Skip to content
Throughscan

Reads

Real reads on real support data.

Examples of what the assessment surfaces, from real support data. Every number is computed in code and labeled by where it came from. The first one runs on a public dataset, so the method is fully reproducible; the rest come from real audits, anonymized with the team's sign-off.

Read 01 · public support channel

The median first answer was 2 hours. The p90 was nearly two days.

Source and method. The public Apache Tomcat user-support mailing list, pulled as a native mailbox export, January through June 2024: 205 question threads, 926 messages, answered by maintainers and the community. Throughscan read the mailbox directly. Every number below is computed in code over the whole set, not estimated, and the data is public, so it is fully reproducible. One honest caveat kept up front: this is a community help list, so "first answer" means the first human reply in the thread, not an agent's SLA clock. The pattern it shows, a healthy-looking median hiding a brutal tail, is the same one a support team sees in its own numbers.

measured = computed in code generated = a draft to confirm strategy = judgment

The number an average hides measured

Median time to a first answer was 2 hours 10 minutes, which on its own says a responsive channel. But the median hid the tail. Look at the same data three ways:

Bars are to scale. The average people report is 611% above the median, and one in ten waited nearly two days.

611%
the average runs above the median, pulled up by the slow tail
23%
of questions got no reply at all (47 of 205)
20.5%
of askers came back within a week, the hidden-rework signal

What customers actually came for measured

The top five contact reasons were 56% of all threads. That concentration is the opportunity: a few well-aimed guides cover most of the volume.

Shares of all 205 threads. A separate 15.1% were release and security announcements (broadcasts, not questions), which the engine set aside rather than counting as support.

The one move generated + strategy

Set the response target at the p90, not the average, and staff or document toward the tail, because the tail is where customers actually churn, and an SLA pinned to a flattering average institutionalizes ignoring them. Then attack the two leaks the data names: the ~23% of questions that get no answer, a triage and ownership gap, and SSL/TLS configuration, the biggest single driver, with one authoritative known-good config guide.

The honest part measured

10.2% of questions did not sort cleanly into a reason and were left uncategorized rather than force-fit. That uncategorized share is the honesty check: a read that forces every ticket into a category is hiding its misses. The cleaner and more ticket-shaped the export you upload, the tighter that number gets, and the numbers above become yours instead of a public list's.

Want this on your own data, where the categories are tighter and the response-time numbers are yours? Upload one export and see where you stand.

Start the free assessment