Call Center OutsourcedPowered by StealthAgents

Call Center Outsourced evidence brief · Desk review · Published

Where does outage evidence break between the client and the vendor?

A shared-timeline approach to outage evidence: compare the two clocks, trace the fallback path, protect promises, and reconcile before normal work resumes.

Where does outage evidence break between the client and the vendor? editorial illustration
On this page
  1. Key takeaways
  2. Question and scope
  3. What the evidence supports
  4. Define the outage boundary
  5. Compare the two clocks
  6. Find the last good record
  7. Trace the fallback queue
  8. Prove what customers were told
  9. Separate outage from performance
  10. Count the promises affected
  11. Review the manual notes
  12. Check the access log
  13. Handle a partial outage
  14. Reconcile after recovery
  15. Test the incident timeline
  16. Document the shared facts
  17. Use the record in the client review
  18. Fix the weakest evidence
  19. Limits of the evidence
  20. Manager review questions
  21. Conclusion
  22. Put this into a support lane
  23. Related operating guides
  24. FAQs
  25. Sources

Key takeaways

  • Use one stated clock and explain every difference between the two systems.
  • Trace one customer contact through every fallback location.
  • Count the promises affected and show what customers were actually told.
  • Reconcile and sign off before the queue returns to normal work.

Question and scope

This brief asks where outage evidence breaks between a client's systems and an outsourced queue. The population is one outage during a declared period, and the unit is a customer contact or promise affected by the failure. The question is whether the two organizations can reconstruct the same timeline. The analysis covers evidence, not fault. Record the systems, the start time, and the first affected case. Keep the definition with the incident plan so both teams use the same words.

What the evidence supports

Federal contingency-planning guidance calls for tested recovery procedures and documented roles, and governance frameworks call for traceable evidence. Neither source defines a shared outage record between a client and a vendor. The operating inference is to compare the two clocks and the two case records. The gap between them is where disputes begin. Record the customer impact at the moment of declaration, not only at the end.

Define the outage boundary

Write a definition that names the systems, the users, and the customer impact that count as an outage. A slow report is different from a failed case system. Record who declares the outage and who ends it. The same definition should be used by the client and the vendor. Name the clock used in the timeline and convert the other logs to it.

Compare the two clocks

The client's monitoring system and the vendor's queue log may record different start times. Compare them at the minute level and explain the difference. A time-zone error or a delayed alert can move the boundary by an hour. The shared timeline should use one stated clock. Attach the last good case and the first fallback row to the timeline.

Find the last good record

Identify the last case that was fully recorded before the failure. Use it as the anchor for the manual entries that follow. Check that the case number, the owner, and the promise are consistent. A fallback record that cannot be tied to the last good case is hard to reconcile. Trace a second contact through a different fallback location to test the map.

Trace the fallback queue

List every place work went during the outage: a spreadsheet, a shared inbox, a phone list, or another system. Name the owner of each. Reconstruct the path of one customer contact through the fallback. The trace shows whether the queue knew where its work was. Compare the approved message with the recording or transcript for the same hour. A tail that appears only during the outage may be a fallback routing effect.

Prove what customers were told

Collect the approved message, the time it was used, and the channel where it appeared. Compare it with what agents actually said in the sampled contacts. A message that was written but never delivered does not protect the customer. The record should show both the plan and the delivery. Show the outage interval as a separate band in the service report.

Separate outage from performance

A queue can miss its service level during an outage without failing at its normal work. Report the outage-affected interval separately from the rest of the day. Check whether the same agents handled the fallback correctly. This separation prevents an infrastructure failure from being scored as an agent failure. Count the customers behind the promises so a single case is not overcounted.

Count the promises affected

List the promises whose windows fell inside the outage. Mark whether each was kept, renegotiated, or missed. Include the promises made during the outage, because they may carry a different risk. The count should be by customer, not only by case. Mark each manual entry with the source that supports it. The count should separate promises made before the outage from those made during it.

