INDEPENDENT WORK SAMPLE / SYNTHETIC DATA
A falling score.
A different mix.
This example shows the style of investigation readout you can expect: a clear observation, transparent calculations and a next step proportional to the evidence. It is an independently constructed demonstration, not a client engagement.
1. The question
Handoff satisfaction falls from 75% to 66% between two weeks. Did the experience worsen, or did the type of cases change?
2. The evidence
| Week | Case type | Satisfied | Responses | CSAT |
|---|---|---|---|---|
| 1 | Routine | 90 | 100 | 90% |
| 1 | Complex | 60 | 100 | 60% |
| 2 | Routine | 36 | 40 | 90% |
| 2 | Complex | 96 | 160 | 60% |
CSAT = satisfied responses ÷ all survey responses. Week 1: 150 ÷ 200 = 75%. Week 2: 132 ÷ 200 = 66%. The complex-case share rises from 50% to 80%, while each subgroup’s satisfaction rate stays unchanged.
The finding: the overall decline is explained by the deliberately constructed change in case mix. This dataset does not show a deterioration within either subgroup.
The boundary: it does not establish why complex cases have lower satisfaction or whether a real handoff loses context.
3. The next step
Before recommending training or a new tool, inspect matched conversations for repeated explanations, lost troubleshooting and unclear ownership. Keep the issue type and complexity comparable. If that evidence supports a structured-summary pilot, agree the outcome and guardrails before running it.
4. What is missing
This example has no transcripts, survey-invitation counts, nonrespondent outcomes or repeated-customer identifiers. It cannot establish causal impact, response bias or an improvement from any intervention.
5. What I contributed
The example connects an operational question with a segmented comparison, a reproducible calculation and a bounded recommendation. The interactive concept lets you vary the mix and see the result. It runs in the browser with fixed subgroup rates; it does not call a live AI model.