The tail got slower
36 h → 52 h95th-percentile resolution time rose by 16 hours while the median fell from 10 to 7 hours.
Examine elapsed resolution time →Explore / Measurement
Judge AI-enabled transformation by the value it creates, the burdens it moves and the control it preserves.
A decision before a dashboard
An invoice-exception pilot is ready for review. The sponsor wants to extend it to higher-value invoices. What would you authorize?
Read the whole operating story
A useful scorecard connects the outcome, its cost, the people carrying the work and the boundaries that must hold.
The evidence behind the headline
Keep the trial bounded. Repair the exception and escalation paths, then reevaluate before expansion.
The tail got slower
36 h → 52 h95th-percentile resolution time rose by 16 hours while the median fell from 10 to 7 hours.
Examine elapsed resolution time →Quality deteriorated
97% → 92%First-time-right completion fell by 5 percentage points after 30 days of follow-up.
Examine first-time-right completion →The capacity story changed
400 h − 310 h = 90 hAdded review, correction and support absorbed most of the estimated release. Only 45 hours were put to agreed use.
Examine net capacity released →Required intervention was missed
14 of 100Audited trigger-positive cases missed their required escalation. Read this as 14% of the audit, with its sampling limits.
Examine missed required escalations →Keep scope bounded while a named owner repairs and retests exceptions and escalation. Confirm acceptable quality, workload and control evidence before widening authority. Severe harm that cannot be contained would justify stopping.
All figures are invented for decision practice. Two 1,000-case invoice-exception cohorts are assumed matched for teaching, with 30-day outcome follow-up. Initial no-touch completion is 760/1,000; it excludes later review and correction and does not meet the full bounded straight-through success definition. Of 90 net hours released, 45 were used: 50%. Released capacity is not cash saving. The escalation audit includes 100 trigger-positive cases; no population estimate or statistical significance is claimed.
Build a scorecard
Start with a suggested set, then adjust it to the process, risk and people involved. Each measure needs an owner and a response.
value
What evidence shows that the benefit is real, arrives on time and survives full operating cost?
m01
Net realized cash benefit
m02
Cash benefit realization
m03
Cost per correctly completed case
m04
Net capacity released
m05
Released capacity put to use
m07
First-time-right completion
m28
Sustained benefit retention
service
What evidence shows that the whole service is getting faster, more dependable and easier to use?
m03
Cost per correctly completed case
m06
Elapsed resolution time
m07
First-time-right completion
m08
Aged work backlog
m09
Repeat contact for the same issue
m22
Human oversight effort
m25
Time to restore the service
scale
What evidence would justify giving this operation more scope or authority?
m03
Cost per correctly completed case
m07
First-time-right completion
m11
Harmful decision rate
m12
Authority boundary breaches
m13
Missed required escalations
m15
Complete decision trace coverage
m22
Human oversight effort
knowledge
What evidence shows that decisions can rely on current, authorized and usable knowledge?
m10
Decisions reversed for original error
m16
Critical knowledge health
m17
Evidence-grounded output pass rate
m18
Knowledge-related case failures
m19
Knowledge update lead time
m23
Accepted operating ownership
m27
Time from signal to validated improvement
ownership
What evidence shows that the business can carry the changed operation after the project leaves?
m07
First-time-right completion
m20
Effective workflow adoption
m21
Demonstrated role proficiency
m22
Human oversight effort
m23
Accepted operating ownership
m24
Overdue critical dependencies
m25
Time to restore the service
sustain
What evidence shows that the improvement endures and the operation learns from experience?
m01
Net realized cash benefit
m07
First-time-right completion
m18
Knowledge-related case failures
m25
Time to restore the service
m26
Recurring incident rate
m27
Time from signal to validated improvement
m28
Sustained benefit retention
Selected scorecard
Your selection stays in this browser tab. No account or data submission is needed.
Agree these before the next review
Comparable work and period
Target, tolerance and trigger
Authority to respond
Scale, repair, restrict or stop
Explore the measures
Open a measure to see exactly what it means, what evidence it needs and the decision it should inform. Read it alongside its companions.
