Skip to main content
A barn swallow in flight with its wings spread, drawn in dots

SIEM migration checklist: inventory, rebuild, parallel run, cutover

A SIEM migration checklist in seven steps: inventory and prune rules, decide which data moves, rebuild detections by intent, run in parallel, then cut over.

Guide
Updated 9 min read

Summary

Migrate a SIEM in seven steps: inventory what the current platform actually detects, retire the rules nobody uses, decide which data has to move, rebuild the remaining detections by intent rather than syntax, run both platforms side by side until their alerts agree, plan where historical logs will live, and cut over one priority tier at a time. Most of the effort goes into the detection rebuild, and most of the risk sits in the cutover, so plan the parallel run and the rollback before the first source moves.

Key takeaways

  1. Export and audit every detection before planning the data move. Rules with no alerts in the past 6-12 months are candidates to retire, not to translate.
  2. Decide per source whether it must be ingested into the new platform, queried where it already lives, or archived. Moving every log by default is what makes migrations long and expensive.
  3. Rebuild each detection from what it was meant to catch, then replay it against real history before it goes live. Line-by-line query translation carries old noise into the new platform.
  4. Judge the parallel run on alert parity per Tier 1 use case, and keep the old SIEM able to take over again until the rollback window after cutover has passed.
  5. Set up the export of logs into storage you own at the start of the migration, because export features usually don't backfill data ingested before they were turned on.

The short version

A SIEM migration is a detection project with a data project attached. The logs are the visible part, but what the security team depends on is the set of rules that turn those logs into alerts, and those don't carry over between platforms. UltraViolet Cyber's migration guide puts the rewrite of rules, parsers, and correlation logic at 40 to 60 percent of the total project timeline.

The pressure to migrate is real for some teams. Palo Alto Networks ended IBM QRadar on Cloud and the QRadar cloud-native SIEM on April 14, 2026, and ends QRadar EDR and XDR on August 31, 2026, with no-cost migration to Cortex XSIAM for eligible customers. Others are moving because ingest-priced storage forces them to leave sources out; legacy SIEM vs modern SIEM covers that architecture shift, and Splunk alternatives compares the replacement options for Splunk customers.

Whatever the reason, the order below keeps detection coverage intact: audit and prune first, move only the data that has to move, rebuild what remains, prove it in parallel, then cut over.

Step by step

1. Inventory what the current SIEM actually does

Export every enabled detection with its query, schedule, threshold, and severity. In Splunk, a REST search over saved/searches and admin/macros returns enabled alert rules and the macros they call; Microsoft documents the exact query, and it needs a Splunk admin role to see every alert.

For each rule, record which log sources it reads, when it last fired, and what happened to its alerts. Then list the sources that feed no rule, dashboard, or report. Those are the first candidates for cheaper storage or for dropping.

2. Retire before you move

Microsoft's Splunk and QRadar migration guides both say not to migrate detections blindly and to review any rule that hasn't alerted in the past 6-12 months. Retire rules whose alerts analysts routinely close without action, and dashboards nobody opens.

Sort what's left into tiers. UltraViolet Cyber's framework uses three: Tier 1 must be live at cutover, Tier 2 follows within 30 days of cutover, and Tier 3 within 90 days or gets re-evaluated. Map Tier 1 to MITRE ATT&CK so the gaps you already had are visible instead of copied.

3. Decide which data has to move

Platforms differ on whether data must be centralized before a detection can use it. Microsoft Sentinel analytics rules require the data type to be present in the Log Analytics workspace before a rule is enabled, so every source a rule reads has to be connected and ingesting first.

Agent-native platforms don't require that for every source. A Cotool detection can call a connected tool's read-only actions directly, so it can check sign-ins in an existing SIEM or users in the identity provider without ingesting that tool's logs first. Cotool also lists an existing Splunk deployment as a source it queries in place. That changes the plan: ingest the sources that need fast search and long retention, and leave the rest where they live until there's a reason to move them. What security logs to store helps with that split.

