How to tell when a useful metric is becoming the target

A useful metric becomes dangerous when people can improve the indicator while weakening the outcome it is meant to represent. Stress-test the measurement and the consequences attached to it. Examine actual work before blaming individuals for responding to contradictory incentives.

Reinforcement is a consequence-dependent increase in the likelihood of behavior, so something counts as reinforcement by its effect rather than by a manager’s positive intention. This definition follows APA Dictionary of Psychology: reinforcement (American Psychological Association (2018), defining publisher entry; no single inventor claimed). A reward is not reinforcement unless behavior becomes more likely. For this metric review, ask which behavior becomes more frequent under the actual consequence, not which behavior the manager hoped to encourage.

A private moment that makes the issue visible

Composite scenario, not a client case: You share a strong dashboard result at home, then receive a customer complaint about the same process. The team has closed more cases by moving difficult work into another queue. The visible indicator improved while the underlying obligation remained.

State the decision before discussing personalities

The review concerns incentive misalignment. Kerr’s 1975 account of rewarding one behavior while hoping for another provides the research frame; reinforcement is the established learning construct used below.

Proposed method: metric outcome-stress test

State the intended outcome. Explain the result the indicator should represent. Identify who benefits and which obligation must remain protected. Ask the receiving person what must be true before they regard the work as complete. Record a condition that the current measure does not capture. Use that condition to choose cases for examination.
Map the measurement rule. Check the numerator, denominator, timing, and exclusions with responsible owners. Record changes that affect comparison. Inspect a period boundary and an exception category. Ask whether delayed, transferred, or reopened work remains visible. Preserve the calculation used for each sampled record so a second reviewer can reproduce it.
Find a plausible gaming route. Ask how someone could improve the number without improving the result. Use the question defensively to detect and remove a weakness. Describe the weakness without teaching staff to conceal it. Keep exploit details in the authorized repair discussion. Ask whether the shortcut is possible under the current rule before deciding whether anyone used it.
Inspect real work cases. Trace favorable and unfavorable results through the process. Distinguish measurement failure, conflicting incentives, and substantiated misconduct. Compare the system record with the actual recipient experience. Let the employee explain a disputed case with its context. A favorable count and an unresolved obligation can coexist without proving deliberate misrepresentation.
Revise and review consequences. Adjust the measure or reward through the appropriate process. Keep a route for exceptions and check for new unintended effects. State when the revised rule begins and how older figures will be labeled. Review any pay or employment implications with the appropriate specialists. Require a fresh case check after the next reporting cycle.

Trace the indicator back to the customer obligation

Before changing the target, write down the result a customer or downstream team should receive. For a support process, that might be a resolved problem with an accurate record. For a sales process, it might be a commitment delivery can fulfill. These are proposed descriptions to refine with the responsible owner. Keep the outcome separate from the event your system can easily count. A closed ticket, signed order, or completed call can be useful evidence without being the whole obligation.

Build a small case comparison from records you are authorized to inspect. Include a case counted as successful that still required substantial follow-up. Include a case recorded as late where the employee protected an important requirement. Ask the process owner to explain both. Do not select only the cases that support abolishing a target you already dislike. The comparison should expose what the measure misses and whether that omission matters for its intended use.

Kerr’s discussion of contradictory reward systems offers a reason to inspect this mismatch, rather than assume that repeating a quality slogan will resolve it. Its examples do not establish the cause of a particular reporting anomaly. Use the paper as a prompt to compare what the organization says it values with what its formal decisions reward. The source supports that distinction; the case audit remains your proposed application.

Examine what happens at the reporting boundary

Draw the path from initial request to final outcome. Mark every place where a record changes status, moves between teams, or disappears from the report. Ask who can make each change and whether the receiving team accepts the work. If a request leaves the queue before anyone has resolved it, inspect the receiving obligation. A transfer should not become successful completion solely because the first team’s dashboard no longer contains the item.

Check how reopened work is represented. Preserve its link to the original case while allowing the process owner to distinguish a new problem from an unfinished one. Avoid inventing a universal reopening window. Choose a review period that fits the actual service and document why it was selected. Ask what happens to cases spanning reporting periods. Otherwise, an apparent improvement may reflect work moving into a different period rather than a better result for the recipient.

Keep measurement defects separate from allegations about people. A confusing status definition, missing integration, and deliberate falsification require different responses. If there is evidence of misconduct, retain the records through the authorized employment and legal process. Do not turn an exploratory workshop into an accusation meeting. Ask staff to describe awkward cases without requiring them to confess a motive you have not established.

Test the proposed replacement before attaching consequences

Write a candidate definition and calculate it on the same historical cases. Show which cases change category and which remain unchanged. Ask the people doing the work whether the revised measure creates a different shortcut. A more complicated formula can still omit the obligation you care about. Require the owner to explain the result in ordinary language before using it to determine rewards, staffing, or judgments about individual performance.

Consider a paired review: the indicator plus a small examination of completed work. Specify how cases are chosen and who checks them. Treat this as a proposed safeguard, not a validated scoring system. Avoid adding an expanding list of measures every time an exception appears. If each measure requires another to explain it, reconsider whether a single score is suitable for the decision at all. Some consequential decisions need a short case account alongside a number.

Fischhoff’s hindsight research is relevant to the review record. Preserve the rationale for the original measure before the latest complaint makes its weaknesses seem obvious. Identify which problems were known and which emerged only after implementation. The question is what to repair now and what design practice to improve next time. An honest account can recognize a reasonable initial choice while requiring a correction once new evidence is available.

Make the revised target understandable to the people using it

Explain the intended outcome, the changed counting rule, and the date the new comparison begins. Show a worked case with identifying details removed. Ask a team member to calculate the result and explain the outstanding obligation. Resolve differences before the reporting cycle closes. Keep older results labeled under their original definition; do not splice incompatible figures into a trend that appears continuous.

For a hypothetical handoff review, suppose an employee transfers a complex request to a specialist. The first team records the transfer, the specialist accepts ownership, and the customer still awaits an answer. Record those facts separately. You can recognize a good handoff without declaring the whole request resolved. If the specialist rejects the handoff, name who must restore ownership. This is a process-design example, not a claim about how every support system works.

At the next review, compare the indicator with the underlying cases again. Ask whether the correction protected the customer obligation or shifted unfinished work somewhere less visible. Include an example where the revised rule was inconvenient but useful. Decide whether to keep the rule, narrow its use, or stop using it for individual evaluation. The deliverable is a defensible measurement decision with a known limitation, rather than a dashboard that looks uniformly favorable.

Put the boundary into words

This number is intended to represent this outcome. Show a case where improving the number could weaken that result. We will inspect the work and the incentives before using the indicator to judge performance.

Work through a bounded example

Hypothetical application: A completion target excludes reopened requests. The stress test reveals that early closure can improve the measure. The company reviews the definition and the handoff with the authorized owners rather than asking employees to protect quality while rewarding the opposite.

Check the failure mode before scaling

Do not provide a playbook for hiding poor results. Avoid assuming every metric difference proves manipulation. Use appropriate employment, financial, and legal review for actual reward changes or allegations.

Review the work, then revise the arrangement

Review the business outcome alongside the indicator and inspect exceptions. Preserve changes to the measurement rule so improved reporting is not confused with improved performance.

A related research boundary appears in Kerr (1975): On the Folly of Rewarding A, While Hoping for B. Use it to question the design rather than attach a scientific label to an individual. The sources support the distinctions discussed here. They do not establish that the proposed worksheet or script has been experimentally tested as a complete intervention.