16,000 tickets analyzed across January 1, 2026 to March 31, 2026, on a synthetic dataset of customer support operations.
Chat is the structural problem.
Chat carries 43% of volume but generates 50% of operational cost, 60% of negative CSAT, and 100% of reopens. Fixing chat is the single biggest lever in support operations this quarter.
Four channels compared on CSAT, handle time, reopen rate, and volume share.
| Channel | Tickets | Volume % | CSAT % | AHT (min) | Reopen % | % of cost |
|---|---|---|---|---|---|---|
| Phone | 2,902 | 18.1% | 87.4% | 9.2 | 0.0% | 13.1% |
| 6,118 | 38.2% | 71.3% | 12.0 | 0.0% | 36.0% | |
| Social | 163 | 1.0% | 70.6% | 10.0 | 0.0% | 0.8% |
| Chat | 6,817 | 42.6% | 53.1% | 15.0 | 39.7% | 50.1% |
Phone clears 87% CSAT. Chat clears 53%. Same period, same customers.
The 34-point spread is structural, not seasonal. Chat is the only channel with reopens, the longest handle time, and the most volume, all at once.
Same data sliced by both dimensions. Each cell is CSAT % with ticket count below.
| Channel | Account | Payment | Login | Delivery | Refund | Other |
|---|---|---|---|---|---|---|
| Phone | 85% 468 | 87% 466 | 88% 471 | 88% 502 | 87% 493 | 88% 502 |
70% 1,029 | 72% 1,034 | 73% 985 | 71% 1,048 | 71% 1,010 | 72% 1,012 | |
| Social | 75% 24 | 74% 27 | 72% 29 | 68% 22 | 77% 35 | 54% 26 |
| Chat | 50% 1,143 | 54% 1,138 | 52% 1,107 | 57% 1,164 | 51% 1,109 | 53% 1,156 |
Phone is green on every issue. Chat is red on every issue.
Channel choice dominates issue type. There is no problem type chat handles well, and no problem type phone handles poorly. The fix lives in the channel, not the queue.
Volume, CSAT, and handle time traced by month.
Y-axis ranges are tight on purpose. Auto-scaling to zero would hide the actual movement (or absence of it).
CSAT moved 1.4 points across the entire quarter. There is no Q1 decline to explain.
Any aggregate 'things are getting worse' narrative is a channel-mix artifact. Heavier-chat months look worse without anything actually changing in operations.
Ticket count and CSAT plotted against handle-time bin.
CSAT collapses 22 points as handle time grows from 5 to 25 minutes.
Resolution quality drives satisfaction, not greeting speed. Chat sits at 15 minutes (the collapse zone). Phone sits at 9 (the high-CSAT zone). First-response time, by contrast, shows no CSAT relationship at all.
X axis: escalation rate. Y axis: reopen rate. Bubble size: ticket volume.
Account and payment escalate at 2x the baseline rate.
These are complexity issues. Customers need tier-2 expertise, but they're filing through first-touch chat first. Account and payment are 33% of volume but generate 48% of all escalations.
Delivery and refund reopen at ~20%, well above baseline.
These are wait-state issues. Resolution depends on external events the agent cannot trigger in-session (a refund clearing, a package arriving). Synchronous chat is the wrong channel for them.
Things this dataset cannot answer, and what to do about it.
Some of what looks like channel variance may be operator variance. We cannot separate them without an agent identifier on each ticket.
Phone, email, and social all show 0% reopen, which is almost certainly an instrumentation gap rather than perfect resolution.
Reliable channel comparisons require agent-level instrumentation.
Without agent_id, team, and tenure, we are at the ceiling of what this dataset can prove. The next layer of analysis (individual variance, training cohorts, tenure curves) is gated on capturing those fields.