PrerequisiteCompleted Module 2 (Process and Baseline)
0 / 2 completed
Learning objectives
Use the SDD’s value-calculation formula to estimate minutes saved per case, monthly released hours, and capacity value.
Find the common error in a metric card and write a corrected version.
Label numbers as "estimate" or "validated" and never present an estimate as a guarantee.
Value Calculation and Metric Cards
25 min
The Value-Calculation Formula
Value-calculation formula
Minutes saved per case = baseline full-process minutes − AI-assisted full-process minutes
Monthly released hours = minutes saved per case × monthly case volume ÷ 60 × actual adoption rate
Estimated capacity value = released hours × assumed hourly cost
Estimated net value for a period = attributable, realized benefit − all incremental cost in the same period
Warning
The "AI-assisted full process" must include input preparation, waiting, review, correction, and rework, not just the few seconds it takes AI to generate an answer. Capacity value is not the same as cash savings, and double-counted benefits must never be added together.
Worked example (synthetic data, from SYN-CS-001)
Baseline full process = 21 minutes (BL-001 from Module 2)
AI-assisted full process = 6 minutes (including AI generation, human verification, and necessary correction)
Minutes saved per case = 21 − 6 = 15 minutes
Assume monthly volume = 800, actual adoption rate = 70%
Monthly released hours = 15 × 800 ÷ 60 × 0.7 = 140 hours
Assumed hourly cost = NT$450 (see cost_assumptions.csv, an estimate)
Estimated capacity value = 140 × 450 = NT$63,000/month (an estimate, not a guaranteed cash figure)
Key takeaway: Estimated capacity value is the value of hours freed up to do other work, not a cash-savings figure that would appear on a financial statement; the two must never be conflated.
35 min
Exercise: Find the Error in a Metric Card
Below is a first draft (v1) of a metric card. Try to spot at least one problem yourself before reading the revised version (v2) and its correction notes. This is this course’s "one fully worked example with a deliberate error and its correction notes" exercise.
Metric card v1 (a flawed draft)
Metric name: Improved support efficiency
Target: Improve support efficiency
Figure: Efficiency improved 30% after AI adoption
Data source: A verbal report from the support manager
Owner: (blank)
Metric card v2 (corrected)
Metric name
Average handling time for a knowledge Q&A case (minutes/case)
Definition
Total minutes from when an agent starts searching the knowledge base to when the reply is sent, including search, verification, drafting, and any necessary rework.
Unit and numerator/denominator
Unit: minutes/case; numerator = total handling minutes; denominator = number of cases.
Data source and period
baseline_log.csv (2026-01 to 2026-03, 30 synthetic observed samples).
Estimate/validated label
Currently an "estimate"; 30 samples is a teaching-material scale, not yet an enterprise acceptance sample size, and must be remeasured with real data before formal adoption.
Owner and limitations
Owner: the project’s technical lead (synthetic role). Limitation: does not cover multi-turn conversations or non-text input.
Warning
Correction notes: v1’s "30% efficiency improvement" has no unit, no numerator/denominator, no data source, and no estimate/validated label: it violates IL 101’s quality rule against "just writing improved efficiency." v2 replaces it with a decomposable, verifiable "handling time (minutes/case)" figure, plus a data source, period, estimate label, and owner: a metric card that can survive being questioned in a gate review.
1Exercise Write a metric card for your own project: first write a v1-style version that "sounds reasonable but cannot be verified," then revise it using v2’s six fields.
2Self-check Check whether your metric card has a unit, numerator/denominator, data source, observation period, owner, and an estimate/validated label; it is only complete once all six are present.
Check yourself
Metric card v1 says "Efficiency improved 30% after AI adoption." Which IL 101 quality rule does this most clearly violate?
Key takeaway: A good metric card must let someone who was not on the project verify how the number was calculated; a number that cannot be verified does not belong in a charter.
