Test competing explanations
5 min read
Lesson objectives
- Build rival explanations for backlog and identify evidence that could strengthen or refute each.
- Recompute output per productive hour and propose a temporary decision with limits.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
When absence and backlog rise together, a team may conclude that absence caused the backlog. Coincidence is a reason to investigate, not proof of cause. Case mix, a tool outage, demand or the definition of a closed case may also have changed. Write two or three hypotheses before requesting more data and state what each predicts. A loss of available staff hours predicts fewer total productive hours with roughly stable output per hour. More complex work predicts fewer completed cases per hour or longer handling time. Faulty routing predicts a queue in one channel and idle capacity in another. If the proposed indicators cannot discriminate among hypotheses, a longer version of the same report will not help. Request a different measurement. The adviser's role is to identify information that could change the decision, rather than collect only evidence that supports the sponsor's first account.
Use an evidence matrix with claim, support and alternative explanation. For quantitative data ask about period, denominator and limitations. In manager interviews ask who was selected, who was not asked, when the observation was made, and what behaviour lies behind a judgment. One interview does not represent every worker, yet it should not be dismissed because it is not a number; it may reveal a process step the system fails to record. Direct observation can expose waiting between steps, but one unusual day may mislead. Compare independent sources when possible: absence and shift logs, case and outage logs, team accounts and sampled files. Agreement among different sources increases confidence but does not turn correlation into a controlled experiment. When they conflict, record the conflict and its possible causes, then design a small test such as a random sample of cases before and after a procedure changed.
Output per hour is not a measure of a person's worth and cannot fairly assess an individual without context. Here it is a process diagnostic: team cases closed divided by the same period's productive team hours. Extra shifts may replace absent hours, so total hours can remain constant while the reported absence rate rises. Inspect also whether closure quality deteriorated or reopens increased; apparent output can hide rework. When comparing weeks, hold the case definition constant and show arrivals, closures and opening backlog. The accounting identity is closing backlog = opening backlog + arrivals − closures ± adjustments. If it does not balance, investigate channel transfers, cancellations and reopened cases before judging performance. Attach source and extraction date to each measure and separate the observed pattern from its possible cause in the advisory memo. A clean calculation based on inconsistent definitions does not become valid merely because it is precise.
A small discriminating test can reorder the options. Select, for example, 40 cases from each period with comparable channels and types. Compare handling time, waiting and reopens, keeping worker identities out of the sponsor report. If time rose for comparable cases after the new verification step, the process account grows stronger and a controlled adjustment is plausible. If case mix changed while time within each type stayed stable, revisit skill capacity and demand instead of blaming that step. In either case acknowledge unobserved factors and avoid a categorical causal claim. A one week trial for a small team could declare success in advance as reduced new backlog without higher reopens or errors, and compare a similar week. Set a stop rule if quality worsens or the test exposes worker privacy risk. Do not propose intrusive individual monitoring when aggregate process records answer the question; more intrusive processing needs specialist review under applicable context.
Lesson summary
Build rival explanations for backlog and identify evidence that could strengthen or refute each. Recompute output per productive hour and propose a temporary decision with limits.