PrerequisiteComplete Module 4 (Testing and Failure Analysis), with a failure log in hand
0 / 2 completed
Learning objectives
List the full cost picture of an AI-assisted approach beyond tooling fees.
Convert the Module 4 failure analysis into human-review and exception-handling cost.
Complete a cost/value sheet with estimate tags, ready for the Module 6 Evidence Review.
Full cost and net value
60 min
Cost items that get overlooked
The cost of an AI approach is not just tool subscriptions or API fees. Sponsors are most often misled by seeing "the tool is cheap" without seeing how much human effort review, operations, and data preparation actually require.
Cost item checklist
Item
Description
Tooling/API cost
Model or tool usage fees, computed from usage
Human review cost
The rework and review time shown in the Module 4 failure log, converted to labor cost
Operations cost
Ongoing labor for monitoring, version updates, and document maintenance
Exception-handling cost
The extra time needed for human-handoff cases (Module 3)
Data-preparation cost
One-time and ongoing effort to clean, label, or refresh the knowledge base or training data
Warning
Common mistake: listing only the API fee for "the AI generating an answer" as the cost, without converting the human-review and rework time shown in Module 4 into a cost: leaving the sponsor thinking the approach is nearly free.
Key takeaway: A cost estimate that counts only tooling and ignores review and operations labor will lead the sponsor to underestimate the real investment.
60 min
Complete the cost/value sheet
Worked example (synthetic data, continuing the numbers from IL 101 Module 3 and this course's Module 4)
Capacity-value estimate (IL 101 Module 3) = NT$63,000/month (estimate)
Human review cost: per the failure log, missing-information and conflicting-sources categories together had 2 critical errors out of 40 cases = 5% needing major rework, estimated at an extra 20 review hours/month x NT$450/hour = NT$9,000/month
Tooling cost: estimated NT$4,000/month (per cost_assumptions.csv)
Data-preparation cost (one-time): knowledge-base cleanup and labeling, estimated 80 hours x NT$450 = NT$36,000 (one-time, not a monthly cost)
Net-value estimate for the period = 63,000 - 9,000 - 4,000 = NT$50,000/month (estimate), with the one-time data-preparation cost to be amortized separately
Warning
Common mistake: adding a one-time cost (e.g. 80 hours of data cleanup) directly to the monthly recurring cost without stating an amortization method, which understates or overstates net value. One-time and recurring costs must be listed separately.
Info
Every number must be tagged "estimate" or "verified," with its data source and observation period noted, continuing IL 101's quality rule; a number with no supporting evidence must never be presented as "verified."
1Exercise Complete a cost/value sheet for your own project: cover at least tooling, human review, operations, exception-handling, and data preparation, and compute a net-value estimate for the period.
Check yourself
A cost/value sheet adds "80 hours of knowledge-base cleanup" (a one-time cost) directly to the monthly tooling fee to produce a single "total cost" number. What is the problem?
Key takeaway: A net-value estimate must list one-time and recurring costs separately and clearly tag each number as estimate or verified before it can be a reliable input to the Evidence Review.
