Exception routing calibration, worked step by step: a safety case misrouted to the warranty queue

Short answer

The records that explain how an exception was handled are the ones that let someone replay it: the case as it arrived, the rule that fired and its version, the queue and SLA it got, the calibration sample that caught it, each reviewer's label, the ruling that settled their disagreement, and the rule change that followed. The worked case below follows one safety report misrouted to a warranty queue through those seven records, then shows how to recognize similar cases at a client.

Visual briefFull-size diagram
A case timeline that branches at the routing rule: the warranty branch shows a 19-hour wait, then the calibration sample, three reviewer labels split two to one, the baseline ruling and the reordered rule, each with the record that captured it.
Illustrative. The provider, queues, times and labels are fictional and show how one misroute is captured in records.

This worked example is part of the exception routing calibration guide, which sets out what to inventory before a routing tool is replaced. Here the task is narrower: show, with one hard case, why the sequence of decisions is the evidence and the final rule set is not.

The case: a sparking unit filed as a warranty claim (Illustrative)

A fictional 380-person equipment service provider runs two queues among several. Safety hazard cases target a 4-hour first response. Warranty claims target 2 business days. A customer writes through the web form: "Warranty claim: unit sparking when switched on."

The case fits both queues. The words "warranty claim" describe what the customer wants; the word "sparking" describes a hazard. Which queue it lands in depends on rule order, and the right answer depends on a judgment the rules did not encode. That is the exception: a case that fits two service queues with different SLAs. This is not a real company or a completed deal.

Step by step: who decided and which record caught it

StepWhat happenedWho or what decidedRecord that captured it
1. ArrivalCase created from the web form; priority set to normal by a form defaultForm configurationCase record: text, channel, product, timestamp, priority
2. RoutingRule 1 matches "warranty" and assigns the warranty queue before rule 2 ("sparking", "smoke", "burning") is checkedRouting tool, by rule orderRule execution log with rule ID and version, if the router keeps one
3. SLAThe warranty queue's policy applies; first response comes after 19 hoursSLA policy orderResponse-target policy on the case; policy list on that date
4. SamplingThe monthly calibration sample of 60 routed cases includes this oneQuality operations teamCalibration session with case IDs
5. LabelsThree reviewers label it independently: two say Safety, one says WarrantyReviewersIndividual labels with reviewer, comment and date
6. RulingSafety terms override product-claim terms, whatever else the case saysQuality operations headBaseline ruling in the review meeting
7. CorrectionThe safety rule moves above the warranty rule; a safety-first routing label is added; next month's sample is re-checkedWorkflow engineering, on the rulingNew rule version with a change comment citing the meeting

Steps 2 and 3 are not hypothetical mechanics. In Zendesk, ticket triggers are checked in order from first to last each time a ticket is created or updated, so the first matching rule can decide the queue before a later, better rule is reached. After triggers run, the ticket goes through the SLA system: policies are compared from the top of the list, the first match applies, and a ticket holds one SLA policy at a time. A blank Priority field means no SLA applies at all, which is why the form default in step 1 matters too.

Why this sequence is useful evidence

Each record answers a different question a reviewer would ask:

  • Steps 1 to 3 show the problem as the system saw it: what was known, what the rules did with it, and what it cost in time.
  • Steps 4 and 5 show people disagreeing about the same facts, which is the hardest part of routing to capture and the most informative.
  • Step 6 shows an authorized judgment that resolves the disagreement and states a principle.
  • Step 7 shows the principle turned into a rule, and the re-check shows whether it worked.

Calibration tools often keep steps 4 to 6 apart from everyday quality data. In Zendesk QA, for example, reviewers rate the same conversations and compare scores and comments, a session can mark a baseline review, and calibration results appear only in the calibration dashboard, not in internal quality scores. A report built from quality scores alone would never show the disagreement.

Together the seven records connect a business exception to an authorized action and a checked result. That is the kind of operational history described on why operations data is valuable and in exception handling records for AI agents.

The three records most often lost

Ask about these first, because they disappear quietly:

  1. The rule version at step 2. Engines often log the outcome (queue assigned) but not which rule produced it. Keep only today's rule set and the case looks as if it was always routed correctly.
  2. Individual labels at step 5. Many teams keep only the consensus. The split vote is what shows the case was genuinely hard.
  3. The ruling at step 6. Rulings are often written in meeting notes or chat, with no link to the rule change they caused.

