Come, take your chai and think about what a junior SOC analyst actually does with a phishing alert. Honestly, most of the first fifteen minutes is copying. Copy the sender’s address into a reputation site. Copy the link into another tool. Check whether the attachment hash is known. Search the mail system to see who else received it. Paste everything into the ticket. Only then does the thinking start.
SOAR, which stands for security orchestration, automation and response, is the idea that a machine should do that copying. Let me explain the three words, because they are different things.
Orchestration, automation, response
Orchestration means connecting your tools so they work as one flow: the SIEM, the email gateway, the endpoint agent, the firewall, the ticketing system, the threat intelligence service. Each has an API, and the platform calls them in sequence.
Automation means doing the individual steps without a person: look up the hash, query the mailbox, isolate a laptop.
Response is the action end: block a sender, disable an account, quarantine a message, open an incident.
A playbook is the written logic that strings them together, usually drawn as a flow with branches. “When a user reports a suspicious email: extract links and attachments, check reputation, search all mailboxes for the same message, if verdict is malicious then pull it from inboxes and block the sender, else close with a note to the user.”
A worked example with small numbers
Suppose a mid-sized company in Nagpur gets 60 user-reported phishing emails a day. Manual handling takes about 20 minutes each, so that is 1,200 minutes, or 20 analyst-hours daily. That is more than two analysts doing nothing else. If a playbook handles the lookups and the mailbox search automatically and a human only reviews the 10 that come out as suspicious or uncertain, at 10 minutes each, the human workload drops to under two hours. These figures are invented for illustration, but the shape is typical: the saving comes from removing repetitive enrichment, not from replacing judgement.
There is a second gain that people forget. Speed. A malicious email delivered to 200 inboxes does its damage in the first hour. A playbook that pulls it from every mailbox within two minutes of the first report is worth far more than a faster analyst.
Good candidates for automation
- Enrichment: attach reputation, geolocation, asset owner, user department, recent logins to every alert before a human sees it.
- Deduplication and grouping: twenty alerts for the same host become one case.
- Containment of high-confidence, low-regret actions: quarantine a known-bad file, block a known-malicious domain.
- Housekeeping: open the ticket, assign it, notify the right channel, record timings.
Things to keep a human in front of
Disabling the CFO’s account, isolating a production server, blocking a whole IP range used by a partner. These have a high cost if the automation is wrong. A sensible pattern is the approval gate: the playbook prepares everything, shows the evidence, and waits for one click from an authorised person. Over time, as a particular action proves reliable, you may relax the gate, but that should be a deliberate decision recorded somewhere.
How it goes wrong
Automating a bad process just makes the bad process faster. If your detection rules produce 90 per cent false positives, a playbook that auto-blocks on those alerts will block real business traffic, repeatedly. Fix the detection first, or restrict automation to enrichment until the verdicts are trustworthy.
Playbooks also rot. The email gateway changes its API, someone renames a field, a vendor retires an integration, and the flow quietly fails at step four. Treat playbooks like software: version them, test them, assign owners, monitor failures, and make failed automations raise a visible alert instead of silently doing nothing.
The platform itself holds powerful credentials, since it must be able to disable accounts and change firewall rules. That makes it an attractive target. Give each integration the narrowest service account that works, protect the platform with strong authentication, and log every action it takes.
Finally, there is a skills risk. If analysts only ever see pre-enriched cases, they may lose the habit of understanding what is behind the verdict. Rotate people through writing and reviewing playbooks so the logic stays understood, not just trusted.
Starting small
You do not need to buy an expensive platform to learn this. Pick one alert type with high volume and a stable manual routine. Write the steps on paper exactly as an analyst performs them today. Mark which steps only gather information, and automate those first, using whatever your SIEM or ticketing tool supports. Measure the time per case before and after. Only then add containment actions behind an approval gate.
Many security tools now bundle simple automation, so you may already own part of what you need. And note that newer AI-assisted features in security products can summarise alerts or suggest next steps; treat their output like a junior colleague’s draft, checked before action, particularly when the action is irreversible.
Figures and names in the examples are illustrative.
Leave a Reply