Loan Cycle-Time Measurement System and Capability Study
A Black Belt is measuring cycle time for small-business loan applications. The primary metric is elapsed hours from application submission to credit decision. Two data sources exist: an automated workflow timestamp and a manual log maintained by credit analysts. A crossed Gage R&R study uses 10 applications, 3 analysts, and 2 trials. The study reports %GRR = 28% of the 72-hour tolerance, ndc = 3, repeatability = 9%, reproducibility = 26%. Review of the raw data shows analysts disagree on when the clock starts for 4 of 10 applications: some use submission, others use 'all documents received.' The team plans a capability study next week and wants to report Cpk against a 48-hour customer promise. Applications fall into two streams: standard (70%) and exception (30%). Exception applications require manual underwriting and have a different workflow. No decision has been made about whether to pool the streams.
1. Interpret the Gage R&R and decide whether the measurement system is acceptable for the capability study.
Read the worked answer
The study says the measurement system is not ready for a capability study. %GRR = 28% of the 72-hour tolerance is in the marginal-to-unacceptable range, and ndc = 3 means the system can distinguish only about three levels of application cycle time, well below the 5 normally needed. Reproducibility (26%) dominates repeatability (9%), which points to disagreement between analysts, not instrument noise. The raw data confirm this: four of ten applications have ambiguous start events. Capability indices are attenuated by measurement variation; a Cpk calculated now would understate true capability and could trigger unnecessary process changes. The right sequence is: (1) define the start event with the customer and process owner, (2) run an attribute agreement analysis on the start decision, (3) retrain and simplify the log, (4) repeat the GRR. Target %GRR <10% and ndc >=5 before reporting Cpk. If the business cannot wait, report the capability result with a measurement-variation caveat and use it only for prioritization, not for a go/no-go decision.
2. Recommend how to define the metric and handle the two application streams.
Read the worked answer
Start by resolving the customer-facing definition. If the promise is 48 hours from submission, the primary metric must start at submission, even if underwriting cannot begin until documents are complete. The 'documents received' event is a separate internal metric that explains delay. Do not pool standard and exception applications: exception applications have a different workflow and 30% of volume, so pooling mixes two distributions and inflates overall variation. Report capability separately by stream, or fit a model with stream as a covariate. For the dashboard, show both the overall customer-facing cycle time and the split by stream, because an overall average can improve while exception applications worsen. Update the data dictionary and train analysts on the start/stop rules. This is a definition and stratification decision, not a statistical one; it determines whether the capability number means anything.
3. What sampling and analysis plan would you set for the baseline capability study?
Read the worked answer
After the MSA is acceptable, define the baseline population and sampling frame: all small-business applications over a stable period, excluding any weeks with known process changes. Because cycle time is likely skewed, use an individuals chart or a weekly Xbar-S chart to check stability, then report median, 90th percentile, and Ppk; do not rely only on Cpk. Sample enough per stream to estimate the 90th percentile with reasonable precision; a practical minimum is 30-50 applications per stream, more if the distribution is heavy-tailed. If sampling from a queue, sample randomly across weeks and include all shifts; avoid consecutive-only samples that capture one operator. Record submission channel, completeness at receipt, and exception reason so the team can later explain variation. Recheck the MSA after any workflow change. The analysis plan should specify non-normal capability methods and a phase split if the process is not stable.