River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Exception Handling Process Template

Three documents and three sheets that score every recurring exception on frequency and approval rate, so the ones acting like policy get written into it.

Free download  ·  No account needed

A return that arrives 38 days after purchase gets argued the same way every time, because nobody has counted how often it happens or how often it gets approved. Vantage Outfitters, a 40-person apparel retailer, logged six months of return exceptions instead of guessing. Final-sale returns came in more than once a week, the highest volume of any type, and looked like the policy gap. Only 26.5 percent got approved. It was working as intended, argued fresh every single time.

Frequency alone ranks that pair backwards. Price match requests, honored within 7 days of purchase, came in less than once a week, well below final-sale's 1.31. But 89.5 percent got approved, meaning the store was already saying yes almost every time it was asked. Clearing a frequency bar and an approval-rate bar together is not a judgment call being made fresh each time. It is a rule the business already follows in practice, just never written down, which is why price match qualifies for promotion and final-sale returns do not.

The same split appears in statistical process control. NIST's own handbook defines a common cause as a source of variation affecting every instance, one that is difficult to eliminate by reacting case by case. A special cause, by contrast, is intermittent and does call for individual judgment. A frequent, consistently-approved exception behaves like the first kind, and belongs next to the SOP library as a written rule instead of a recurring argument. One that fails the bar stays the second kind, kept as a named category in the Decision Tree rather than re-argued from scratch.

The pair frequency alone gets backwards

Frequency Analysis scores every type; Standardization Candidates lists what actually qualified.

Frequency Analysis

Illustrative rows for a fictional apparel retailer, Vantage Outfitters, 26 weeks of history.

Exception typeRequestsApprovedRatePer weekVerdict
Late return (31-45 days)887989.8%3.38Qualifies
No receipt, has confirmation email615590.2%2.35Qualifies
Final sale item, unworn34926.5%1.31Stays case-by-case
Price match after purchase191789.5%0.73Qualifies
Damaged item, partial refund88100.0%0.31Below frequency bar

Final sale outranks price match on volume alone, 34 requests against 19. Scored on frequency and approval rate together, price match qualifies and final sale does not: a type denied three requests out of four is a working judgment call however often it gets asked.

Qualifies means at least 0.5 requests a week and at least 80 percent approved, together.

Standardization Candidates

The three types that cleared both bars, with the rule change proposed.

Exception typeProposed standard ruleTickets coveredHours saved / 6mo
Late return (31-45 days)Extend the return window from 30 to 45 days with a receipt888.8
No receipt, has confirmation emailAccept an order confirmation email as equivalent to a receipt616.1
Price match after purchaseHonor price match requests made within 7 days of purchase191.9

168 tickets over six months stop needing individual manager review once these three become written policy instead of a request, at roughly six minutes of review time each.

Each row states the rule itself, not just a faster approval, because the log already proved these are not judgment calls anymore.

What's in the pack

01

Exception Policy

What counts as an exception, the frequency and approval-rate bars a recurring one has to clear, and what happens next.

02

Decision Tree

The order to work a request, so a known category gets the same ruling no matter who is on duty.

03

Approval Note

The write-up format for one ruling, short enough for the ticket and specific enough for the next similar case.

04

Exception Log

Every request logged, approved and denied both, because a denial rate only means something measured against what actually got refused.

05

Frequency Analysis

Each exception type scored on how often it comes up and how often it gets granted, ranked on both together.

06

Standardization Candidates

The types that cleared both bars, with the specific rule change proposed, ready to announce and roll out.

How it works

  1. 1

    Send the history

    Tickets, emails or a spreadsheet covering the last two or three months of exception requests, approved and denied both.

  2. 2

    Every request gets logged

    Each one typed into the Exception Log with its type and ruling, building the real denominator the frequency analysis needs per type.

  3. 3

    Types get scored

    The Frequency Analysis ranks each type on how often it comes up and how often it gets approved, together rather than as two separate readings.

  4. 4

    Qualifying types become policy

    Standardization Candidates lists what clears both bars with a proposed rule change, and the Decision Tree updates so the next request is answered consistently.

Frequently asked questions

What counts as an exception versus a normal request?

A request to deviate from a documented standard process, made by someone who already knows the standard exists. If nothing is written down for the situation at all, that is a documentation gap, which the SOP gap audit finds, not an exception to log here.

Why log the denied requests, not just the approved ones?

Because a denial rate only means something measured against what got refused. A log of approvals alone cannot tell a type granted 90 percent of the time from one granted 100 percent, and that gap is exactly what decides whether it becomes policy.

What stops two approvers from ruling differently on the same type?

The Decision Tree, which sends a known category to the criteria the prior cases were judged against instead of a fresh read. Consistency across whoever is on duty is the entire reason a category gets written down at all.

How many exception requests do I need before this works?

A few months of history is enough to see which types clear the bars. Fewer than that still runs, it just means one unusual week carries more weight in the count, so treat an early result as provisional.

What happens to a type that almost qualifies?

It stays in the Decision Tree as its own category with its own criteria, reviewed again next quarter rather than promoted or dismissed on one pass. Policies drift, and so does what counts as routine.

What format are the downloaded files?

Word (.docx) for the three documents, CSV (.csv) for the three sheets, zipped together and needing no conversion. Excel, Numbers and Google Sheets open the sheets directly, and the format documents open in Word or Pages.

What does 'Edit with AI' actually do?

It creates a private workspace with this exact pack installed, then asks for your exception history: tickets, emails or a spreadsheet, approved and denied both. River logs each one, scores the types, and proposes which ones to promote.

Find out which exceptions already are your policy

Send the exception requests from the last few months, approved and denied both, and see which types clear the bar to become a written rule.

Edit with AI