← All reports

🎧 Customer Support report

Mon 8/10/26 to Sun 8/16/26

Report generated: Mon 9/21/26 @ 10:02pm Central

Last sync: Mon 9/21/26 @ 6:04am Central, 8,719 tickets.

Replayed: tickets and messages after that date are excluded, but each ticket's status is its status today, not its status then.

❤️ HEALTH🔍 DIAGNOSIS🌟 PRAISE

❤️ HEALTH

Is the desk ok right now?

📊 Activity

DayMsgs in<1h1-2h2-4h4h+Waiting
Mon 8/102578210
Tue 8/1125155200
Wed 8/1231173111
Thu 8/1333136610
Fri 8/1428112600
Sat 8/15000000
Sun 8/16000000

Business hours only. Rows do not sum to messages in: 34 more were closed without a reply, nearly all of them a client saying thank you. The ones that were not are under Bad closes.

🚨 Needs attention now

4 messages are waiting on a reply. Longest first.

WaitingCompanySubject

🔇 Bad closes

⚠️ 79 of 79 closed threads in this window are unjudged. An unjudged thread cannot appear as a bad close, so what follows is a floor, not the number.

To finish the window: python3 freshdesk/thread_evaluation.py --as-of 2026-08-17 --threads threads.md, read and judge them, then --verdicts verdicts.jsonl, then re-run this report.

📈 Against your own baseline

This week4w baseline
Tickets in82 ↑66+24%
Messages in142 ↑115+23%
Closed79 ↑64+22%
Still open3
Median response38m ↓45m🟢 -14% better
p90 response2.9h ↓4.3h🟢 -33% better
Still waiting1

Response times are per inbound message, not per ticket, and include messages still waiting at their elapsed-so-far duration. Dropping unanswered ones would mean ignoring a customer improved the percentile.
p90 is the number to watch: a fast median with a slow p90 is a desk that looks healthy and has a tail of people who were left.
Rows with for a baseline are cohort-aged: this week's tickets are days old and earlier weeks' have had a month to close, so comparing them would read as worse every week no matter what.

⏱️ Response distribution

Wait
p5043mhalf were faster
p751.8h
p903.8h10% waited longer than this
p9515.8h
p9916.3d1% waited longer
max29.3d
AnsweredMessagesShare
< 1 business hour27260%
< 4 hours41290%
< 1 business day42894%
< 3 business days43595%
> 7 business days163.5%

456 inbound messages over the trailing 4 weeks. A single week is too few messages for a p99 to mean anything.

⚡ What changed

🔍 DIAGNOSIS

Why, and who it is coming from.

🏢 Who

Ranked by change against each company's own 4-week baseline, so a client going from one ticket to six outranks a steady heavy user.

CompanyWeekBaselineChange
firstchoicebackground.com3114.8+110%
kennia@19
samanthad@5
paige@4
essentialscreens.com51.0+400%
clientcare@2
dhill@2
etaylor@1
deltastar.com20.0new
dswhr@1
lchand@1
fairscreen.com20.2+700%
shannon@1
support@1
argusverify.com20.8+167%
eric@1
larissa@1
viewhouse.com10.0new
hr@1
sierrausd.org10.0new
hwade@1
lotuscmc.com10.0new
megan@1

🏷️ Tags applied

Computed from the store, not inferred.

💬 Themes

Inferred: read from the ticket bodies, not computed.

SLA expiry is not firing, so clients chase us to close their own orders

Seven tickets from one client asking us to close orders that had already run past their SLA. The pattern is always the same: attempts stop, the order sits open, and the client notices before we do.

Work we completed is not visible on the client's side

Five tickets where the verification was done and the client could not see it, so they re-opened the question. Two of them are the same order reaching two different people at the client.

Dates we report are wrong often enough to cause reinvestigations

Four tickets in one week, and a wrong date is the one error that comes back as a dispute rather than a question.

Parchment runs without the client's approval, then stalls with no fallback

Four tickets. One client was billed for third-party reports they had not approved, twice, and nobody has a plan for a Parchment order that never returns.

An invalid contact stops an order instead of escalating it

Five tickets where the only contact was bad and nothing routed the order to research. The expected behaviour is a flag for manual contact research.

The inbound call path drops work it has already done

Three tickets where a verifier did speak to us and the result did not survive.

A phishing email impersonated a client

Two tickets on the same attempt: a forged Adobe Sign request for a "vendor contract" sent in a client's name. Robin asked whether it was genuine instead of opening it, and the client confirmed it was not theirs.

🌟 PRAISE

Client praise, in their words.

No noteworthy praise this week.

Written by a Claude Code routine. Not for distribution outside Argus.