most recent and most powerful GPT-model. In this case, it was clear early on that GPT-3.5 would be the most viable and cost-efficient solution.”

Result: The advantage of failing forward

The GPT-model had to be able to generate accurate and up-to-date product copy, propose a structure for marketing documents, and even source lay-out and visuals from pre-existing templates to create a highly finalized proposal. Before publication, the company’s (human) experts check the proposal and validate it to ensure accuracy and compliance.

As a result, the workload of the sales and marketing teams has shifted from manually searching and processing data to validating proposals generated by the GPT-model. This means there’s more time for creativity and quality control, ultimately speeding up the go-to-market for certain products as well.

The project is limited to one region for now. But as its potential benefits for the larger group become more apparent, the company’s headquarters are paying increasingly close attention.