Keep only the case and nobody can see why it took 19 hours. The system map shows which system usually holds each of these and the identifier that joins them.

How to recognize similar cases at a client

A workflow automation advisor can spot this pattern in a short conversation without looking at a single case. Listen for:

  • Two or more queues with different response targets, and keyword or category rules that decide between them.
  • A regular review or QA sampling routine, monthly or quarterly, with named reviewers.
  • Rule changes that someone can tie to a review finding ("we moved the safety rule up after the March review").
  • A dated records event, such as a routing tool replacement or the end of a QA tool contract.

Similar exceptions appear in other workflows: a procurement request that fits two approval paths, covered in procurement approval exceptions, or a closure call that could go to retention or cancellation, as on the call center closure recordings page.

What a partner can say, and what stays unknown

Describe the case history at metadata level: which queues, how many review cycles, how many cases per sample, whether individual labels and rulings survive, which systems hold them and who owns them. Never pull cases or labels to demonstrate the point. The referral brief template gives the fields.

The introduced company needs 50+ full-time employees at peak (contractors excluded), a documented operating history, rights to license and an authorized contact. Companies in the U.S., Canada and the EU qualify, as do companies in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia when at least 70% of the data is in English. Unknown until SourceX reviews it: whether customer case text can be used, how it would be de-identified, SourceX's current data needs and any offer. Partners sign the Channel Partner Agreement before submitting; the first valid introduction within the 90-day window is credited when the company submits a verified application.

Next step

Refer a company. No cap on rewards. Default reward on a company's first qualifying transaction: the greater of $50,000 plus 5% and 15% of Actual Supplier Payout, with no cap; further qualifying transactions first agreed within 12 months of the first Closing pay 15%. Eligibility and payment conditions apply, including the USD 100,000 Minimum Transaction, and a signed Schedule A or addendum can differ. A lead, meeting or signature alone does not trigger payment.

If a client can still replay its hard routing cases from arrival to rule change, register as a partner, check the company on who qualifies and read the program terms before you describe them to the quality operations head.

  1. Step 1Share your linkSend your personal link to a company you know.
  2. Step 2Company appliesThe company applies itself at /apply.
  3. Step 3Company signs and is paidThe company signs, delivers its data and is paid by SourceX under its agreement.
  4. Step 4You get your rewardthe greater of $50,000 plus 5% and 15% of the actual supplier payout, with no cap. A reward becomes payable after SourceX has paid the referred company under its signed agreement and the payout conditions in the program terms are met, including finance approval and your tax and payment details.

Common questions

Is one misrouted case enough to interest anyone?

On its own, no. What matters is that the company kept the same sequence for many cases over several calibration cycles: the arrival, the rule, the labels, the ruling and the change. One worked case is how you explain to the client what to look for.

What if reviewers only recorded the agreed label, not their individual ones?

The case still shows a ruling, but not the disagreement that led to it. Note it in the brief as consensus only, with the dates. SourceX's review decides how useful that is.

Does the client have to use Zendesk for this to apply?

No. Zendesk is used here because its trigger, SLA and calibration behavior is documented. Any case manager with ordered rules and SLA policies raises the same questions; ask the owner how their platform records each step.

Sources

  1. support.zendesk.com/hc/en-us/articles/5600997516058-About-SLA-po…
  2. support.zendesk.com/hc/en-us/articles/4408886797466
  3. support.zendesk.com/hc/en-us/articles/7043724530842-Setting-up-c…
  4. refer.sourcex.si/terms

Free resources

Know a company with valuable proprietary data?

Become a referral partner, get your link and introduce the company’s CEO (in an owner-run company, usually the owner). Referral partners can join from most countries. We can't work with partners in countries or regions under U.S. sanctions or on U.S. restricted-party lists. The companies you introduce should be based in the U.S., Canada and the EU, or in the UK, Australia, New Zealand, India, Singapore, South Africa, Latin America, the UAE and Saudi Arabia with at least 70% of their data in English.

Refer a company →

I own a business

Explore licensing your company's data directly to SourceX for AI training and evaluation. Start a short assessment; no uploads needed.

Start an assessment