A request closes on Thursday. The export shows a creation date and a closed status.
Between those fields, someone reviews the request, asks for clarification, and checks the finished work. The export contains none of those steps.
Before choosing process mining software, establish what your records can show. A clean spreadsheet can still leave most of the process invisible.
Use the checklist below to assess one service workflow. The worksheet and acceptance decisions are proposed methods. Every request, timestamp, and business rule in the examples is fictional. No customer export or employee activity record supplies the examples.
Define the Question Before Inspecting Columns
Start with a question narrow enough to answer from recorded activity: which recorded steps occur between request receipt and closure?
Keep the intended conclusion beside it. You want to compare sequences. You are not yet estimating effort, assigning responsibility for delay, or deciding which work to automate.
An operations automation audit can identify a workflow worth examining. This readiness check asks whether its existing records support that examination. It does not collect desktop activity or build an integration.
Write the period, request types, and included systems above the worksheet. For this fictional review, use service requests created and closed within a defined sample window. Exclude open requests from the sequence comparison, but count and disclose them separately.
That boundary limits the finding. You cannot describe all incoming demand using only completed requests. Likewise, a single team's export cannot establish what happens in a system the export never includes.
Keep the selection rule fixed while assessing the sample. If you change it, save the new rule with the revised result.
Give Each Row a Meaning
Microsoft's Power Automate documentation identifies three essentials: a case identifier, an activity name, and time information. A case represents one process instance. An event log has one timestamp per event; an activity log has start and end timestamps. Microsoft: data requirements.
For this example, one case is one service request. The activity names describe recorded milestones. The timestamp records when the source says each milestone occurs. A separate event identifier helps us trace individual rows; it is an editorial addition to the example.
| Event ID | Case ID | Activity | Event timestamp |
|---|---|---|---|
| E101 | SR-101 | Received | 2026-10-20T09:00:00Z |
| E102 | SR-101 | Reviewed | 2026-10-20T09:30:00Z |
| E103 | SR-101 | Completed | 2026-10-20T11:00:00Z |
| E104 | SR-102 | Received | 2026-10-20T10:00:00Z |
| E105 | SR-102 | Clarification requested | 2026-10-20T10:20:00Z |
| E106 | SR-102 | Clarification received | 2026-10-21T08:00:00Z |
| E107 | SR-102 | Completed | 2026-10-21T09:00:00Z |
These seven rows describe two cases. They provide no paired activity start and end times. Treat the gap between Received and Reviewed as an interval between recorded milestones. Do not call it thirty minutes of review work.
If your source offers activity intervals, record what starts and stops each interval. An open task can contain interruptions. A label such as Started needs a source definition before it becomes evidence of active work.
Keep the Case Intact Across Handoffs
Trace SR-101 through every included record. Its case identifier should continue to mean the same request after review, reassignment, and completion.
Now propose two failure examples. A service desk creates a new identifier when it transfers a request. Another export uses the company identifier for every request from that company. The first splits one case. The second joins separate cases.
Neither problem is repaired by renaming a column Case ID.
Ask the source owner which identifier represents the process boundary you selected. If a transfer creates another identifier, look for an explicit relationship in the source. Record that relationship before treating both identifiers as one request.
Keep the original identifiers beside any approved mapping. A reviewer should be able to trace the assembled case back to its records. Do not join two requests because their descriptions sound similar.
Write a rule for reopened requests too. Does reopening continue the original case, or begin another service episode? Choose the rule that fits the question. Document the choice before comparing cases with different histories.
Mark identity unresolved when the source cannot establish continuity. A guessed join can make a tidy sequence that never occurred.
Look for History, Missing Events, and Cutoffs
A current-state table may omit the history process mining needs. Microsoft also warns that some events of interest may never be logged. Microsoft: locating log data.
Return to SR-102. It contains clarification milestones but no Reviewed milestone. Several explanations remain possible in this fictional case: review was skipped, review was unrecorded, or the export omitted it.
Write all three possibilities down. Check the source record and export selection before choosing one. If the source contains the event, mark the export repairable from source. If no evidence survives, leave the event unknown.
Inspect both edges of the sample window. A request received before the window may appear to start at review. A request still open at the cutoff may appear to stop after clarification. Those paths need explicit boundary labels.
Compare the sample's case list with the source's list under the same selection rule. Reconcile missing identifiers before relying on the count. Do not use a larger row total as proof that the export contains every required event.
Keep a short exclusion note: which cases were omitted, why, and whether their absence changes the question you can answer.
Separate Duplicate Rows From Repeated Work
Add a fictional export error: E102 appears twice with identical contents. The source owner confirms both rows refer to the same stored event. The reviewed sample can retain one event while recording the removal.
Now change the example. SR-101 contains another Reviewed event, E108, after a correction. Its activity label matches E102, but its event identifier and business context differ. Removing it would erase repeated work.