28 carefully defined measures across six areas.
1
Value & economics
Is the initiative delivering an economic benefit that can be evidenced?
Verified incremental cash benefits minus incremental transition and operating costs, over the same period. Report period and cumulative currency values.
Finance-approved benefit ledger, invoices and payroll/spend records; explicit counterfactual. Exclude monetized spare capacity unless cash expenditure actually changes.
Benefit owner + Finance / monthly
Operate; improve
Separate cash savings, avoided future cost and revenue contribution. Reconcile overlap across initiatives; revise the case when full costs erase the benefit.
m03
m04
m07
Build the operating case around evidence, cost and ownership.
2
Value & economics
Are cash benefits arriving at the pace assumed in the approved case?
Verified gross cash benefits realized to date / approved gross cash benefits due to date x 100. Use the same benefit categories and price basis.
Versioned benefit profile and Finance ledger. Show the currency gap and plan revision history. A zero planned benefit gives N/A.
Benefit owner + Finance / monthly
Operate; improve
Investigate delivery, adoption and assumption gaps. Above 100% may reflect a weak forecast; this ratio excludes cost, so review net benefit alongside it.
m01
m20
m28
Return to the value hypothesis when the expected benefit does not appear.
3
Value & economics
Does the operation produce an acceptable outcome at a sustainable unit cost?
Total in-scope operating cost / cases meeting the acceptance criteria after the agreed follow-up window. State currency per case.
Include labor, AI/platform use, oversight, support and remediation across successful and failed work. Report implementation spend separately; use a matched case cohort.
Process owner + Finance / monthly
Operate; improve
Segment by complexity and channel. Lower cost from excluding hard cases is a coverage change; make it visible before deciding to scale.
m07
m11
m22
Test the economics of the complete operating service.
4
Value & economics
Has labor capacity actually been freed across the whole process?
Hours needed at the baseline labor rate for current volume and case mix minus actual total process labor hours. Report signed hours.
Validated time study/work logs plus review, exception, support and correction work, including off-system work. Compare equivalent output and quality.
Operations owner / monthly
Pilot; operate
A negative result means more labor was needed. Capacity becomes economic value only through a specific use or spend change; document where it goes.
m05
m07
m22
See which handoffs or hidden work absorb the apparent saving.
5
Value & economics
Has released time been used for an agreed business purpose?
Evidenced released hours actually used for agreed additional work / positive net hours released in the same period x 100.
Capacity plan, work allocation and delivery evidence. Track reduction in overtime or staffing expenditure separately. Zero or negative net release gives N/A.
Business owner / monthly
Operate; improve
Name the receiving work and verify its output. Allocation alone is insufficient. Reconcile a result above 100% before reporting; do not count it again as cash.
m01
m04
m07
Assign responsibility for turning released capacity into useful work.
6
Work & customer outcomes
Are customers getting a timely outcome, including difficult cases?
Median and 95th percentile elapsed hours from valid receipt to accepted resolution. Include waiting, handoffs and reopening time.
Workflow event log; a fixed receipt cohort and stated follow-up cutoff. Report still-open cases and their age. Separate business time from calendar time.
Process owner / weekly
Pilot; operate; improve
A faster median with a slower tail can hide a growing exception problem. Examine case mix and the aged queue before expanding volume.
m07
m08
m09
7
Work & customer outcomes
Is the initial completion dependable enough to avoid later correction?
Completed cases requiring no correction or reopening during the follow-up window / completed cases with a fully observed follow-up window x 100.
Quality checks and reopen/remediation logs. Define acceptance and the follow-up window before the test; disclose cases whose outcomes remain unknown.
Quality + process owner / weekly
Pilot; operate
Investigate failure causes and delayed errors. Higher throughput is useful only alongside acceptable quality; do not improve the rate by suppressing reopenings.
m03
m06
m11
Make acceptance and quality checks part of everyday operation.
8
Work & customer outcomes
Is unfinished work accumulating beyond the service promise?
Open cases older than their agreed case-class threshold / all open cases at the same snapshot x 100. Also show the count and oldest age.
