Lesson 46 — Designing an Operational Microsoft Sentinel Workbook
Module 4 taught us to investigate incidents.
Now we build the operational views that help analysts spot patterns, understand activity and decide where to look next.

What you will learn
Plan a workbook that serves a real SOC need.
Learning objectives
- Explain how an operational workbook differs from an incident investigation.
- Define the audience and decisions a workbook should support.
- Choose appropriate telemetry, measures, time ranges and visualisations.
- Design useful filters and drill-down paths.
- Consider query performance, access and data quality.
- Sketch a workbook specification before building it.
Why this is not Lesson 8 again
Lesson 8 introduced Sentinel workbooks and dashboards. Module 5 moves from knowing the feature to designing useful operational reporting.
Our goal is not a page full of attractive charts. It is a page that helps the SOC make better decisions.
The Agent Foskett question
What a workbook does
Microsoft Sentinel workbooks provide interactive visualisations and reporting over relevant data sources. They can combine text, queries, parameters and visual elements to support monitoring, analysis and investigation.
A workbook is a view of data, not a replacement for detection rules, incident management or raw-event validation.
Pick the audience first
- SOC analyst: recent suspicious activity, drill-down links and investigation pivots.
- SOC lead: queue trends, coverage and operational bottlenecks.
- Detection engineer: rule quality, noise and data availability.
- Management: carefully defined, high-level measures with limitations explained.
Start with decisions, not visuals
| Question | Useful measure | Possible display |
|---|---|---|
| Are failed sign-ins increasing? | Failures by hour | Time-series chart |
| Which accounts have repeated failures? | Failures by account | Ranked table |
| Where should an analyst investigate? | Source IP, account, application and time | Detailed table |
| Is data arriving? | Latest event timestamp by source | Status table or text |
An operational workbook blueprint
Define each metric precisely
“Failed sign-ins” is not enough. Specify the table, result conditions, time window, grouping and whether counts represent events or unique users.
An attractive metric without a definition can mislead the person relying on it.
Choose the right visual
| Display | Best used for |
|---|---|
| Time-series chart | Change over time |
| Bar chart | Comparing categories or top contributors |
| Table | Detailed entities, evidence and drill-down |
| Number / KPI | A clearly defined summary value |
| Text / guidance | Interpretation and next investigation steps |
Build a simple data foundation
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
SigninLogs
| where TimeGenerated > ago(24h)
| where ResultType != "0"
| summarize FailedSignIns = count()
by bin(TimeGenerated, 1h)
| order by TimeGenerated ascThis example counts sign-in records with nonzero ResultType values, grouped hourly. Confirm the meaning of result codes and the source table for your environment before publishing it as a business metric.
Build the drill-down
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
SigninLogs
| where TimeGenerated > ago(24h)
| where ResultType != "0"
| summarize Failures = count(),
SourceIPs = dcount(IPAddress),
LastSeen = max(TimeGenerated)
by UserPrincipalName
| where Failures >= 5
| order by Failures descThe threshold of five is an illustrative reporting filter, not a Microsoft-recommended detection threshold. A ranked table should lead analysts to relevant detail, not imply every row is malicious.
Use parameters carefully
Time-range, account, device, application or workspace parameters can make one workbook useful across many investigations.
Choose sensible defaults and label filters clearly. Ensure the query logic respects the selected parameter and handles empty selections safely.
Time windows matter
A one-hour incident triage view and a 30-day trend view answer different questions. Make the selected period visible, and avoid comparing periods without accounting for different durations or ingestion delays.
Avoid confusing data coverage with zero activity
A blank chart might mean no events, no permissions, a missing connector, a query error or delayed ingestion. Show data freshness and document source dependencies where possible.
Plan for data access
Workbook readers need appropriate access to the workbook and the underlying data. Validate the experience with a representative analyst account rather than assuming the creator’s permissions match everyone else’s.
Think about performance
- Filter by time as early as possible.
- Summarise before visualising large event sets.
- Limit unnecessarily wide or high-cardinality result tables.
- Avoid repeating expensive queries across multiple visuals.
- Test the workbook with realistic data volumes and time ranges.
Do not turn dashboards into detections
A chart showing a spike is an investigation lead, not proof of compromise. If the behaviour requires timely automated detection, consider a properly tested analytics rule rather than relying on someone to notice a workbook.
A useful operational layout
Workbook design review
| Check | Question |
|---|---|
| Purpose | Can the user name the decision this view supports? |
| Definitions | Are all counts and filters explained? |
| Coverage | Are required connectors and tables present? |
| Freshness | Can users tell whether the data is current? |
| Performance | Do queries remain responsive at realistic scale? |
| Access | Can intended readers see the underlying data? |
| Action | Can analysts drill down and investigate? |
Agent Foskett exercise
Your SOC lead wants a workbook that helps analysts notice repeated identity failures and identify the accounts worth investigating.
- Write the primary decision the workbook should support.
- Choose three measures and define them precisely.
- Pick a visual for each measure and explain why.
- Run the two example KQL queries in a suitable environment.
- Describe one useful parameter and one data-quality warning.
- Sketch a drill-down from the trend chart to an individual account.
- Explain what evidence would be needed before calling an account compromised.
Common mistakes
- Adding charts before defining the investigation question.
- Presenting counts without definitions or time ranges.
- Using a pie chart when a time-series trend is needed.
- Hiding data gaps and permission limitations.
- Building expensive queries that nobody can use.
- Treating a dashboard anomaly as a confirmed attack.
Agent Foskett takeaway
A good workbook does not merely display what happened. It helps the analyst decide what to investigate next.
The logs already knew. The dashboard still needs to ask the right questions.
