Lesson 45 — The Complete Microsoft Sentinel Investigation
One incident. Several alerts. Multiple entities.
Now put everything together — from first triage to a defensible investigation outcome.

What you will learn
Work an incident end to end.
Learning objectives
- Triage an incident without treating its severity as proof.
- Investigate alerts, entities and related telemetry using targeted KQL.
- Reconstruct an evidence-based timeline and determine the scope.
- Preserve findings, document uncertainty and choose proportionate response actions.
- Write a defensible final disposition and handover.
The Module 4 finale
Lessons 36–44 covered each part of the workflow. Now you will work one fictional case from the first incident notification to the final analyst decision.
There is no magic button that performs an investigation for you.
The case: a quiet Monday morning
At 02:14 UTC, Microsoft Sentinel opens a high-severity incident involving alex@contoso.com and DEVICE-27. Alerts reference suspicious sign-in activity, mailbox changes, PowerShell and an external connection.
The incident looks serious. But severity is only a starting signal; the evidence must establish what actually occurred.
The investigation map
Step 1 — Read before acting
- Record incident ID, owner, severity, status and creation time.
- Identify the earliest and latest underlying activity, not merely the first alert.
- List mapped entities and any obvious gaps.
- Check whether containment or automation has already changed the case.
Step 2 — Ask the right priority question
A high-severity alert on an ordinary workstation may differ from the same activity involving a privileged account or business-critical server. Consider privilege, asset value, evidence confidence, business impact and possible spread.
Step 3 — Review the alerts as leads
| Alert time (illustrative) | Detection | Question to investigate |
|---|---|---|
| 02:08 | Suspicious mailbox activity | Who changed the mailbox and from where? |
| 02:14 | Suspicious PowerShell | Which account, parent process and device? |
| 02:17 | Suspicious network connection | Which process initiated the connection? |
Alert times do not automatically equal event times. Validate the timestamps in source telemetry.
Step 4 — Pivot through the entities
These IPs are documentation-only examples. A shared entity is a lead, not proof of a shared attacker.
Step 5 — Query the sign-ins
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
SigninLogs
| where UserPrincipalName =~ "alex@contoso.com"
| where TimeGenerated between (
datetime(2026-10-07 01:30:00) ..
datetime(2026-10-07 03:00:00)
)
| project TimeGenerated, UserPrincipalName, IPAddress,
AppDisplayName, ResultType, ConditionalAccessStatus
| order by TimeGenerated ascConfirm whether the account signed in successfully, from which addresses, and whether the activity preceded the mailbox alert. Verify data availability in the connected workspace.
Step 6 — Investigate endpoint execution
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
DeviceProcessEvents
| where DeviceName =~ "DEVICE-27"
| where Timestamp between (
datetime(2026-10-07 01:30:00) ..
datetime(2026-10-07 03:00:00)
)
| where FileName in~ ("powershell.exe", "pwsh.exe")
| project Timestamp, DeviceName, AccountName, FileName,
ProcessCommandLine, InitiatingProcessFileName
| order by Timestamp ascDeviceProcessEvents is Microsoft Defender XDR advanced hunting telemetry. Run this query in the appropriate Defender hunting environment, or use an available equivalent table in your connected Sentinel workspace. Do not assume every Sentinel workspace contains this table.
Step 7 — Test the network relationship
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
DeviceNetworkEvents
| where DeviceName =~ "DEVICE-27"
| where Timestamp between (
datetime(2026-10-07 02:00:00) ..
datetime(2026-10-07 03:00:00)
)
| project Timestamp, DeviceName, RemoteIP, RemoteUrl,
InitiatingProcessFileName, InitiatingProcessCommandLine
| order by Timestamp ascCorrelate process, account, device and network timestamps. A connection alone does not establish data exfiltration or attacker control.
Step 8 — Reconstruct the activity timeline
| Time (UTC) | Event | Evidence / status |
|---|---|---|
| 01:57 | Successful sign-in from unfamiliar IP | Identity telemetry; verify context |
| 02:08 | Mailbox rule activity | Alert plus audit evidence needed |
| 02:14 | PowerShell on DEVICE-27 | Endpoint process evidence |
| 02:17 | External connection from DEVICE-27 | Network event; attribution unverified |
| 02:31 | Potential follow-on sign-in | Check for continued activity |
This is a teaching scenario, not a confirmed real-world case. Preserve event timestamps and clearly distinguish observed activity from analyst interpretation.
Step 9 — Challenge the initial story
- Was the sign-in genuinely unauthorised or merely unfamiliar?
- Is the mailbox change tied to the same authenticated session?
- Was PowerShell run by the same identity or a different local process?
- Is the destination actually malicious, and what did the connection do?
- Are there other affected users, devices or resources?
- What evidence contradicts the current hypothesis?
Step 10 — Preserve important findings
Bookmark or otherwise preserve the 01:57 sign-in and any other relevant hunting result using supported tooling. Record the query, time window, entity, source, reason for relevance and unresolved questions.
Avoid saving every result simply because it appeared in a query.
Step 11 — Write the analyst record
Step 12 — Determine scope
Search for additional accounts using the same source IP, other devices associated with the identity, related processes, and subsequent network activity. Expand the investigation window only when evidence warrants it.
The incident originally named one device. That does not prove only one device was involved.
Step 13 — Coordinate response
- Escalate according to the organisation’s severity and authorisation procedures.
- If compromise is supported, coordinate proportionate containment with the incident owner.
- Record who approved each action, what was done and whether it succeeded.
- Continue monitoring for recurrence or newly identified entities.
Step 14 — Decide whether the case is ready to close
| Question | Closure requirement |
|---|---|
| What happened? | Evidence-supported conclusion or documented limitation |
| How far did it spread? | Reasonable scope assessment |
| What was done? | Verified actions and owners |
| What remains? | Residual risk and follow-up assigned |
| Why this classification? | Explicit justification in case notes |
Do not equate a quiet dashboard with verified containment. If material risk remains, continue or escalate the case.
Possible outcomes
- True positive: evidence supports an actual security event; response and residual risk are documented.
- Benign positive: detected behaviour occurred but is established as authorised or benign.
- False positive: detection logic or interpretation incorrectly indicated the concerning behaviour.
- Unresolved: evidence remains insufficient; use the appropriate escalation or case process.
Example final case summary
Hands-on challenge
- Identify the earliest observed activity, not the earliest alert.
- Choose two entities and explain a targeted pivot for each.
- Run the sign-in query in an authorised test environment.
- Identify one gap between alert and event chronology.
- Write one observed fact and one hypothesis.
- Describe a response decision that would require approval.
- Draft a closure recommendation that accounts for unresolved questions.
What the investigation taught us
An incident is a case, alerts are leads, entities are pivots, KQL interrogates telemetry, timelines order the evidence, bookmarks preserve important discoveries, and notes make reasoning reproducible.
The goal is not to close quickly. It is to reach a conclusion the evidence can defend.
Agent Foskett takeaway
Read the incident. Follow the entities. Challenge the timeline. Verify the response. Then make the call.
The logs already knew. The analyst had to ask the right questions.
