Automating service tickets in construction and installation: 10 minutes to 10 seconds

Why service tickets are complicated

A service ticket sounds simple: a technician fills in a form, someone in the office posts it to the right project. In practice, service tickets arrive in dozens of formats, from seven different mailboxes per region, often with missing or wrongly filled fields. A simple ticket costs ten minutes of admin: opening it, looking up the project, determining the cost type, entering hours in Bouwworks or AFAS, checking the subcontractor link.

At 3,500 service tickets a year, a realistic volume for a mid-sized installation company, that comes down to nearly 600 hours of admin work. Time that doesn't go into work preparation, planning, or customer contact.

What our AI coworker does

The AI coworker listens on the mailboxes you already have. For every incoming ticket the same thing happens:

• Classification: is this a service ticket, a quote, a complaint, or something else?
• Extraction: date, technician, GLN, project code or reference, hours, materials.
• Validation: is the project correct? Does Bouwworks know this subcontractor? Are the hours plausible?
• Posting: service ticket created in Bouwworks, linked to the project, with the right cost types.
• Confirmation: a reply email to the subcontractor with the project reference.

A concrete case

A Dutch construction company historically processed service tickets with multiple FTE, about ten minutes per ticket. After six weeks of implementation, the AI coworker was running in production. In the first 2.5 months the system processed 462 tickets with an exception rate of 1.92 percent and 4.36 percent human intervention. Average handling time per ticket: ten seconds.

Implementation cost: €15,000 one-off, €800 per month. Payback on this one process: under a year. The time freed up goes into planning own staff onto service tickets, a process now being automated as the next step.

What we've learned

Three lessons from implementations like this.

Start small. One mailbox, one ticket type, one ERP integration. Not seven mailboxes at once. The AI coworker learns the patterns in the first few weeks; that knowledge then scales easily to the rest.

Make the escalation explicit. Not 'a human should see this' as a vague outcome, but 'project unknown, suggestion: project X based on address'. The person doesn't have to investigate, just confirm.

Let the technician feel that it's working. A quick confirmation email with the project reference matters more than you'd think, it makes visible that the admin side of the work is now fast too.

Integrations

We connect natively with Bouwworks, AFAS Profit and Bizzstream for project admin. For companies using XML mail integration and GLN mapping for subcontractor data, we build that mapping into the design, no separate portal for everyone to maintain.