By Michael Henry, CEO, Accelerynt

In every engagement, Accelerynt looks for ways to improve the customer’s security operations, and during an initial review of a Microsoft Sentinel queue we watched one type of alert come back again and again. A sign-in would match a Conditional Access exception the organization had approved, and an alert would fire. A skilled analyst on the outsourced L1/L2 team would gather the evidence, confirm the exception was fine, and close the alert, and then the next one would arrive.

To us the work created value for one party only, the outsourced operations team that was paid to do it. It was simple, linear, and repetitive, which is our test for what a machine should do. So we built from one design principle: automate everything upstream of the human decision, keep the decision with a named person, and have Sentinel record every action.

MSSP Alert named Accelerynt one of this year’s Top 10 Most Innovative MSSPs for that work, and the case study is Routine Alerts vs. Human Decisions. The principle is the innovation.

The Alert-Volume Problem Was an Evidence Problem

That customer knew what its environment was supposed to look like, but the knowledge sat outside anything a system could check, so every deviation looked new and every triage started from zero. Once we wrote down the intended state and automated the comparison, the work fell off along with the noise.

Other teams carry the same burden. Microsoft’s Digital Defense Report this year says experienced analysts “could be spending more time reconstructing context than disrupting adversary activity,”1 and CrowdStrike’s 2026 threat report shows why the job keeps getting harder: 82% of detections in 2025 were malware-free, up from 51% in 2020.2 Attackers work through valid credentials and trusted tools, and defenders are left to read intent in actions that look normal.

Automate What Is Simple, Linear, and Repetitive

Work that passes the test sits upstream of the decision. For this customer that meant working out what the environment was supposed to look like, gathering the evidence an analyst would otherwise collect by hand, closing the alerts that matched a known-good state someone had approved, and adding context to whatever was left, so that the analyst opens a case with the evidence in hand. All of it now happens before a case reaches a person, which is the SOC’s version of shifting left.

Some responses can run on their own too. Blocking an address or revoking a session is a decision the team can make once, in advance, for a given detection, and from then on a named person has made the call and the automation carries it out. The open source identity playbooks we publish work that way.

Keep the Judgment With a Person

Judgment is different. Auto-remediation and auto-rollback hand it to a machine, and that adds risk where a team can least afford it. An unauthorized change worth reverting is an incident worth investigating, and when a machine reverts the change on its own the alert closes and the investigation never starts. The Conditional Access exception shows the other half of the risk, because a change that looks unauthorized may be one the organization approved.

It Runs on What the Customer Owns

The automations run natively in the customer’s own Sentinel workspace. The customer kept the platform they had, added zero agents, and kept their log data in their own environment. Their playbook library has grown to about 30 automations, each mapped to how that organization operates, and because the automations and the records they produce belong to the customer, the results are durable and auditable. For a security leader, that changes where to look for the next improvement, which may be sitting in the platform the team runs today.

What Improved

The number of alerts an analyst had to touch fell 75%, because automation now handles the commodity, low-risk alerts before a person sees them, and the time went back to investigation and engineering, the higher-risk work that needs judgment. Time to resolve fell as well, because the evidence that used to take longest to gather is there when the case opens. The customer’s own team gained too. They build expertise working alongside our engineers, and they can pull the record of what an automation did, when it ran, and who authorized it from their own SIEM, which is the record auditors and insurers ask for.

Our industry still reports activity, such as alerts triaged and tickets closed, and those numbers measure volume. Microsoft’s report makes the same point, that rule counts and alert counts have stopped being a useful measure of SOC maturity.1

Count the Alerts That Needed a Decision

Take a week of alerts and sort them into two groups. The first group turned out to be the company’s own activity, and it can be automated once the intended state is written down. The second group needed someone to decide something, and the analysts’ hours belong there.

The teams that pull ahead will automate the routine and put a named person on every change. The machine gathers the evidence, and a person makes the call.

Sources

1. Microsoft. (2026, October). 2026 Microsoft Digital Defense Report (p. 54). https://microsoft.com/mddr

2. CrowdStrike. (2026, February). 2026 Global Threat Report (p. 9). https://www.crowdstrike.com/en-us/global-threat-report/

← Previous

Security Leaders: How Do You Measure Readiness?