Threat Hunting – Art or Science?

Raj Samani, Head of Strategic Intelligence, McAfee LLC provides insights on some of the characteristics of good threat hunters using McAfee’s recent report.

Raj Samani, Head of Strategic Intelligence, McAfee

Security professionals are in a fight every day to track down criminals who would disrupt their organization. Attackers nearly always have the element of surprise in their favor, but threat hunting can throw the attackers off their footing.

So what are the characteristics of good threat hunters? We recently surveyed more than 700 IT and security professionals, to identify insights and lessons for organizations looking to understand and enhance their threat hunting capabilities. One of the key questions was the level of maturity of the organization’s threat hunting activity. Ranging from Level 0, where the organizations rely primarily on automated alerting (i.e. little or no routine data collection) and typical tools include IDS, SIEM and Anti-virus, to Level 4, where organizations automate the majority of successful data analysis procedures and use high or very high levels of routine data collection, these self-reported assessments provide useful insight into the current nature of the threat hunt and reveal some surprises about how organizations are investing for future improvement.

Some of the key findings include:

  • The most mature threat hunting organizations are twice as likely to automate parts of the investigation process, spend 50% more of their time actually hunting, and as a result 70% of them are closing investigations in a week or less, compared to only 50% of the less-mature organizations.
  • Mature organizations are three times more likely to consider every level of the identification and investigation processes as viable for automation, especially sandboxing, endpoint detection and response, and user behavior analysis.
  • Tool emphasis changes with experience. Sandboxing was the number one tool for Tier 1 and 2 analysts of all sizes and maturity levels, but Tier 3 and 4 analysts use sandboxing as part of a broader mix of tools.
  • Immature companies are trying to use the same tools as the most mature companies, but without the same results. Adopting new tools without changing the processes for hunting and incident response is rarely successful, as success requires an upfront investment in architecture and optimized processes.
  • Threat hunters in mature SOCs spend 70% more time on customization of tools and techniques. Custom scripts and Security Information and Event Management (SIEM) are heavily used to automate manual and ad hoc processes.

Observe, Orient, Decide, and Act

Human decision-making can be the critical advantage in many security scenarios, tilting the playing field in your favor. U.S. Air Force Colonel John Boyd first documented the four fundamental parts of this process, which are Observe, Orient, Decide, and Act (OODA).

Effective security operations teams are leveraging this process to exploit their adversaries’ weaknesses, supported by automated processes, machine-driven analytics, and curated threat intelligence. Threat hunters often begin with the assumption of a breach or compromise, following clues and personal intuition, and later turning successful hunts into automated rules. Hunting is a human-centric activity, using a wide range of tools and information to seek out hidden threats to the organization.

Based on the survey results, threat hunting begins as an ad hoc process in the least-mature organizations, then swings strongly towards process development before eventually finding an appropriate balance between process and ad hoc in the most mature hunters. Immature organizations tend to aggressively give their hunters sophisticated tools and data, with limited success. As they mature, hunters refine their processes and hunting techniques, adding automation and analytics to help manage the vast amounts of security data. By Level 4, hunters have significantly increased their effectiveness as they selectively use tools and data appropriate to their environment and likely attack vectors.

 As a case in point, our survey revealed that at Level 1, only 40% of processes are automated, compared with more than 70% by Level 4. This embrace of automation, combined with effective and skilled identification of patterns of anomalous behavior, results in a synergy between hunting and incident response that delivers faster triage, shorter case closure times, and a much higher percentage of root-cause determination. Our survey showed that more than 70% of mature SOCs closed cases in less than 7 days, compared to 25 days for the least mature ones, and determined root cause 70% of the time, compared to just 43% for least mature ones.


Threat hunters are using a wide range of tools and techniques to find, contain, and remediate cyberattacks. As they mature in the role, their effectiveness increases as they are augmented by human-machine teaming, combining human judgment and intuition with machine speed and pattern recognition.

One of the key characteristics of mature hunters is the way they leverage automation to improve manual steps in the process, customize scripts for their environment, and quickly test new ideas. In mature environments, leading hunters make use of a wide variety of tools and data sources, continuously updating and improving them and generating a positive OODA loop.

For less mature organizations, copying the tools and techniques of the leading hunters is not sufficient. Adding new tools without changing the OODA cycle is unlikely to produce positive results. Sandboxing, automation, and analytics can empower these less-experienced hunters, but organizations that have not invested in architecture and defined processes that support that automation will experience diminished results.

Threat hunting is here to stay, and is no longer an esoteric practice limited to a few of the edgier practitioners. Over the next few years, expect to see threat hunting as part of most organizations’ analytics driven security operations, backed by extensive automation and machine analytics.