A week of warranty requests
An equipment supplier's manager must choose manual or automated warranty checks this week. Customers paid for equipment with warranty support included. Instant replies promise less staff work, but an unsupported rejection could leave a covered customer paying for repairs.
This hypothetical example assumes 120 requests, eight staff hours available, and 70 registration records missing a warranty_end value.
The invoice says yes; the rule says no
Those 70 requests include invoices establishing coverage. The automated rule reads only the registration record and treats a blank warranty_end as expired. It drafts a rejection despite the invoice. Unavailable evidence has become negative evidence.
Like an inspector facing an unreadable label, the system needs a way to ask for more evidence.
Count the work after the reply
Manual handling takes three minutes per request, including reading its invoice: 120 × 3 = 360 minutes, or six hours.
Automation takes 60 minutes to configure and check. Assume each rejected draft needs six minutes to trace the blank field, inspect the invoice and correct the reply. Holding all 70 rejections for human approval gives 60 + 70 × 6 = 480 minutes, or eight hours. The other requests need no additional handling.
Both options fit, but manual handling leaves two hours available. Choose it this week; reply speed alone misses the correction work.
Inspect the decision before sending it
Compare each rejected draft's warranty_end input with its invoice's coverage dates. A covered invoice paired with a blank field and rejection exposes the faulty rule. Hold that reply, arrange covered service, and route missing evidence to human review.
Now change this hypothetical week: only 20 requests need review. Automation takes 60 + 20 × 6 = 180 minutes, or three hours. With the same review safeguard, choose automation. The options tie at 50 reviews; below that, automation saves time. Keep manual handling at or above that threshold.