Case Study: How a Logistics Company Cut Delivery Reconciliation Time by 38% with an AI Agent

1. Pain: 8,000 shipments and a pile of disconnected documents

Nam, the owner of a logistics company in Long An, handles more than 8,000 deliveries every month. At month-end, his operations team must open delivery records, inspect proof-of-delivery photos, verify surcharges, and compare everything with rate cards and transport invoices.

The process is still running on manual labor. Staff spend five to seven days searching for missing signatures, missing photos, incorrect order codes, or wrongly entered surcharges. When documents are scattered across email, Zalo, Google Sheets, and the operations system, delivery reconciliation automation is held back by technical debt and layers of patchwork.

2. Agitate: Slow reconciliation means revenue stuck in the queue

For Nam, each incorrect record is more than a cell that needs editing. It may be a surcharge that never reaches an invoice, a shipment that must be reviewed again, or a customer waiting for confirmation. Accounting issues invoices late, cash arrives later, and employees work overtime while artificial KPIs fail to show how much revenue the company has missed.

Before the pilot, 14% of records required another review. During peak periods, the bottleneck was reconciliation, not delivery capacity. Hiring more people would increase fixed costs without removing the root cause. That is a half-baked optimization pattern that can quietly erode margins and make the company pay for avoidable mistakes.

3. Solve: Deploy an AI Agent in 3 measurable steps

Step 1 – Collect and classify documents: The AI Agent receives delivery records, proof-of-delivery photos, surcharge sheets, and invoices from selected sources. It extracts order IDs, signatures, delivery times, and document status. Records missing a photo or signature are immediately added to an exception list.

Step 2 – Reconcile under policy control: The Agent compares delivery data with contracts, rate cards, and the operations system. Valid cases continue through the workflow; unusual surcharges, wrong routes, or amount variances are flagged for human approval. HimiTek uses OpenClaw Gatekeeper, 9router v0.4.66, and LiteLLM dual-instance failover to control AI calls, apply rate limits, rotate API keys, and enforce a hard budget cap such as 5 USD per virtual key or developer per month.

POLICY = {
    'budget_cap_usd': 5,
    'allow_tools': ['read_document', 'compare_rate_card', 'create_exception'],
    'require_approval_for': ['approve_surcharge', 'issue_invoice']
}

def route_case(case):
    if case['missing_documents'] or case['amount_variance'] > 0:
        return 'human_review'
    return 'continue_workflow'

The Reasoner only analyzes; the Actuator executes scripts. Elevated tools remain locked by default and are enabled only through a whitelist or explicit permission. This separation reduces the risk of prompt injection taking over the company VPS.

Step 3 – Track exceptions and recovered money: The workflow creates an exception queue, reminds responsible teams to provide missing documents, and updates reconciliation status in real time. The company starts with one delivery route or customer group and tracks three metrics: reconciliation days, recheck rate, and recovered surcharge value.

After eight weeks, reconciliation time fell from six days to approximately 3.7 days, a 38% reduction. The recheck rate dropped from 14% to 5.8%. The company saved approximately 45–60 million VND per month by reducing missed surcharges and overtime, while issuing invoices earlier without hiring additional staff during peak season.

4. CTA: Turn reconciliation into earlier cash flow

If your company processes thousands of shipments but still reconciles them through spreadsheets and scattered messages, start with one route or customer group. HimiTek can help map the document flow, define exception rules, and measure results through processing days, discrepancy rate, and actual savings.

The goal is not to create an impressive AI project on paper. The goal is to remove wasted manual hours, collect every valid surcharge, and move invoices out of the queue sooner.