Review the manual notes

Manual notes are often short and inconsistent because they were written under pressure. Sample them against the phone log and the customer message. Mark entries that lack a time, an owner, or a next step. Do not treat a later cleanup as the original evidence. Record the edit reason and the person who made it. The sample should include at least one entry written by each person on the fallback.

Check the access log

Confirm who could enter the fallback record and when. A record that anyone can edit without a trace is weak evidence. Compare the edit times with the shift and the outage window. The access log should explain any change that altered a promise or an owner. Test the record against each function that was unavailable. A record with no access trail should be treated as an estimate, not as proof.

Handle a partial outage

Some systems may fail while others keep working. Record which functions were available and which were not. A partial outage can produce a record that looks complete but is missing one field, such as the disposition or the verification step. Test the record against the function that failed. Record the reconciliation owner and the time it was completed. The recording and the fallback entry should agree on the time and the next step.

Reconcile after recovery

Reconcile the fallback entries with the main system before normal work resumes. Look for duplicates, missing cases, and actions taken twice. Give the reconciliation an owner and a completion time. A queue that resumes without reconciling carries two versions of the same case. Note every disagreement beside the fact rather than removing it. The person who reconciles should sign a short statement that the two records now agree.

Test the incident timeline

Build a timeline from the first alert to the final reconciliation and mark the source of each entry. Ask the client and the vendor to review it separately before the joint meeting. Differences in the timeline should be explained, not smoothed over. A timeline that both sides accept is the foundation of the review. Attach the client and vendor source documents to the shared record.

Document the shared facts

Write one outage record with the agreed start, end, affected cases, customer messages, and recovery actions. Keep the client's and the vendor's source documents as attachments. The shared record should not assign fault; it should establish what is known. Disagreements can be noted beside the facts. Ask for one system or process change with an owner and a date. The shared record should show which facts were confirmed by both organizations.

Use the record in the client review

Present the shared record with the promises kept and missed. Show what the fallback protected and what it could not. Ask for a specific change to the plan or the systems. A review that only describes the outage will not prevent the next one. Test the fix in the next exercise and record the result.

Fix the weakest evidence

Choose the part of the record that failed most often, such as the timestamps or the manual owner field. Fix that before adding a new tool. Test the fix in the next exercise. A small improvement in the weakest link is worth more than a polished report. State the limits of the record in the same document as the timeline. The fix should be small enough to test before the next outage exercise.

Limits of the evidence

This brief cannot determine fault or a contractual remedy for a specific outage. The records may be incomplete, and the two organizations may use different definitions. A shared timeline can still leave the customer impact uncertain. Use the evidence to improve the plan and state what it cannot prove. Record the answers with the shared timeline and the promise count.

Manager review questions

Managers should ask which systems failed, which fallback path was used, and who owned the customer promise during the gap. They should also ask whether the customer message matched what agents said and what the reconciliation found. The remaining question is which evidence gap will be fixed first. Record the answers with the shared timeline. The record should let a new reviewer follow the same path without asking either team.

Conclusion

Outage evidence breaks where the clocks, the case records, and the customer messages diverge. Build one shared timeline, trace the fallback path, and reconcile before normal work resumes. Separate the outage interval from ordinary performance. The record is useful when it helps both organizations protect the next customer promise. A shared outage record is useful when it improves the next response.

Put this into a support lane

Bring one queue, its reporting extract, and the manager who owns the operating decision.

Review the queue

FAQs

Whose outage record should be authoritative?

Neither side alone. A shared timeline built from both source logs, with disagreements noted, is the strongest record.

Should the outage interval count against the queue scorecard?

Report it separately so an infrastructure failure is not scored as an agent performance failure.

Sources

  1. NIST SP 800-34 Rev. 1: Contingency Planning Guide for Federal Information Systems
  2. NIST SP 800-61 Rev. 3: Incident Response Recommendations
  3. GAO-20-283G: Assessing Data Reliability
  4. ISO 18295-1:2017 Customer contact centres