Heraeus: Improving the Sales Lead Pipeline With LLMs

When it comes to implementing Generative AI, Heraeus’s goals were twofold from the beginning: 

1. Provide secure Generative AI tools to a larger audience within their operating companies.
2. Challenge said companies to identify and develop specific use cases that generate value for the business. 

We sat down with Richard Kroegel, Senior Data Architect for the Heraeus Digital Hub, and Jannik Beers, Commercial Excellence Manager, to zoom in on one use case in particular — the Generative AI-driven lead list generation tool. We discussed how the company is using this momentum to experiment with new technology in a governed environment and identify and develop more use cases within Heraeus.

The Use Case: Sales Lead Identification With LLMs 

Heraeus has many business activities and their target customers correspond to high quality specialty segments, so it’s traditionally been hard to identify potential customers. Before Dataiku, their process for identifying sales leads was very informal and very manual (i.e., searching on Google, brainstorming, etc., which took a lot of time), so the goal was to automate this process via LLMs, using specific selection criteria. 

The team asked themselves, “Can we use Generative AI’s creativity to generate new sales leads?” The answer turned out to be yes. Heraeus partnered with Dataiku to develop a prototype use case which identifies sales leads and fact-checks them thanks to external knowledge. This was another reason the use case was chosen — the data requirements didn’t rely on internal data or connecting internal data with external data. It only relied on external data and knowledge. As a result, the team would be able to create a list of leads repeatedly and efficiently for a target market in order to determine, for example, where to sell their products next. 

So, how does the use case actually work?

1. Heraeus sales employees input selection criteria (e.g., businesses that are active in a specific therapeutic end-market as well as located in a defined region) and any optional open-ended questions (e.g., What are the main products developed by this company?). Then, an LLM is used to generate sales leads (name and homepage URL) corresponding to user-provided selection criteria.
2. The LLM suggests sales leads that are then searched via an automated check implemented in the project, using the Crunchbase API.
3. In order to validate the output and protect against any LLM hallucinations, the selection criteria for the sales leads identified on Crunchbase is automatically double checked via Crunchbase, internet search results, and the company’s website. 
4. Complementary questions are answered based on internet search results and the company’s website. 

Impact to the Business: 60%-70% Time Saved & Improved Lead Quality

Given that the former sales lead identification process was immensely manual and time-consuming, the estimated time saved with the new process leveraging LLMs in Dataiku is 60%-70%. Heraeus has been able to identify targets that they would not have been able to without the tool, demonstrating the efficiency of the use case and quality of the results. 

Dataiku features like automation and the possibility to quickly deploy interfaces for end users with Dataiku applications and webapps enabled Heraeus to get this use case up and running very quickly — from initial discussion to implementation in five weeks — in order to focus on the use case and LLM without having to worry about the automation. The team was also very focused on Responsible AI, mitigating the risk of erroneous or obsolete information and always keeping a human in the loop for validation. 

Timing was fast and pivotal here. Dataiku made it possible in such a short timeframe and, from a technology capability point of view, we would not have been able to do it without them.

-Jannik Beers, Commercial Excellence Manager, Heraeus 

Further, different operating companies within Heraeus work completely independently. Each company has its own sales department, its own business development group, and therefore different markets and requirements. With LLMs, there wasn’t a need to define a rigid structure, but rather there was a certain level of flexibility with regard to the input, and the team then used the interpretation capabilities of the LLM. So far, the feedback from the end users of the use case (sales and business development) has been extremely positive. 

Working with Dataiku, there was always an open-minded atmosphere and the colleagues demonstrated that they are real experts in this space. Plus, the speed of getting the solution built speaks for itself.

-Jannik Beers, Commercial Excellence Manager, Heraeus 

AI Democratization in Practice

In the beginning of 2023, the Commercial Excellence team mapped certain use cases to optimize sales and marketing topics with AI. This team, on the business side of Heraeus that supports the operating companies and acts more like a think tank to leverage optimization potentials and deploy best practices, began testing this idea of a sales lead identification use case on their own with ChatGPT (outside of Dataiku). They started experimenting autonomously with some prompts and were able to preliminarily assess the feasibility of the use case before involving data scientists.

They soon realized that with the limited capabilities on the tech side, they wouldn’t get the results they desired (hallucination was the biggest issue with plain ChatGPT). That’s when the Commercial Excellence team approached the Digital Hub to get them involved and the Digital Hub team, in turn, looked to Dataiku to automate and operationalize the use case. With traditional machine learning use cases, this type of democratization and collaboration is more complex because teams are confronted with advanced models and scientific concepts that are harder to explain. Here, LLMs act as a very powerful catalyst for AI democratization. They are very versatile and can be used across a variety of parts of a use case, enabling the team to accelerate their work. 

Looking to the Future

In the months to come, Heraeus hopes to widen the scope and extend the power of this use case to identify stakeholders to target for business development teams, craft the first outreach messages, and so on. For Heraeus, this use case opens the door to a completely new world of opportunities with regard to faster, more efficient customer communication. They are keen to experiment with AI use cases to improve the efficiency of sales and marketing processes as well as increase the customer support quality. They are open-minded about the future of Generative AI, adhering to the goals of saving time to be more efficient or increasing the quality of information to make more qualified decisions.