Agent Foskett Academy • Microsoft Sentinel • Module 4 • Lesson 45

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.

Module 4 finale — investigate the evidence, not just the alert.
Agent Foskett Academy complete Microsoft Sentinel investigation
What you will learn

Work an incident end to end.

✓ Triage and prioritise
✓ Investigate alerts and entities
✓ Reconstruct the timeline
✓ Document, respond and close

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

1. TRIAGE the incident ↓ 2. READ the alerts and timestamps ↓ 3. PIVOT across users, devices and IPs ↓ 4. QUERY underlying telemetry with KQL ↓ 5. RECONSTRUCT the timeline ↓ 6. PRESERVE evidence and notes ↓ 7. SCOPE and coordinate response ↓ 8. VERIFY and decide closure

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)DetectionQuestion to investigate
02:08Suspicious mailbox activityWho changed the mailbox and from where?
02:14Suspicious PowerShellWhich account, parent process and device?
02:17Suspicious network connectionWhich process initiated the connection?

Alert times do not automatically equal event times. Validate the timestamps in source telemetry.

Step 4 — Pivot through the entities

alex@contoso.com │ ├── 203.0.113.50 (sign-in IP) │ └── possible related activity │ └── DEVICE-27 (endpoint) ├── PowerShell process └── network destination 198.51.100.20

These IPs are documentation-only examples. A shared entity is a lead, not proof of a shared attacker.

Step 5 — Query the sign-ins

Identity pivot — sign-in evidence
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 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 asc

Confirm 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

Endpoint pivot — process evidence
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 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 asc

DeviceProcessEvents 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

Endpoint pivot — network evidence
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 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 asc

Correlate 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)EventEvidence / status
01:57Successful sign-in from unfamiliar IPIdentity telemetry; verify context
02:08Mailbox rule activityAlert plus audit evidence needed
02:14PowerShell on DEVICE-27Endpoint process evidence
02:17External connection from DEVICE-27Network event; attribution unverified
02:31Potential follow-on sign-inCheck 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

[Observed] 01:57 UTC: successful sign-in for alex@contoso.com from 203.0.113.50. [Hypothesis] May precede mailbox manipulation. The link is not yet established. [Next step] Compare mailbox audit actor, session context and affected account. [Response] No containment claimed until its execution and result are verified.

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

QuestionClosure 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

Incident: Fictional Sentinel investigation Initial concern: identity + mailbox + endpoint Confirmed: relevant sign-in and process events Not yet established: single attacker across events Scope: affected account and endpoint reviewed; additional entity pivots documented Actions: only verified actions recorded Disposition: hold / escalate if material questions remain unanswered Follow-up: assigned owners and monitoring

Hands-on challenge

  1. Identify the earliest observed activity, not the earliest alert.
  2. Choose two entities and explain a targeted pivot for each.
  3. Run the sign-in query in an authorised test environment.
  4. Identify one gap between alert and event chronology.
  5. Write one observed fact and one hypothesis.
  6. Describe a response decision that would require approval.
  7. 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.

Module 4 complete
You have followed an incident from initial triage through alert and entity analysis, KQL, timeline reconstruction, evidence preservation, response and a defensible disposition.
Sentinel Academy Home