Meet Heidi: Heidelberg Materials’ new AI voice agent supporting order management

• In a pilot project in Australia, Heidelberg Materials is testing Heidi, an artificial intelligence (AI) voice agent that assists in managing order requests. Heidi communicates in natural language, works with SAP in the background, and hands calls over to customer service agents when needed. 

We spoke with Nimit Kamdar, Global Program Lead for AI in Customer Service, about why the company introduced Heidi, how the pilot works, and what the teams learned so far.

Why did you start exploring AI for customer service?

We think that great customer service today is not about isolated actions or channels. It’s about bringing together everything customers expect: a seamlessly connected service that is fast, consistent, and personalised. AI excels in delivering across these dimensions. The AI voice agent pilot was a key step in our omnichannel strategy. Together with other channels, it gives customers flexibility in how they interact with us, and ensures every touchpoint feels connected and intuitive. 

Our customer service setup varies between countries, but customer service agents mainly place and manage orders for customers – for example, change quantities, change dates, or cancel orders. During busy hours, customers might wait in the queue, and we want to avoid that.  

However, AI is not here to replace people. Customer service agents are efficient already, but their time is limited. When AI handles routine tasks, customer service agents have more mental bandwidth to enhance customer relationships and solve complex cases that require human empathy. They are also less exposed to fatigue caused by continuous and repetitive phone calls and data entries.

Which part of the customer journey does Heidi support?

The focus is on order management. Customers can choose to wait for a customer service agents or immediately speak with Heidi by pressing 1️⃣. Heidi can retrieve the customer’s orders, identify the project and delivery details, and execute changes in SAP.  

If Heidi cannot complete the task or the customer wants to speak to a customer service agent, the call is handed over seamlessly with a summary of the conversation, so the customer does not have to start the conversation from scratch. This eliminates one of the most common sources of frustration in such processes: being passed from an AI to a customer service agent and having to repeat the same information all over again.

What was customer feedback like?

The pilot started with around 35 customers. Now, about 500 callers from 250 customers have access to Heidi. Since this type of service is new in our industry, customers were curious how it would sound and behave.

They appreciate that the voice is natural and dialogues are flexible, unlike traditional Interactive Voice Response (IVR) systems. Also, it helps them skip the queue during peak times, with quick and smooth handovers to humans, when needed.

Today, the reliability of intent recognition and order identification is above 90% and increases as we refine the system. Not all customers want to speak with AI, but many appreciate not having to wait.

What is the technical setup behind Heidi?

Heidi is technically orchestrated across multiple systems which handle telephony, conversation logic, and business processes. Within the conversation flow, we leverage a mix of generative AI and deterministic logic. The AI can speak naturally, but the order-management steps follow a controlled sequence to avoid mistakes.

Who was involved and how was the project built?

Australia was chosen for the pilot due to its highly structured ready-mixed concrete processes, strong local accountability, and committed executive sponsorship. This is essential: without organised backend processes, even the best AI cannot perform well. We worked in a collaborative cross-functional set up, with a lean core team of experts, strongly supported by internal digital teams, project leads from our provider and our Australian business. Our secret: we genuinely care for customer experience and want to make Heidi a success. 

You also work on AI for email processing. What does that look like?

Processing customer emails is time consuming, especially when the information given is incomplete. AI now reads these emails, extracts the fields needed to create an order, and creates the order in SAP. For incomplete emails, AI asks the customer to clarify missing information. This significantly reduces manual workload. We also ensure that a human is always involved for oversight.

Could Heidi become a chatbot?

Yes. Since all logic is text-based, converting the voice agent into a chat agent channel would require almost no redevelopment. 

What are the next steps for Heidi?

We are strengthening the solution in Australia by adding targeted features that enhance the customer experience and automate additional steps, preparing Heidi for thoughtful global scaling grounded in customer needs and business value.