We, Arvato SCS Healthcare and Arvato SCS Digital, have worked on a modular solution to automate email routing in our SAP CRM system. The project is called AI Email Routing and it is divided into 3 modules/phases:

1. Language Detection of incoming emails (e.g. from hospitals, pharmacies, wholesalers etc.)

2. Content Recognition of incoming emails (e.g. from hospitals, pharmacies, wholesalers etc.)

3. Automation of email processing (e.g. order entry, document provision etc.)

The project was piloted (Phase 1 and 2) for a global medtech company and one of Arvato SCS Healthccare's biggest customers (Cardinal Health).

While most countries where Arvato offers customer service on behalf of Cardinal Health use dedicated email addresses (.es, .fr etc.) that could be used for simple routing rules in SAP CRM, multi-language countries or regions such as Switzerland (German, French, Italian) and BeNeLux (French, Dutch) bring some challenges to these simple routing rules. Therefore, a more sophisticated, AI-backed solution with Azure cognitive services, was developed to automatically detect the language and, based on the language label, route the incoming email into the respective country/langue main queue within the CRM system where customer service agents with the respective language skills take care of the email (either answer it or re-route into further sub-queues). 

Phase/module 2 identifies the email content based on categories that were derived from the queue-structure in the CRM-system. Thus, it goes beyond Phase/module 1 as it identifies the content category which helps to further route into the respective sub-queue. Typical categories are manual orders, GHX-orders (can be processed automatically), returns, invoices, no_action and a few more categories, subject to the country-specific queue setup in SAP CRM.

While Phase/module 1 is only live for countries and regions with multi-language setup, i.e. Switzerland and BeNeLux, Phase/module 2 is live for more than 10 countries including e.g. UK, IE, GER, Austria Switzerland, BeNeLux, France, Spain, Portugal, Italy, Poland and the Nordics countries.

First qualitative feedback from agents and floor managers is positive and quantitative reports show good performance of the AI-models. Reduction of routing steps per mail and reduced throughput-times are also already measureable. Nevertheless, model retrainings might follow in a few months when more data has been collected. 

Next steps for Arvato SCS Healthcare is to roll this concept out to further clients in Harsewinkel. In parallel, a side project will be started to look into Phase 3 which is about automation possibilities for email processing (1st use case: change and/or creation of customer master data). 

For further information on AI Email Routing, please reach out to

Jens Wagner (Product Manager/Project Manager at Arvato SCS Healthcare) or

Regina Gerber (Product Owner at Arvato SCS Digital)