AI Adoption Series
A Vixxo Senior Customer Service Representative transformed a time-intensive repair review into a five-minute AI-assisted decision, without losing the human oversight needed for objective, well-supported outcomes.
By the Vixxo Team | Featuring Akram Saleh, Sr. Customer Service Representative, Coffee Equipment
Key Takeaways
- A Vixxo team member built an internal artificial intelligence (AI) agent to determine whether a repeat coffee equipment repair call is a "re-dispatch," where a client shouldn't be billed twice for one unresolved issue.
- The agent cut investigation time from roughly three hours to about five minutes by cross-referencing technician notes, equipment manuals, and 30 days of service history.
- Over six months of testing, the agent’s determinations consistently aligned with expert review, demonstrating a high level of accuracy.
- Because the tool relies on asset-tagged equipment data, Vixxo sees a path to applying it across other trades outside of coffee equipment
The Problem: WHEN EVERY RE-DISPATCH REQUIRED HOURS OF REVIEW
In facilities management, a "re-dispatch" happens when a technician is sent back because the original issue was never actually fixed. When Vixxo confirms a re-dispatch, the client is billed only for parts, and the service provider (SP) can invoice only for parts, since the work wasn't completed correctly the first time.
For a large, multi-site coffee equipment client with standardized machines across its footprint, that call happens constantly. Reps on Vixxo's coffee team, who aren't trained technicians themselves, manually reviewed technician notes, pulled 30 days of history per asset, and consulted technical support to decide whether a callback was legitimate. Each case could take up to three hours, and coordinators often handled 20 - 30 cases a day.
The Build: An Agent That Reads Manuals So Coordinators Don't Have To
Akram Saleh, a Sr. Customer Service Rep supporting Vixxo's coffee equipment accounts, proposed a fix: build an AI agent trained on equipment manuals, parts lists, part costs, and repair procedures, then have it do the cross-referencing automatically.
Given a new work order, the agent pulls the asset ID, reviews the prior 30 days of service calls, and tags every related visit and note. It produces two outputs: a short summary for quick review, and a detailed report citing the specific manual pages, part numbers, and repair steps relevant to the case. If a required procedure, like a pressure test, was never logged, the agent flags the gap. It then drafts a message to the service provider laying out the visit history and the discrepancy in plain language, reframing findings as "here's what we're seeing" rather than "here's what you did wrong."
| Step | Before the AI Agent | After the AI Agent |
|---|---|---|
| Investigation time | Up to 3 hours per case | About 5 minutes per case |
| Research method | Manual note review, back-and-forth with tech support | Automated 30-day history and manual cross-reference |
| Documentation | Informal, inconsistent | Standardized report citing manual pages and part numbers |
| SP communication | Ad hoc, sometimes contentious | Consistent, evidence-based, coaching-oriented |
What the Agent Delivers: Example output
Here’s a condensed, anonymized look at one real determination involving an automated bean-to-cup coffee brewer. The equipment had three “excessive noise” and inconsistent-grounds calls involving the same subsystem within 26 days. Each visit resulted in cleaning and calibration, with no parts replaced.
Sample Agent Output: Quick Summary
Determination: Redispatch (parts only, no travel/labor)
Pattern: Same equipment and subsystem, three calls in 26 days, each a calibration or re-seat with no parts. The issue kept returning within days of each visit.
Root cause: A worn or improperly seated mechanical component that calibration alone cannot fix.
Recommendation: Process as a redispatch and scope the next visit for a component repair instead of another calibration.
Sample Agent Output: Technician Brief
"This asset (ID #123456) has had three right-tower grinder noise calls in 26 days — 04/17 and 05/08 were
clean-and-calibrate of the right mill, and 05/14 found the grinder auger not fully engaged and re-seated it. The loud/metal-on-metal grinding and coarse grounds keep coming back, so please do not re-calibrate again. Recommendation: replacing the right grinder hardware failed component. Run a sustained set of brews to confirm the noise is gone and the grounds hold to spec, and get a named store-lead sign-off before close."
Accountability, Not Blame
One design choice stands out: findings are attributed to the service provider organization, not an individual technician, keeping the feedback loop collaborative rather than punitive.
"Without our vendors, we can't do work, and without us, our vendors don't have work. At the end of the day, it's a partnership. It's a 50/50 for everything."
Akram Saleh, Sr. Customer Service Representative, Vixxo Facility Solutions
Why It Started With Coffee, and Where It Could Go Next
The agent works well in coffee equipment because service requests are tied to a specific asset ID, letting it build a precise, unit-level history instead of guessing which machine at a location is involved.
Vixxo already manages more than 2.2 million assets and supports over 20,000 coffee equipment installations, with roughly 38,900 annual repairs and 5,600 preventive maintenance visits across its portfolio, giving this tool room to scale well beyond one account. Saleh sees the same logic applying anywhere equipment is asset-tagged, including HVAC, where most sites run a single system.
The tool is still early, released within the last couple of months. Saleh's benchmark for success is three to six months of real-world use, and Vixxo plans to revisit this story with outcome data once that window closes.
Curious how AI-assisted decision support could work across your facilities program?
Vixxo's teams work from a growing set of AI-enabled tools built to reduce friction, not just cut costs.
Talk to VixxoFrequently Asked Questions
What is a re-dispatch in facilities management?
A re-dispatch occurs when a technician returns to a site because the original repair issue was not fully resolved. In these cases, the client is billed only for parts, and the service provider can invoice only for parts, not travel or labor, since the first visit did not complete the job.
How does Vixxo's AI agent decide whether a callback is a valid re-dispatch?
The agent reviews the equipment's asset ID and the prior 30 days of service history, cross-referencing technician notes against equipment manuals, required repair procedures, and part records. It produces a summary and a detailed report citing the specific manual steps and any documentation gaps.
Could this kind of AI tool work outside of coffee equipment?
Yes. The approach depends on asset-tagged equipment data rather than the trade itself, so it applies well to categories like HVAC, where each site typically has one system, and Vixxo expects to extend the model to additional trades and accounts over time.
Does the AI agent replace human judgment in these disputes?
No. The agent accelerates research and documentation, but coordinators and technical support still review findings, and communication with service providers is framed as coaching rather than an automated verdict.
Vixxo asset and service volume figures reflect internal company data. Client and service provider identities in this article have been anonymized to protect confidentiality.
