Introduction
AB Werkt: a multilingual AI helpdesk on Salesforce Service Cloud
Sector: staffing and employment services
Solution: Salesforce Service Cloud, Agentforce, Einstein Case Classification, Omni-Channel
In brief: TopDesk replaced, WhatsApp intake in 16 languages, the whole organization on one platform, an Agentforce agent that handles questions autonomously.
About AB Werkt
AB Werkt places and seconds workers throughout the Netherlands. The helpdesk processes a high volume of questions: pay, sick leave, contracts, facilities, HR and logistics. The flex workers come from all over Europe and speak more than 16 languages. That made the helpdesk not only busy, but multilingual too.
The challenge
The helpdesk ran on TopDesk. That system was separate from the channel employees prefer to use (WhatsApp), had no AI support, and no connection to the rest of the data within AB Werkt (the MSR platform with Mysolution). In practice, that meant:
- Agents copied emails manually from Outlook into the CRM.
- Cases were distributed inconsistently, without fixed routing logic.
- There was no digital self-service channel for employees.
- Escalations to the back office ran through separate email chains outside the system.
- Teams worked in different tools, and nobody was looking at the same case.
AB Werkt wanted one platform that could handle multilingual intake, route cases intelligently, and give every department (helpdesk, HR, payroll, logistics) visibility into the same case.
The approach
Cloudmuscle replaced TopDesk entirely with Salesforce Service Cloud. The whole organization now works from one platform. Cases come in through three channels and are automatically classified and routed.
WhatsApp intake via Agentforce. On AB Werkt’s business WhatsApp number, an Agentforce AI agent handles all incoming messages. The agent holds a conversation in the employee’s own language, collects the required information, and automatically creates a case in the helpdesk queue. Attachments the employee sends along are linked directly to the case. Agents reply to the employee straight from the case record, with automatic translation into the right language via an Einstein prompt.
Email-to-Case. Emails to the helpdesk address are automatically converted into cases. The manual step of cutting and pasting from Outlook is gone. Email cases land directly in the helpdesk queue and move on to Omni-Channel.
Einstein Case Classification. Every case is analyzed for category, subcategory and language. That prediction determines the priority in Omni-Channel and the language of the outgoing messages. The agent checks the prediction when picking up the case, so the model stays accurate. To train the model, we migrated historical TopDesk cases.
Omni-Channel routing. Cases run through the helpdesk queue to the next available agent, with a maximum of five active cases per person. Priorities (P0 through P3) are assigned automatically based on the AI prediction, with manual escalations always coming out on top.
Custom object for second-line escalation. A custom object, “Interne Vraag”, handles escalation to the back office within Salesforce. The separate email chains between departments are a thing of the past, and the full case context travels along.
What it delivers
For the agents: cases come in classified and prioritized in a single queue, and WhatsApp replies go out directly from the case.
For the employees: 24/7 contact via WhatsApp in their own language, an automatic confirmation of their request and clear status updates. Support without a language barrier.
For the back office: escalations through Interne Vraag instead of email chains, with the full context included.
For management: a complete audit trail on every case (status, owner, category, subcategory, priority) and insight into the quality of the AI predictions.
The result
- TopDesk fully replaced. The whole organization works from Salesforce.
- Manual copying of emails into the CRM has been eliminated.
- WhatsApp is live as a new contact channel, AI-driven, in 16 languages.
- Case classification went from manual and inconsistent to AI-predicted and verified.
- Second-line escalation runs in a structured way within the platform.
