Information security metrics have a bad rep. Mention metrics to a CISO and immediately his thoughts may well turn to sigmas, standard deviations and, probably, probability. To many, metrics equals statistics.
There's no denying that proven economic principles can, and should, be applied to information security investments. At the same time, a bumper crop of valuable metrics exist that don't require classes on Nobel Prize-winning theories or a working knowledge of the Greek alphabet. You've actually already sowed the seeds of these less dense but equally valuable metrics. They're sitting in your log files, on your network, in the brains of your business unit managers, just waiting to be harvested. You won't need computational prowess to exploit this crop's value, just some legwork and—this is key—the most effective presentation tools.
See the best of CSO's security metrics coverage in our Security metrics: Critical issues updated roundup
Useful Books on Security Metrics
By Andrew Jaquith (Addison-Wesley Professional, 2007)
by George Campbell (Security Executive Council)
by Debra Herrmann (Auerbach, 2007)
Jaquith has sharp, sometimes contrarian opinions on what makes a good metric and what makes for good presentation of metrics. For example, he thinks annual loss expectancy (ALE), a tool used to measure potential losses against probability of losses occurring over time, is useless, because in infosecurity, the L and the E in ALE are wild guesses. Quoting Geer, he says, "The numbers are too poor even to lie with."
He also thinks CISOs are too apt to dumb down visual representations of metrics for their executive counterparts, mistaking simplicity for clarity. He holds a particular grudge against the overuse of the "red, yellow, green" representation of metrics to signify high, medium and low numbers. "A CEO's favorite visualization of metrics is a stock chart, a 1-inch square that contains a month's worth of opening and closing prices, a trend line and several other indicators. Maybe 50 or more data points right there. Don't tell me they can't handle complex data. They can, as long as it's presented well."
By no means does Jaquith (or CSO for that matter) think these five metrics are the final word on infosecurity. Quite the contrary, they're a starting point, relatively easy to ascertain and hopefully smart enough to get CISOs thinking about finding other metrics like these, out in the vast fields of data, waiting to be reaped.
Metric 1: Baseline Defenses Coverage (Antivirus, Antispyware, Firewall, and so on)
This is a measurement of how well you are protecting your enterprise against the most basic information security threats. Your coverage of devices by these security tools should be in the range of 94 percent to 98 percent. Less than 90 percent coverage may be cause for concern. You can repeat the network scan at regular intervals to see if coverage is slipping or holding steady. If in one quarter you've got 96 percent antivirus coverage, and it's 91 percent two quarters later, you may need more formalized protocols for introducing devices to the network or a better way to introduce defenses to devices. In some cases, a drop may stir you to think about working with IT to centralize and unify the process by which devices and security software are introduced to the network. An added benefit: By looking at security coverage, you're also auditing your network and most likely discovering devices the network doesn't know about. "At any given time, your network management software doesn't know about 30 percent of the IP addresses on your network," says Jaquith, because either they were brought online ad hoc or they're transient.
How to get it: Run network scans and canvass departments to find as many devices and their network IP addresses as you can. Then check those devices' IP addresses against the IP addresses in the log files of your antivirus, antispyware, IDS, firewall and other security products to find out how many IP addresses aren't covered by your basic defenses.
Expressed as: Usually a percentage. (For example, 88 percent coverage of devices by antivirus software, 71 percent coverage of devices by antispyware and so forth.)
Not good for: Shouldn't be used for answering the question "How secure am I?" Maximum coverage, while an important baseline, is too narrow in scope to give any sort of overall idea of your security profile. Also, probably not yet ready to include cell phones, BlackBerrys and other personal devices, because those devices are often transient and not always the property of the company, even if they connect to the company.
Try these advanced versions: You can parse coverage percentages according to several secondary variables. For example, percentage coverage by class of device (for instance, 98 percent antivirus coverage of desktops, 87 percent of servers) or by business unit or geography (for instance, 92 percent antispyware coverage of desktops in operations, 83 percent of desktops in marketing) will help uncover tendencies of certain types of infrastructure, people or offices to miss security coverage. In addition, it's a good idea to add a time variable: Average age of antivirus definitions (or antispyware or firewall rules and so on). That is, 98 percent antivirus coverage of manufacturing servers is useless if the average age of the virus definitions on manufacturing's servers is 335 days. A star company, Jaquith says, will have 95 percent of their desktops covered by antivirus software with virus definitions less than three days old.
One possible visualization: Baseline defenses can be effectively presented with a "you are here" (YAH) graphic. A YAH needs a benchmark—in this case it's the company's overall coverage. After that, a business unit, geography or other variable can be plotted against the benchmark. This creates an easy-to-see graph of who or what is close to "normal" and will suggest where most attention needs to go. YAHs are an essential benchmarking tool. The word "you" should appear many times on one graphic. Remember, executives aren't scared of complexity as long as it's clear. Here's an example: plotting the percentages of five business units' antivirus and antispyware coverage and the time of their last update against a companywide benchmark.