4. Rebuild detections by intent

Translate what each rule was trying to catch, not its syntax. Microsoft's SIEM migration experience maps Splunk and QRadar detections to out-of-the-box Sentinel analytics rules and Defender XDR detections by intent; it runs on Security Copilot, which must be enabled in the tenant, and it doesn't install connectors or enable rules for you. Rules without a match are rebuilt by hand in KQL.

In Cotool, detections are Python notebooks rather than a proprietary query language, and agents write, test, and tune them. Bring existing rules over, and Cotool replays every new detection against your history before it deploys, so you see what it would have fired on. Cotool also maintains more than 200 signal detections across more than 40 log sources, added when you enable a log stream and backfilled over the last 14 days, which covers single-event rules you'd otherwise rewrite one by one. Detection as code covers the review and versioning side.

5. Run both platforms and compare alerts

Run both SIEMs on the same data for a defined period and compare alert output per Tier 1 use case: what fired in both, what fired only in the old one, and what's new. Don't shut down the old SIEM until Tier 1 use cases are validated in production, not only in testing.

Alert counts alone hide disagreements. Compare dispositions too. Cotool investigates every alert automatically, so during a parallel run each of its alerts arrives with an investigation and a verdict you can line up against how your analysts closed the matching alert in the old queue. Cotool's autotuning then proposes targeted changes from false-positive and benign outcomes and evaluates each candidate against the current logic, tracking labeled true positives retained, before it's published.

6. Plan for historical data and an exit

Decide early where history lives, because exports take longer than expected. Splunk's CLI export returns 100 events by default; set -maxout 0 for an unlimited export, and use dump or the CLI for large volumes. Microsoft recommends the Sentinel data lake as the target for migrated history, with up to 12 years of retention in Parquet.

Plan the next exit at the same time. Cotool's logging sinks mirror newly ingested logs to Amazon S3, Google Cloud Storage, Azure Blob Storage, or S3-compatible storage, as normalized Parquet or JSON Lines or as raw gzipped JSON Lines. Sinks don't backfill, so create one when the first source starts ingesting, not after cutover. Cotool can also keep log data in your own cloud bucket and search it in place; see what a security data lake is for that model.

7. Cut over one tier at a time

Set go/no-go criteria per tier before picking a date. Aim for the contract renewal, but don't let it force a cutover before Tier 1 is validated. Write the rollback plan for critical issues in the first 24 to 48 hours after cutover: who decides, and how the old SIEM resumes alerting. Watch the new platform closely for the first 30 days, including source health; Cotool raises an alert when a source goes quiet, so a broken pipeline after cutover doesn't become a blind spot.

What vendor migration tools do and don't do

Vendor tooling covers the mapping, not the decisions. Microsoft's SIEM migration experience produces recommendations grouped by Content Hub solution and a report of rules with no good match, but someone still connects data sources, enables each rule, and tests it; the documented limitation is that it maps to out-of-the-box Sentinel connectors and rules. Palo Alto Networks' QRadar offer is a services engagement toward Cortex XSIAM. Either way, the tool targets its own platform and query language, so the inventory, pruning, and tiering in steps 1 and 2 stay with your team.

How Cotool approaches a SIEM migration

Cotool is built to run alongside an existing SIEM: bring the existing rules, move sources over at your own pace, and keep detecting on the ones you haven't moved, because detections reach connected tools and federated sources directly. Every alert is investigated by agents, so the parallel run compares verdicts as well as counts. Detections are Python you can keep in your own Git repository and review in pull requests, and logs can stay in your own storage, so the next migration starts from files and code you already hold.

Common mistakes

  • Translating every rule. Porting unused or noisy rules spends the largest part of the project on content that never earned its place.
  • Moving data before deciding what reads it. Ingesting every source into the new platform by default raises cost before the first detection runs.
  • Testing translated rules only on synthetic data. Replay against your own history, or the first week of production becomes the test.
  • Letting the renewal date decide. A renewal date is a target, not a go/no-go criterion; validated Tier 1 use cases are.
  • Exporting history with default limits. A Splunk CLI export without -maxout stops at 100 events.
  • Turning on log export late. Export features that don't backfill leave a gap for everything ingested before they were enabled.

