The 2.7 Percent: A Breach Investigator on the Number Nobody Audits
Archi4 min read
Contents
The number that surprised him#
Marcus Feld is a breach investigator who has worked on incident response for about a dozen years, mostly for mid-sized companies. I asked him for the single figure that changed how he runs an investigation. He gave me 2.7 percent.
That is the share of stolen records, in his own case files, that his team could actually trace to a destination. The rest vanished into caches, resale bundles and re-uploaded copies. The number is his own tally, not an industry study, and he is careful to say so. It still shapes every report he writes.
Q: Why 2.7 percent?
A: We pull server logs, egress records, paste site hits. For every hundred records we confirm left the building, we can say where about two or three went. The other 97 are out there. We just cannot follow them.
Q: What does that look like in practice?
A: A customer table gets copied at 2 a.m. By the time we are called, the file has been split, compressed and posted to two or three forums. Each copy has a different hash. We trace one. The rest are a guess.
Why the untraceable majority matters#
Most breach coverage focuses on the entry point: the stolen credential, the unpatched server, the vendor with weak access controls. Feld says that is only half the job. The exit is where the damage compounds, and it is where records go quiet.
The resale bundle#
Stolen data rarely sells once. A bundle of email addresses and hashed passwords can be resold for months. Buyers test the hashes against other services. Feld has seen one credential dump reappear in three separate incidents, each with a new seller and a new price.
Q: Does that change how you advise clients?
A: It changes the timeline. If you assume the data is gone and will be reused, you force password resets on every service, not just yours. Most companies do not want to hear that because it costs support hours. But it is the only advice that matches the 2.7 percent.
The audit gap#
Feld says few companies measure traceability at all. They count records exposed and move on. The share they can follow is rarely in the incident report, partly because it is embarrassing.
Q: Is that a failure of tools?
A: Partly. Egress logs are often kept for 30 days or less. By the time an investigation starts, the evidence window has closed. The other part is organizational. Nobody owns the question of where the data went after it left.
A small test any team can run#
Feld suggests a simple exercise. Take the last ten incidents your team handled, even small ones. For each, write down how many records you could trace to a specific destination. Divide by the total. He says most teams land under 5 percent.
Q: And then what?
A: You stop treating the breach as a single event. You treat every record as a liability with an unknown address. That shifts spending toward detection on the way out, not just prevention on the way in.
Q: What does detection on the way out look like?
A: Alerts on large outbound transfers, yes, but also on unusual read patterns. A service account reading the entire customer table at 3 a.m. is a signal. Most teams log it. Few alert on it.
The cost of not knowing#
Feld keeps a rough ledger. For a mid-sized client, the response to a 40,000-record exposure ran about 900 staff hours across legal, support and engineering. The traceable portion was a few hundred records. The rest of the budget went to notifications and credit monitoring, which he calls a blunt instrument.
Q: Would better traceability lower that cost?
A: It would lower the guessing. If you know which 3,000 customers are at real risk, you can notify them and skip the bulk mailing. That saves money. It also saves trust, because a vague notice to 40,000 people reads as noise.
Q: What is the one change you would make first?
A: Keep egress logs for a year. That single decision would have moved the number in most of my cases from 2.7 to somewhere in double digits. It is boring. It is cheap compared to a notification campaign.
What the number does not say#
The 2.7 percent is not a measure of how much data leaks. It is a measure of how much an investigator can follow. Feld is clear that the true figure could be higher or lower in other shops, and that his cases skew toward companies with limited logging budgets.
The lesson he draws is narrower. A breach report that says how many records were exposed, and nothing about where they went, is an incomplete document. The gap between the two numbers is where the next incident is already being prepared.
More articles