Queue snapshot, original receipt timestamps and case-class service thresholds. Include suspended cases; disclose exclusions and pause rules.
Service owner / daily or weekly
Operate; improve
Inspect the oldest and highest-impact work. A shrinking percentage can reflect an influx of new cases; use the absolute count when assigning capacity.
m06
m07
m22
Find the ownership or handoff that prevents closure.
9
Work & customer outcomes
Has the customer received a sufficiently clear and complete outcome?
Resolved cases followed by an unplanned same-issue customer contact within the defined window / resolved cases with complete observation x 100.
Linked case and contact records across channels. Set the window by service; distinguish planned follow-up and new issues. Report unmatched contacts.
Service experience owner / monthly
Pilot; operate
Review repeat-contact reasons with customers. Low contact can also reflect difficult access to support; compare complaints and sampled feedback.
m06
m07
m11
Connect internal workflow performance with the customer outcome.
10
Work & customer outcomes
Are decisions being corrected because the original judgment was wrong?
Reviewed decisions overturned for an error under the facts and policy available at the time / reviewed decisions with completed adjudication x 100.
Quality reviews, appeals and adjudication records. Use a representative or stratified sample; report sample size and route-specific results.
Decision owner + Quality / monthly
Pilot; operate
Separate original error from later changes in facts or policy. A low appeal rate alone cannot establish accuracy; improve the decision rule or review path.
m11
m13
m17
Clarify what good judgment requires in this decision.
11
Decision quality & control
What material adverse outcomes are decisions causing?
Reviewed decisions with a confirmed material adverse outcome attributable to the decision / decisions reviewed x 1,000.
Independent quality review, complaints and incident investigation. Define harm and severity by context; disclose sample method and delayed/unknown outcomes.
Business risk owner / weekly + immediate severe-event review
Pilot; operate
Show severity and affected groups separately where appropriate. Rare severe events require individual review; a low average cannot establish an acceptable boundary.
m10
m12
m13
Set the consequences that require intervention or reduced authority.
12
Decision quality & control
Is the system acting beyond its explicitly delegated authority?
Executed actions that breached an applicable approval, access, amount or scope boundary / observed executed actions x 1,000.
Action logs reconciled with the versioned authority policy. Report monitoring coverage, severity and each material event. Separate blocked attempts.
Control owner / continuous alerting + weekly review
Pilot; operate
Contain the affected action and inspect the control failure. Zero observed breaches is inconclusive when logging or boundary tests are incomplete.
m13
m15
m25
Turn decision limits into controls that can be observed and tested.
13
Decision quality & control
Are cases reaching a person when intervention is required?
Audited cases that met a mandatory escalation trigger but were not escalated within its required time / audited cases that met a trigger x 100.
Independent case review against versioned rules plus routing timestamps. State sample size and trigger class. No trigger-positive cases means N/A.
Control owner + process owner / weekly
Pilot; operate
Investigate each material miss and repair detection or routing. Read alongside review workload: escalating everything can overwhelm the people who must act.
m11
m15
m22
Design the trigger, evidence, recipient and intervention right together.
14
Decision quality & control
Which eligible work can complete reliably within approved authority?
Eligible cases correctly completed without unplanned human handling, with required controls satisfied / all cases in the predefined eligible cohort x 100.
Workflow and control logs plus outcome validation after a fixed observation window. Include failures, fallbacks and unresolved eligible work in the denominator.
Process owner + control owner / weekly
Pilot; operate
Keep the eligibility rule visible and stable. Expand authority only when outcomes and controls support it; a higher percentage is not a goal by itself.
m07
m11
m13
Choose how much autonomy the operating evidence justifies.
15
Decision quality & control
Can an accountable reviewer reconstruct a consequential decision?
Sampled decisions with retrievable required context, policy/version, actor, action, rationale/evidence and accountable owner / decisions sampled x 100.
Decision log and trace retrieval tests. Define required fields by decision class; use authorized access and retention rules. Report missing-field patterns.
