The context
An unnamed financial services company faced substantial rework in loan and lease applications. A published iSixSigma case describes a designed experiment spanning its Midwest and Northeast operations. The bank reported that incomplete customer information caused 60 percent of applications to require reprocessing. See the published case.
This is an analysis of that documented case, not a Lean Six Sigma International client project. The source describes the experiment and outcomes, but not a complete DMAIC project record. The measurement, regression, governance and control activities below are explicitly proposed Black Belt extensions, not claims about undocumented work.
The problem
The practical question was not simply whether employees could process applications faster. It was whether the application process could produce usable information without repeated correction.
For a Black Belt charter, the financial hypothesis would connect avoidable clarification work to processing capacity, operating expense and delayed business. Finance would need to establish the value of those effects rather than treating every saved minute as cash savings. Customer completion effort and processing accuracy would remain safeguards against an apparently faster but weaker process.
How the Black Belt would scope it
The proposed boundary would start when a customer receives an application and end when processing accepts a complete submission. Credit decisions, pricing and underwriting policy would remain outside scope.
The sponsor would bring together regional operations, customer-facing staff, form owners, technology, compliance and finance. Each would have a specific decision to make: which information is essential, how customers should supply it, how completeness is judged and how benefits will be verified.
The charter would distinguish three outcomes: completeness at first submission, applications returned for correction and elapsed time to an accepted application. Combining them into one vaguely defined “completion rate” would make the project difficult to evaluate.
The tools applied phase by phase
Define: agree on the decision and boundaries
A proposed SIPOC and cross-functional process map would identify who supplies application information, who checks it and where clarification loops begin. The Black Belt would ask regional staff to walk through actual submissions before suggesting a replacement form.
The Define review would require agreement on the process boundary, financial hypothesis, customer safeguards and accountable process owner. It would also settle whether the intended improvement is a common form across regions or a justified regional variation.
Measure: validate the completeness assessment
The proposed measurement system analysis would test whether reviewers agree on what constitutes a complete application. Reviewers from both regions would independently assess the same anonymized applications, then repeat the assessment in a different order without seeing their earlier decisions.
The study would compare within-reviewer consistency, between-reviewer agreement and agreement with an expert reference. Ambiguous fields would trigger clarification of the scoring rules and another assessment. This follows the approach described in Michael Mueller’s explanation of judgment-based measurement systems.
The collection plan would record application type, region, form version, first-submission status and correction effort. It would preserve separate denominators for fields completed and whole applications accepted.
Analyze: separate process effects from regional differences
For the training extension, the Black Belt would specify a regression model before examining treatment results. Candidate terms would include form features, application type, region and selected interactions. The response definition would determine the model: a binary acceptance outcome should not automatically receive the same analysis as a continuous completeness score.
Model review would consider effect estimates, uncertainty, lack of fit and residual patterns rather than selecting changes from p-values alone. These checks are covered in NIST’s guidance on testing experimental models.
Regional averages would be examined alongside within-region variation. For a continuous-response model, residuals would be checked against predictions, region and run sequence. A high R-squared would not by itself establish an adequate model, as NIST’s model-validation guidance explains.
Improve: test combinations, not isolated preferences
The documented experiment used 16 runs covering five two-level factors: loan versus lease, region, current versus enhanced descriptions, current versus enhanced examples, and the presence of negative examples. Enhanced descriptions and positive examples drove improvement. The source also discusses an interaction between region and negative examples. See the experimental design and findings.
A Black Belt reviewing this design would inspect its alias structure before attributing an apparent effect to one factor. Fractional designs can confound effects, making some explanations inseparable without assumptions or additional runs. NIST explains this limitation.
For a confirmation experiment, the proposed plan would randomize form variants within region and application type, specify replication and sample size, and retain a comparison condition. Region would be a comparison or blocking variable, not something randomly assigned to customers. The design would follow the objective-setting and variable-selection sequence in NIST’s experimental-design guidance.
Implementation would require approved wording, controlled form versions, staff practice and withdrawal of obsolete forms. Regional managers would receive the evidence and an escalation route for genuine local requirements.
Control: assign ownership and a response plan
The proposed control plan would monitor the proportion of applications requiring correction by region and application type. A p-chart would be appropriate only after checking its assumptions, including independence and a meaningful binary classification. Limits would reflect subgroup sample sizes, following NIST’s proportions-chart guidance.
The process owner would investigate signals, audit form versions and periodically repeat reviewer-agreement checks. Finance would separately track released capacity, expense actually avoided and additional business handled.
What the result was
The source reports application completeness rising from 60 percent to more than 95 percent, less reprocessing and capacity for additional revenue without increased headcount. It provides no dollar benefit. Its opening reprocessing rate and closing completeness measure should not be treated as interchangeable. These are the reported outcomes, not independently verified financial results.
What a trainee should take from it
The visible change was better application guidance. The Black Belt work is establishing trustworthy measurement, separating regional effects, testing combinations and coordinating adoption across departments. A strong project record would make every claimed benefit traceable to a defined measure, an appropriate comparison and an accountable owner.
