All case studies
Logistics·April 18, 2026

Pune manufacturer saved 8 hours/day on ops exceptions with an AI agent

Inventory and supply chain exceptions were eating the ops manager's days. Arthat built an agent that detects anomalies, investigates them across 6 systems, and auto-resolves routine cases while preparing complex ones for human review.

8 hrs/day recovered · 70% exceptions auto-resolved · -42% stock-out risk
Client
A Pune-based auto-components manufacturer
Tools
SAP · Tally · Claude · Gmail · Slack · Zapier · Google Sheets
Pune manufacturer saved 8 hours/day on ops exceptions with an AI agent

Results

8 hrs/day
recovered from ops manager daily fire-fighting
70%
of exception events auto-resolved
-42%
reduction in stock-out risk over 90 days
2 days → 30 min
average exception resolution time

The challenge

A Pune-based auto-components manufacturer (supplying Tier-2 to the automotive majors) had a classic mid-market ops problem: everything worked most of the time, but the 5% of the time something didn't work consumed most of the ops team's energy.

Exceptions included: delayed deliveries from sub-suppliers, inventory mismatches between SAP and physical stock, PO discrepancies, quality rejections, customer complaints about shipment content, invoice mismatches. Each one required the ops manager to pull data from 3–6 systems, understand what happened, and decide what to do.

The ops manager was spending 80% of his week investigating exceptions. Strategic work — supplier relationship management, process improvement, cost optimization — got the leftover 20%.

What we built

Arthat built an exception monitoring and resolution agent that sits across the company's operational systems:

Cross-system anomaly detection: The agent monitors events across SAP (ERP), Tally (accounting), Gmail (supplier communications), shared Google Sheets (production tracking). When an anomaly is detected — an expected delivery not received, an invoice that doesn't match a PO, a stock count variance, a quality rejection — the agent creates an exception record.

Cross-system investigation: For each exception, the agent pulls relevant context from all systems:

  • For a missed delivery: PO details, contracted timeline, supplier contact, recent communications
  • For an invoice mismatch: original PO, GRN, invoice, any communication about changes
  • For inventory variance: recent movements, recent counts, possible reasons

The agent drafts a summary of what happened, the likely cause, and the recommended resolution.

Auto-resolution for routine cases: Common, well-understood exception types get auto-resolved:

  • Small invoice rounding mismatches (under ₹500) → auto-approved
  • Routine delivery delays (under 48 hours with supplier notice) → auto-acknowledged + customer notified
  • Stock variances within tolerance → auto-reconciled with a note
  • Missing GRN on receipt → auto-drafted and routed to the warehouse lead for confirmation

Escalation with full context: For complex exceptions (quality issues, significant delays, disputed invoices, new supplier problems), the agent prepares a structured briefing for the ops manager: what happened, what's been tried, what options exist, recommended action. The manager reads and decides in 5 minutes instead of 45.

Supplier communication drafts: When communication is needed (chasing a delayed delivery, clarifying an invoice, confirming quality requirements), the agent drafts an email in the company's tone, ready for human review + send.

Daily + weekly reports: Every morning, a 1-page report: exceptions opened/closed yesterday, open items aging, trends, top suppliers with exception count, top SKUs with stock-out risk.

Arthat AI

Want results like these at your company?

We built this for them. We can build something similar — or something completely different — for you. Start with a 30-minute discovery call and we'll tell you honestly whether AI is the right tool for your situation.

Ready to automate the work that’s slowing you down?

Book a 30-minute discovery call. We’ll listen, scope, and tell you honestly whether AI is the right tool for the job.

See our work first