Control owner + platform owner / monthly
Pilot; operate
Treat missing evidence as an assurance gap. A complete trace supports review; assess correctness through outcome checks before treating it as trustworthy.
m10
m12
m13
Make evidence usable by the people who hold accountability.
16
Knowledge reliability
Are the knowledge dependencies of the workflow fit for use?
Registered critical knowledge items with a named owner, approved valid version and in-date review / all registered critical items required by the workflow x 100.
Dependency inventory, ownership register and review records. Report unregistered dependencies and missing high-criticality items separately.
Knowledge owner / monthly + material-change review
Design; build; operate
Prioritize gaps by decision impact. A high aggregate does not compensate for one missing policy essential to a high-impact decision.
m17
m18
m19
Define authoritative knowledge and the responsibility for maintaining it.
17
Knowledge reliability
Does the output use current authorized evidence correctly?
Sampled outputs whose material claims are supported by applicable authorized sources and correctly reflect them / outputs evaluated x 100.
Representative test cases, source versions and expert review rubric. Separate factual support from task usefulness; report refusals, omissions and sample size.
Knowledge + quality owner / weekly
Build; pilot; operate
Inspect unsupported claims, wrong context and omissions. Passing this test is only one input to an operating decision; verify the resulting outcome.
m10
m11
m16
Connect evidence quality with the decision the work needs.
18
Knowledge reliability
How often does a knowledge defect contribute to failed work?
Reviewed cases with a validated failure contribution from missing, stale, conflicting or inaccessible knowledge / all cases reviewed x 1,000.
Case review plus root-cause investigation. Permit multiple contributing causes but count each affected case once; disclose unresolved diagnoses.
Knowledge + process owner / monthly
Pilot; operate; improve
Fix the knowledge dependency and retest the affected case class. A reduction may reflect less investigation, so monitor review coverage alongside the rate.
m07
m16
m19
Locate the dependency that repeatedly limits reliable execution.
19
Knowledge reliability
How quickly does an approved change become usable throughout the work?
Median and 95th percentile elapsed time from knowledge-change approval to verified use by every identified in-scope consumer.
Change approvals, dependency map, propagation logs and acceptance tests. Show incomplete changes and their age; distinguish urgent and routine updates.
Knowledge + platform owner / monthly
Build; operate; improve
A fast central update can leave downstream consumers stale. Test retrieval and application in the workflow before marking propagation complete.
m16
m17
m18
Treat knowledge distribution as an operating responsibility.
20
People, adoption & ownership
Is the intended way of working being used across the eligible scope?
Eligible cases actually handled through the approved workflow / all eligible cases assigned to in-scope teams during the period x 100.
Process logs reconciled with intake and sampled off-system work. Define eligibility before rollout; include fallbacks and workarounds in the total.
Business change + process owner / weekly
Deploy; operate
Investigate usability, access and role-fit barriers. Adoption demonstrates use; benefits and quality must be measured separately.
m03
m07
m21
Build adoption into the operating routine and support model.
21
People, adoption & ownership
Can the people assigned to the work handle normal and exceptional situations?
Assigned people meeting the approved role-specific scenario standard / all assigned people required to demonstrate that standard x 100.
Observed task assessments covering normal work, challenge, escalation and fallback. Show unassessed people and assessment age; report at useful role/team level.
Business capability owner / before assignment + periodic review
Build; deploy; transfer
Use gaps to improve support, practice and role design. Course completion alone does not demonstrate the required judgment or intervention skill.
m13
m20
m22
Define the capability required by the changed work.
22
People, adoption & ownership
What human workload is required to keep the service dependable?
Active hours spent reviewing, verifying and correcting outputs / all eligible cases handled in the period x 100. Report hours per 100 cases.
Time sampling plus review, quality and correction logs, including off-system effort. Separate routine review from exception handling and queue waiting.
Operations + workforce owner / weekly
Pilot; operate
Higher effort can indicate necessary control or poor design. Check effectiveness, queue age and staff capacity before changing review requirements.
m04
m07
m13
Give human reviewers sufficient capacity, information and authority.
23
People, adoption & ownership
Can the business responsibly accept the operation from the project?
