fpf.tools
Jump to text
On this page
Copy and source
Original source · lines 3667–3691
Published source confirmed at last check

Source changed 2026-10-14 19:58:00 UTC · snapshot created 2026-10-14 19:58:38 UTC · last check 2026-10-15 00:05:10 UTC

RMP.Application:5.4 - Ask for a comparison the organization can obtain and run

The manager gives a qualified analyst a concrete request:

We need to decide whether to fund one additional hour of approval capacity per working day in each participating branch. The new arrangement adds a trained approver during the stated hour, using the same approval rules; the current arrangement has no additional approver. Our outcome is working time from receipt of an eligible customer request to actual sending, including waits for missing input. We also need the change in the proportion of quotations requiring a customer-facing correction. Please specify which cases and branches the answer concerns, how to assign the arrangements, how to handle unfinished or missing cases, and which comparison and uncertainty calculation can support the decision.

The request supplies current workload and complexity records, staffing arrangements, queues, expected arrival volumes, the reporting deadline and the people who can operate the record. The manager also provides a prospective criterion: the lower limit of the specified two-sided 95% interval for the mean reduction must exceed one working day, and the upper limit of the corresponding interval for the increase in correction rate must be below 0.5 percentage points. These are this case’s operating criteria, chosen with the responsible decision owner before examining intervention results. They are not universal scientific thresholds.

The analyst must explain how the assignment, outcome, dependence, expected precision, observation period, analysis and stopping rule answer this question. “Run an A/B test” leaves those operations unspecified. RMP.2 uses the supplied professional construction to choose an obtainable design or a useful stop; RMP.3 makes its consequential operations executable. Ask the analyst to work through one branch and one request so that the responsible staff can check what they would actually do. A citation or an assisting agent’s fluent explanation does not supply an unavailable design qualification.

For the positive branch, suppose the analyst supplies and justifies the following arrangement for this workplace:

  • Sixteen branches have separate customer populations, queues and approvers. They do not share reserve staff or transfer requests. Eight pairs are formed using pre-intervention workload, complexity and waiting records. Within each pair, random assignment gives one branch the new arrangement and one the current arrangement. Both run over the same calendar period. Requests enter during four weeks; assigned staffing continues for six weeks to cover the planned follow-up. The protocol specifies the limitation to return if sending and correction follow-up remain incomplete, rather than silently dropping those cases.
  • The target is the average effect across these branches at the specified workload, giving each branch equal weight. The analyst first calculates each branch’s mean for all eligible requests arriving during four fixed weeks, then the eight paired branch contrasts. The number of requests improves information about a branch; it does not turn each request into a separately randomized branch.
  • Eligibility is settled at arrival. The total waiting clock includes time obtaining missing input; the analysis does not adjust away approval time, an intended mechanism of the intervention. Follow-up covers sending and the agreed five-working-day correction window. Unsent, withdrawn or untraceable requests have a specified handling rule and remain visible; dropping them because they are inconvenient is not that rule.
  • Both arrangements use the same recording and staff-notification practice. Time comes from the relevant events. An assessor who can examine the quotation without its assignment label applies the specified customer-correction rule. The analysis retains the actually used staffing, other changes and departures from the intended arrangement.
  • Historical variation and the expected volume support the proposed precision under the analyst’s stated assumptions. For this constructed case, the analyst qualifies a paired-contrast model and its interval calculation. Four weeks and eight pairs come from that supplied justification; neither number can be copied into a different workplace without it.
  • The plan fixes the primary outcome, weighting, interval calculation and analysis before results are examined. A serious customer harm stops the operational trial and returns the affected decision; missing records or staffing departures are reported for interpretation. An attractive interim result does not authorize early success stopping or a newly selected primary outcome.

Matching branches reduces known background differences; random assignment supplies the allocation mechanism. Neither guarantees that a realized small comparison is perfectly balanced or protects against a shared queue. The NIST account of randomized blocks explains the distinction between the factor being varied and background variation controlled by blocking. Its design principle is not a qualification of this workplace protocol.

Obtain the complete protocol and its grounds from the analyst before relying on it. The summary above identifies what the manager inspects and operates. It is not a replacement for the setting-specific treatment of unfinished cases, precision and analysis.