Metric 2: Patch Latency
Patch latency is the time between a patch's release and your successful deployment of that patch. This is an indicator of a company's patching discipline and ability to react to exploits, "especially in widely distributed companies with many business units," according to Jaquith. As with basic coverage metrics, patch latency stats may show machines with lots of missing patches or machines with outdated patches, which might point to the need for centralized patch management or process improvements. At any rate, through accurate patch latency mapping, you can discover the proverbial low-hanging fruit by identifying the machines that might be the most vulnerable to attack.
How to get it: Run a patch management scan on all devices to discover which patches are missing from each machine. Cross-reference those missing patches with a patch clearinghouse service and obtain data on 1. the criticality of each missing patch and 2. when the patches were introduced, to determine how long each missing patch has been available.
Expressed as: Averages. (For example, servers averaged four missing patches per machine. Missing patches on desktops were on average 25 days old.)
Not good for: Companies in the middle of regression testing of patch packages, such as the ones Microsoft releases one Tuesday every month. You should wait to measure patch latency until after regression testing is done and take into account the time testing requires when plotting the information. The metrics might also get skewed by mission-critical systems that have low exposure to the outside world and run so well that you don't patch them for fear of disrupting ops. "There are lots of systems not really open to attack where you say, 'It runs, don't touch it,'" says Jaquith. "You'll have to make a value judgment [on patch latency] in those cases."
[Don't miss 68 Great Ideas for Running a Security Department, our exclusive collection of practical wisdom on security metrics and so much more] [7pp PDF - free Insider registration required
Try these advanced metrics: As with baseline coverage, you can analyze patch latency by business unit, geography or class of device. Another interesting way to look at patch latency statistics is to match your average latency to the average latency of exploits. Say your production servers average 36 days on missing patches' latency, but similar exploits were launched an average of 22 days after a patch was made available. Well, then you have a problem. One other potentially useful way to approach patch latency is to map a patch to its percent coverage over time. Take any important patch and determine its coverage across your network after one day, three days, five days, 10 days and so on.
One possible visualization: For data where you can sum up the results, such as total number of missing patches, a "small multiples" graphic works well. With small multiples you present the overall findings (the whole) as a bar to the left. To the right, you place bars that are pieces making up the whole bar on the left. This presentation will downplay the overall findings in favor of the individual pieces. One key in small multiples graphing is to keep the scale consistent between the whole and the parts. This example plots total number of missing patches for the top and bottom quartiles of devices (the best and worst performers). Then it breaks down by business unit who's contributing to the missing patches.
Metric 3: Password Strength
This metric offers simple risk reduction by sifting out bad passwords and making them harder to break, and finding potential weak spots where key systems use default passwords. Password cracking can also be a powerful demonstration tool with executives who themselves have weak passwords. By demonstrating to them in person how quickly you can break their password, you will improve your lines of communication with them and their understanding of your role.
How to get it: Using commonly available password cracking programs, attempt to break into systems with weak passwords. Go about this methodically, first attacking desktops, then servers or admin systems. Or go by business unit. You should classify your devices and spend more time attempting to break the passwords to the more important systems. "If it's a game of capture the flag," Jaquith says, "the flag is with the domain controller, so you want stronger access control there, obviously."
Expressed as: Length of time or average length of time required to break passwords. (For example, admin systems averaged 12 hours to crack.) Can be combined with a percentage for a workgroup view (for example, 20 percent of accounts in business unit cracked in less than 10 minutes). Is your password subject to a lunchtime attack? That is, can it be cracked in the 45 minutes you are away from your desk to nosh?
Not good for: User admonishment, judgment. The point of this exercise is not to punish offending users, but to improve your security. Skip the public floggings and just quietly make sure employees stop using their mother's maiden name for access.
Try this: Use password cracking as an awareness-program audit tool. Set up two groups (maybe business units). Give one group password training. The other group is a control; it doesn't get training. After several months and password resets, try to crack the passwords in both groups to see if the training led to better passwords.
One possible visualization: Both YAH and small multiples graphics could work with this metric. (See the graphics for Metric 1 and Metric 2.)
Metric 4: Platform Compliance Scores
Widely available tools, such as the Center for Internet Security (CIS) scoring toolset, can run tests against systems to find out if your hardware meets best-practice standards such as those set by CIS. The software tools take minutes to run, and test such things as whether ports are left unnecessarily open, machines are indiscriminately shared, default permissions are left on, and other basic but often overlooked security lapses. The scoring system is usually simple, and given how quickly the assessments run, CISOs can in short order get a good picture of how "hardened" their hardware is by business unit, by location or by any other variable they please.
Expressed as: Usually a score from 0 to 10, with 10 being the best. Best-in-class, hardened workstations score a 9 or a 10, according to Jaquith. He says this metric is far more rigorous than standard questionnaires that ask if you're using antivirus software or not. "I ran the benchmark against the default build of a machine with Windows XP Service Pack 2, a personal firewall and antivirus protection, and it scored a zero!" Jaquith notes.
Not good for: Auditing, comprehensive risk assessment or penetration testing. While a benchmark like this may be used to support those advanced security functions, it shouldn't replace them. But if you conduct a penetration test after you've benchmarked yourself, chances are the pen test will go more smoothly.