Critical operating responsibilities with a named accepting owner, authority, capacity and tested support route / all required critical responsibilities x 100.
Responsibility register, signed acceptance and rehearsal evidence. Include benefits, decisions, knowledge, controls, service support and improvement.
Business sponsor + service owner / each handoff + monthly
Design; deploy; transfer
Resolve any unowned critical responsibility before handoff. Confirm ownership in a real incident or rehearsal; a completed register is only supporting evidence.
m21
m24
m25
Make the transfer of authority and responsibility explicit.
24
People, adoption & ownership
Which unresolved commitments are putting the next operating step at risk?
Critical dependencies unresolved after their agreed due date / critical dependencies due in the review period plus unresolved carryover x 100.
Versioned dependency register with owner, due date and acceptance evidence. Show count and impact; log date changes and cancellations.
Transformation lead + business sponsor / weekly
Frame; design; build; deploy
Resolve the dependency or change scope and sequencing. Do not move due dates silently. Delivery readiness supports a decision; it does not prove operating value.
m16
m21
m23
Use the transformation phase guides to sequence the necessary commitments.
25
Resilience & improvement
Can the operation recover to an acceptable customer outcome?
Median and 95th percentile elapsed time from service loss/degradation onset to verified restoration of the agreed service level.
Incident timeline and service checks. Segment by severity; show open incident age. Separate restoration from permanent repair and drill results.
Service owner / each incident + monthly
Deploy; operate
Validate fallback and customer impact. Fast technical recovery is insufficient if affected work remains stuck or incorrect.
m08
m11
m26
Design recovery around the operating outcome and its dependencies.
26
Resilience & improvement
Are known failure causes continuing to disrupt the service?
Incidents in the period with a validated cause previously observed within the declared lookback window / all incidents in the period x 100.
Incident/problem register and consistent cause taxonomy. A 90-day lookback is an example, to set by context. Also show counts, severity and investigation coverage.
Service + problem owner / monthly
Operate; improve
Prioritize permanent fixes to repeat causes. Rising reporting can improve visibility; distinguish detection changes from a true deterioration.
m25
m27
m28
Turn recovery experience into changes that prevent recurrence.
27
Resilience & improvement
How quickly does an operating signal produce a demonstrated improvement?
Median and 95th percentile time from the first qualified signal to evidence that the implemented change meets its acceptance criteria.
Signal, decision, change and evaluation records. Predefine qualification and validation windows; report open signals, their age and stopped attempts.
Process improvement owner / monthly
Operate; improve
Deployment is a milestone within the interval. Preserve evaluation quality; inspect delays in decision rights, testing or ownership.
m07
m26
m28
Convert feedback into a governed and evaluated change.
28
Resilience & improvement
Does a demonstrated improvement endure after initial stabilization?
Validated benefit per comparable unit at the later review / validated benefit per comparable unit at the stabilization checkpoint x 100.
Same benefit definition, matched case mix and quality/control limits at both checkpoints. Set dates upfront; 30 and 90 days are examples. Non-positive initial benefit gives N/A.
Benefit + process owner / agreed review checkpoints
Improve
Investigate erosion from drift, workload, knowledge or support changes. Read the underlying benefit and guardrails; the ratio can exceed 100%.
m01
m07
m27
Check whether the improvement survives routine operating conditions.
No measures match. Try a broader word or clear the filters to see the full directory.
Use with judgment
Agree the eligible work, period, numerator and denominator, source, missing data and owner. Set targets and intervention triggers for your operating context.
Use a comparable baseline and examine case mix, uncertainty and open work. A before-and-after difference alone does not establish that the initiative caused it.
Separate cash, avoided cost and capacity. Pair outcomes with quality, control and human effort. Review severe incidents individually; a good average cannot cancel them.
These 28 measures are RePerspective Labs operational definitions. They support judgment and local design; they are not a certified standard, universal benchmark or maturity score.
Readiness and delivery evidence differ from achieved outcomes. Disclose sampling, unknown outcomes and relevant segments. Agree the owner and the action before reporting a KPI.
From measurement to action
I help leaders connect measures with the decisions, responsibilities and changes behind them.