Research question and scope
This protocol measures how often a customer address change is completed without a second verification. The unit is one address change request tied to one account. A bypass means the record shows no confirmation through a separate channel before the change took effect. The measure describes the verification control, not the intent of the customer.
Set the scope to one order desk, one account system, and a fixed period. Record the approved verification methods and the change policy in force. If the policy changes during the period, split the sample at the effective date. A different method creates a different denominator.
Keep the measure separate from a fraud loss count. A change can be verified and still be part of a scheme, and an unverified change may be entirely legitimate. The purpose is to test the control before a loss occurs. Report fraud outcomes separately.
Name the decision the result will support before collection starts. The client may use it to add a callback step, restrict same-thread confirmation, or require a hold on open shipments. A report without a decision leaves the desk with the same risk. Write the intended action beside the scope.
Record whether the account is consumer or business before sampling. A business account may have an authorized contact who can confirm a change, while a consumer account usually needs the holder. Record the account type and the authority rule. Mixing the two rates hides where the control actually fails.
Population and unit of analysis
Define the population as every address change request that reached the desk during the period. Include changes made by phone, chat, email, and the customer portal. The unit is the request, not the order or the customer. A customer who changes a billing and a shipping address in one request is one unit when the change is recorded together.
Include requests that were refused and requests that were later reverted. Mark the outcome on each record. Exclude a change made by the client's own finance team through a system that does not use the desk. Report the excluded count so the desk's share is visible.
Exclude a duplicate request for the same account and address within the same session. Link it to the original and count it once. Duplicate handling is a separate efficiency measure. State the exclusion rule before sampling so a second reviewer applies it the same way.
Decide how to treat a change that only corrects a postal code typo. Include it but mark it as a minor correction. A minor correction may still redirect a parcel, but it carries a different risk from a full street change. Report the two groups separately.
Decide whether a change made at the customer's request during a return is in scope. A return address is not the same as a delivery address, and the authority may sit with a different team. Record the case type and the team that handled it. The exclusion should be written before the sample, not after a difficult case appears.
Definitions that make the measure reproducible
Define a second verification as a confirmation through a channel different from the one that carried the request, with the person and time recorded. Define a partial verification as a same-channel reply or an unrecorded callback. Define a bypass as no second verification before the change took effect. Publish one example of each.
Define the change effective time as the moment the account or order record was saved with the new address. A verification completed after that time does not protect the change. Record both times. The gap between them is part of the result.
Define the approved channels in writing, such as a callback to the number on file or a code sent to the verified contact. A reply in the same email thread is not a separate channel. Record the channel used and the reference. This rule is the core of the measure.
Define what counts as a completed second check. A call that rings out is not a confirmation, and a voicemail is not a conversation. Record the outcome as confirmed, attempted, or unavailable. A failed attempt is different from a check that never happened.
Decide how to classify a change requested by a third party, such as a gift recipient or a reseller. Keep it in the sample and mark the requester type. A third-party request has a different authority question and should not be merged into the customer rate. Report the third-party share beside the main result so the reader can judge the mix.
Sampling and data collection
Draw the sample from the change log, the order system, and the contact queue for the full period. Record the export date, the filters, and the row count. Do not build the sample from fraud alerts alone. An alert-based sample cannot show the changes that were never flagged.
Capture the request identifier, the account, the requester, the channel, the requested address, the effective time, the verification method, the confirming person, and the shipment state. Add the order edit reference where one exists. Store the exports with the report. The verification field and the effective time are the core evidence.
Check a small overlap sample with a second reviewer. If the reviewers disagree about whether a same-thread reply counts as a second channel, the definition is too vague. Rewrite the rule and retest before the full sample. Agreement matters more than sample size.
Record the time spent matching change records to order edits. A report that requires reading every chat thread will not be repeated. Note the fields the order system does not capture. The collection cost belongs beside the result.
Take a small pilot sample from two channels before the full draw. A phone change and a portal change may require different evidence and different export fields. Record the pilot size and what was missing. The pilot prevents a full sample from discovering a missing field at the end.
Measuring bypassed verification
Report the share of changes with a complete verification, a partial verification, and a bypass. Show the counts behind each share. The three groups describe different levels of control and should not be collapsed. A single bypass rate hides the change that had a same-channel reply.
Break the result down by channel, requester type, and shift. A portal change may carry its own login check, while a phone request may rely on a callback. A night shift may lack the person who can approve a hold. Compare only groups with a similar method and volume.
Report the share of bypassed changes that affected an open shipment. A change to a future order is different from a change that reroutes a parcel already in transit. Include the carrier intercept cost where one was incurred. The shipment link shows where the risk becomes money.
Report the share of changes that were reverted within thirty days. A revert may signal a mistake, a fraud attempt, or a customer who changed their mind. Include the reason where recorded. The revert rate belongs beside the bypass count, not in place of it.
Report the time between the address change and the first customer contact about it. A quick contact may mean the confirmation message was unclear, while a late contact may mean the parcel arrived at the wrong place. Record the median and the longest gap. The timing helps separate a communication issue from a control failure.
Limitations and next action
The measure cannot show whether a change was fraudulent or whether a customer understood the check. It also cannot show a verification completed in a system that the desk does not record. Treat the result as a measure of the recorded control. Note the channels and account types that were not included.
Give the result to the order desk owner and the account security owner with one change to test. The change might be a required callback field, a hold on open shipments, or a same-channel block. Record the effective date and repeat the measure with the same definitions. A controlled change is easier to judge.
Review the boundary cases before tightening every channel. A few business accounts with a named authorized contact may explain a large share of partial records. Fix the authority rule first if that is the pattern. The goal is a verified change before the parcel moves, not a lower bypass number by definition.
Set the next measurement date and keep the same channels and account types. Record any change to the verification method in the note. Compare the bypass rate and the shipment-edit rate together. A drop in one without the other may mean the recording improved rather than the control.