Checklist

  1. Export all enabled detections, macros, and lookups with owners and last-fired dates.
  2. List every log source with its daily volume and the rules, dashboards, and reports that use it.
  3. Retire rules with no useful alerts in 6-12 months and dashboards nobody opens.
  4. Assign the rest to Tier 1, 2, or 3 and map Tier 1 to MITRE ATT&CK.
  5. Mark each source: ingest, query in place, archive, or drop.
  6. Rebuild Tier 1 detections by intent and replay each against real history.
  7. Set up log export to storage you own before cutover.
  8. Run both platforms on the same data and compare alerts and dispositions per Tier 1 use case.
  9. Write go/no-go criteria and a rollback plan for the first 24 to 48 hours.
  10. Cut over Tier 1, then move Tier 2 within 30 days and Tier 3 within 90 days.
  11. Monitor source health and alert volume closely for 30 days after cutover.
  12. Keep the old SIEM readable until its retention obligations end.

FAQ

How long does a SIEM migration take?
A SIEM migration's length depends mostly on how many detections have to be rebuilt, because rewriting rules, parsers, and correlation logic typically takes 40 to 60 percent of the project timeline in UltraViolet Cyber's field guide. Pruning unused rules first and querying some sources in place instead of moving them both shorten it. Plan for a parallel run and at least 30 days of closer monitoring after cutover.
Can I translate Splunk SPL rules directly into another query language?
Splunk SPL rules can be translated, but a direct translation copies each rule's thresholds, exceptions, and noise into the new platform. Microsoft's own Splunk guidance says not to migrate detections blindly and to review rules that haven't alerted in 6-12 months. Rebuild each rule from the behavior it was meant to catch, then test it against your own data.
Do I need to move all my historical logs during a SIEM migration?
Historical logs don't all need to move into the new SIEM's search tier during a migration. Keep what compliance and investigations require in low-cost storage, keep the old SIEM readable until its retention runs out, or export to object storage. If you export from Splunk with the CLI, note that it returns 100 events by default unless you set maxout.
What happens to IBM QRadar SaaS customers?
IBM QRadar SaaS products reached end of life on April 14, 2026 (QRadar on Cloud and the cloud-native SIEM) or reach it on August 31, 2026 (QRadar EDR, XDR, and related services), according to Palo Alto Networks, which acquired them. Palo Alto Networks offers eligible customers no-cost migration to Cortex XSIAM. The announcement doesn't affect on-premises QRadar.

Sources

  1. Use the SIEM Migration Experience - Microsoft Sentinel · accessed November 22, 2026
  2. Migrate Splunk detection rules to Microsoft Sentinel · accessed November 22, 2026
  3. Migrate QRadar Detection Rules to Microsoft Sentinel · accessed November 22, 2026
  4. Export Splunk Data to Target Platform - Microsoft Sentinel · accessed November 22, 2026
  5. Microsoft Sentinel Migration: Select a Target Microsoft Platform · accessed November 22, 2026
  6. End-of-Sale Announcement (QRadar SaaS products) - Palo Alto Networks · accessed November 22, 2026
  7. The Complete SIEM Migration Checklist (UltraViolet Cyber) · accessed November 22, 2026
  8. AI-Native SIEM Built for Agents - Cotool · accessed November 22, 2026
  9. AI Detection Engineering with Python Detections - Cotool · accessed November 22, 2026
  10. Detections Platform - Cotool Documentation · accessed November 22, 2026
  11. Signals and Signal Detections - Cotool Documentation · accessed November 22, 2026
  12. Export Logs to Your Storage - Cotool Documentation · accessed November 22, 2026

Keep reading

Attackers are scaling with tokens

Cotool helps defenders operate at machine speed. See how security teams are scaling Detection & Response.

Book a demo