DEV Community

Hieu Luong
Hieu Luong

Posted on Originally published at himitek.com

Case Study: How a Parts Factory Saved VND 420 Million per Year with an AI Agent

#himitek #technology #saas

1. Pain – Operational data was scattered everywhere

Minh, the production manager of a 150-person mechanical parts factory in Binh Duong, used to begin every shift by opening Excel files, searching chat groups, and asking team leaders about paper forms. Product defects were recorded in one place, machine downtime in another, while maintenance history often lived only in the memory of a senior technician.

The factory did not lack data. It lacked connected data. When the same defect appeared for the third time, no one received an early warning. The manager had to run the operation manually, while technicians relied on personal memory and experience to decide what to do next.

2. Agitate – Downtime became a hidden operating bill

Every machine stoppage costs more than lost output. Delayed orders, overtime, additional scrap, and diverted technicians all reduce margin. Two hours spent compiling reports per shift can become dozens of labor hours each week while producing a false KPI: the report is complete, but nobody can prove whether the root cause was fixed.

Minh had patched the process with multiple Excel files and considered a large project without a clear ROI baseline. That is project theater and half-hearted optimization: more budget is spent while manual data entry remains the bottleneck. If the factory keeps taking short-term shortcuts, technical debt grows, repeated defects create avoidable losses, and operations may effectively go down when one key employee leaves.

3. Solve – A three-step AI Agent rollout in 90 days

Step 1: Standardize the input. HimiTek connects internal incident forms to an AI Agent. Each record includes machine ID, production shift, defect type, downtime, and supporting images. The Agent classifies the issue, checks production history, detects recurrence patterns, and assigns the task to the right team.

Step 2: Control actions and spending. The Agent runs through OpenClaw Gatekeeper with 9router v0.4.66 and LiteLLM dual-instance failover. Rate limiting, automatic API key rotation, and a hard budget cap such as 5 USD per month per virtual key prevent runaway loops from inflating costs. Dangerous commands remain locked by default and require a whitelist or explicit permission.

def create_task(issue):
    if issue['severity'] == 'critical':
        return {'team': 'maintenance', 'approval': 'required'}
    return {'team': 'quality', 'approval': 'standard'}

# Gatekeeper checks policy before any command is sent
Enter fullscreen mode Exit fullscreen mode

Step 3: Measure ROI before expanding. During the 90-day pilot, the system reminds owners about deadlines, prepares shift reports, and sends management summaries. Role-based access protects operational data. The factory then tracks three business metrics: response time, repeat-defect rate, and scrap cost.

  • Incident reporting time fell from about 2 hours to 15 minutes per shift.

  • Defect response time decreased by 35%.

  • Repeat defects decreased by 28%.

  • Lower scrap and downtime produced estimated savings of VND 420 million per year.

4. CTA – Start with one process that can prove its value

The factory did not need to replace its entire ERP or purchase a heavy platform. It started with incident recording and maintenance, ran a 30–90 day pilot, and measured the result in money. HimiTek can help identify the operational bottleneck, design a controlled AI Agent, and build an ROI baseline before any wider rollout.

The target outcome is practical: fewer manual reporting hours, fewer repeated defects, less downtime, and VND 420 million retained each year instead of being lost to fragmented processes.

Top comments (0)