AI answers your customer questions, your team handles the rest
AI customer service platform for an installation company. AI reads incoming questions, answers immediately with a confidence score and escalates to a team member when needed. The team only handles the exceptions.

Laravel, React, TypeScript
Web applicatie
The challenge
Installation companies get the same questions every day: when is the engineer coming, what does error code F28 mean, how do I shut off the main water valve, can I schedule a maintenance visit. Those questions come in by email and forms. Every question requires a manual answer from a staff member, even when the answer is standard.
Vermeer wanted to handle that stream of customer questions without hiring extra staff. Customers expect a quick answer. The team didn't have the capacity.
Approach
We built an AI customer service platform that automatically reads, understands and answers incoming questions. The AI knows the company, the services, the service areas and the frequently asked questions. For every answer, the system shows a confidence score, so staff can see at a glance how certain the AI is.
Questions the AI can't answer with sufficient certainty are automatically escalated to a staff member. The team only handles the exceptions.
How it works
- Inbox with AI status - All conversations in one overview. Per conversation, the team sees whether the AI has answered, whether a staff member is needed and how high the confidence score is.
- Confidence scores - Every AI answer gets a score. At 94% on an F28 fault, nobody needs to do anything. For an invoice question the AI doesn't understand, it goes to administration.
- Knowledge base - The AI learns from a central knowledge base of frequently asked questions, procedures and company information. The team adds to it themselves. Every new answer makes the AI smarter.
- SLA monitoring - Conversations that stay open too long are automatically flagged. No customer falls through the cracks anymore.
- Assignment - Conversations are assigned to staff members. Unassigned questions are immediately visible in the inbox.
- Statistics - Insight into how many questions the AI handles, the average response time and which topics come up most often.
The AI in practice
A customer emails: "My boiler shows fault F28." The AI recognizes the problem, explains what F28 means and gives concrete steps. Confidence: 94%. The staff member doesn't have to do anything.
Another customer asks: "When is the engineer coming?" The AI checks the appointment details and answers with the time window. Confidence: 91%. Customer helped, within seconds.
An invoice question about an amount that doesn't add up? The AI escalates that to administration. Not every question belongs with the AI. The system makes that distinction itself.
Result
The majority of customer questions are answered automatically. Customers get a relevant answer within seconds instead of waiting hours. The team spends its time on complex questions and urgent situations, like a gas leak or a broken boiler in the middle of winter. The knowledge base grows along with it: the longer it runs, the less the team is needed.



.png&w=1920&q=75)