Use source identity and context to judge duplicates. Equal case IDs and activity names do not establish that two rows describe one event. Even equal timestamps need inspection when the source records time coarsely.
Preserve the untouched sample and a short transformation note. State which rows were removed, what evidence justified removal, and which rule produced the result. The readiness decision should remain explainable without remembering a spreadsheet operation.
If an export has no stable event identifier, ask what combination of source fields establishes a distinct event. Leave ambiguous pairs unresolved until the source owner answers.
Do not collapse repetition to make the process map easier to read. Repetition is part of the evidence under review.
Check the Clock Before Trusting the Order
For each timestamp column, document the source meaning, time zone, stored precision, and export formatting. Use a synthetic case with known ordering to inspect the conversion.
In the sample table, the Z suffix denotes a zero UTC offset. RFC 3339 defines that suffix. Keep that interpretation explicit. If another source supplies local time without an offset, ask its owner for the applicable zone and date rules before combining records.
Suppose Received and Reviewed both export as 09:00 because the export drops minutes. Their order is unresolved at that precision. Alphabetical sorting would create an order without evidence.
Power Automate's documented variant analysis uses event start time to arrange events. Its sequence therefore depends on the time fields supplied. Microsoft: process variants.
Check tied times, reversed intervals, and events dated outside the selected period. For each anomaly, return to the source value. A formatting repair is useful only if the original precision or zone information survives.
Never manufacture seconds to force a preferred sequence. If ordering remains unknown, state which comparisons the ambiguity prevents. You may still describe observed activity membership while holding conclusions about its exact order.
Decide What Attributes Mean
An event attribute describes an individual event. A case attribute describes the process instance. Power Automate supports first-event and last-event interpretations for case attributes, while event-level functions use the supplied event values. These are documented product semantics, not a rule for every tool. Microsoft: optional attributes.
For the fictional requests, use request type as a case attribute and handling department as an event attribute. Keep personal names out of the exercise. A team-level context field is enough for the questions here.
Suppose request type changes from Standard to Specialist after review. Decide whether your question concerns the type at intake or the type at completion. Record that decision with the selected interpretation.
A column containing multiple values within one case deserves inspection. Do not replace those values with a convenient majority vote. That would introduce another rule whose effect needs explanation.
Keep business meaning separate from import settings. Write the intended meaning in plain language first. Then ask whether the selected tool and configuration preserve it. This checklist relies on the cited documentation checked October 28, 2026; no product installation is tested here.
Use a Readiness Table With Evidence
Give each check one of three statuses. Usable means evidence supports the specific question. Repair from source means the needed evidence exists and a defined correction remains. Unknown means the evidence does not yet support a decision.
These labels are proposed review states, not vendor certification. The table illustrates a fictional sample assessment.
| Check | Status | Evidence or next action |
|---|---|---|
| Case continuity | Usable | SR-101 retains its request identity through completion |
| Duplicate E102 | Repair from source | Source confirms one event; remove the duplicate export row |
| Missing Reviewed for SR-102 | Unknown | No evidence yet distinguishes omission from skipped work |
| Timestamp precision | Usable | Example source retains seconds and UTC values |
| Older activity labels | Repair from source | Approved label dictionary identifies a renamed milestone |
| Request type changes | Unknown | Review owner must choose intake or completion meaning |
| Work outside the request system | Unknown | No supporting record included |
For a label change, preserve both the original label and its approved interpretation. Map two labels together only when the source owner confirms they represent the same activity. Keep different activities separate even when their names look similar.
Attach a source reference to each usable finding. Attach a specific correction to each repairable finding. Attach an owner question to each unknown. The status alone does not explain the decision.
Make a Bounded Readiness Decision
Assess readiness against the original question. There is no useful overall green light if a missing event prevents the sequence comparison you intend to make.
For the fictional sample, SR-101 can support a basic recorded sequence after the duplicate repair. SR-102 needs its missing-review question resolved or an explicit limitation. Neither case supports an employee effort comparison.
Use a business system dependency map to identify which source supplies each missing piece. Use SOP documentation to locate the intended process definition. Neither a dependency diagram nor a written procedure proves that a particular event occurred.
Record the decision as a short statement: approved question, included cases, completed repairs, unresolved limits, and the person accepting those limits. Keep the sample and its transformation notes with that statement.
If unrecorded calls or offline checks matter to the question, acknowledge the coverage gap. Do not infer employee performance, misconduct, or causation from the remaining path.
A useful readiness review ends with a defensible boundary. Know what the log can show.
Sources
- Microsoft Learn: Prepare processes and data. Required fields, optional attribute interpretations, and history limitations. Checked 2026.10.28.
- Microsoft Learn: Analyze processes with variants. Ordering used in Power Automate variant analysis. Checked 2026.10.28.
- RFC Editor: RFC 3339, Date and Time on the Internet: Timestamps. Definition of the Z suffix in timestamp examples. Checked 2026.10.28.