Skip to content
NTPU AI Learn

IL 302 · Module 1

Investment and Value

Make keep/stop investment decisions across a project portfolio, distinguish capacity-released value from actual cash benefit, and plan resource allocation.

Start Section 1
  • Duration240 min
  • LevelArchitecture
  • PrerequisiteIL 301 recommended, or documented enterprise AI pilot experience
0 / 4 completed

Learning objectives

  • Produce a keep, stop, or redesign investment recommendation within a portfolio of multiple AI projects.
  • Distinguish capacity-released value from actual bookable cash benefit, avoiding double counting.
  • Plan resource allocation across projects based on evidence quality and strategic alignment.

Keep/stop decisions across the project portfolio

60 min

After one successful pilot, there is a whole portfolio to manage

IL 301 handles a single project's pilot and adoption decision. By the time an organization reaches IL 302, it usually has several AI projects running at once, at different stages: some have already piloted successfully, some are still prototypes, and some should probably have been stopped already but nobody made that call. Portfolio management answers one question: where should limited money, people, and attention go, and which projects should they be pulled out of?

Four things a keep/stop/redesign decision must look at
DimensionQuestion to ask
Value evidenceIs there traceable evidence the project keeps producing value, or is all that is left the original assumption?
Ongoing costIs the ongoing cost of operations, review, and exception handling still lower than the value it produces?
Opportunity costIf the people and budget this project currently uses were moved to another project, would the whole portfolio be worth more or less?
Strategic alignmentIs the project still aligned with the organization's current strategy, or has the strategy shifted while the project stayed the same?

Warning

Common mistake: treating "it still technically runs" as a reason to keep funding it. A project can run perfectly well technically while its value has already been overtaken by a better alternative, or its operating cost now exceeds its benefit; "it still runs" is not, by itself, a reason to keep it.
  1. Step 1
    List every AI-related project the organization currently runs, and tag each one's stage (pilot / limited adoption / full adoption / stopped but not closed out).
  2. Step 2
    Fill in value evidence, ongoing cost, and strategic alignment for each project, tagging the evidence level (estimate / verified).
  3. Step 3
    Draft an initial keep, stop, or redesign recommendation for each project, to be backed with numbers in the next slide's value ledger.

Key takeaway: A portfolio decision looks at the value of the whole portfolio; it is not enough for each project to individually prove it is still alive.

Capacity-released value vs. actual cash benefit

60 min

Capacity released is not cash benefit: record them separately

The most common AI-project value claim is 'we saved this many hours.' But hours saved is only capacity released: that time has been freed up, which does not by itself mean the organization spent one dollar less, nor that the freed time was actually redirected to something more valuable. Capacity only becomes a realized cash benefit once it is actually cashed in: reduced overtime pay, avoided new hiring, or the same team taking on more work with no added headcount.

Not an adequate value claim

This solution saves 120 hours of manual work per month, equivalent to saving NT$XX in labor cost per month.

Adequate value-ledger wording

Capacity released
120 hours of manual work released per month (estimate, from the baseline vs. AI-assisted full-process minute difference, at a 70% actual adoption rate).
Where the capacity went
40 of those hours were used by the same team to work through cases that would otherwise have been delayed (traceable to the case-volume log). The remaining 80 hours currently have no confirmed destination.
Actual cash benefit
Only the 40 hours tied to increased case throughput can be counted as real work on the backlog for now. There is no evidence of reduced headcount or avoided hiring, so the cash benefit is NT$0 for this period and must not be claimed as equivalent to a labor-cost saving.
Required Value Ledger fields
FieldWhat it means
Minutes saved per itemBaseline full-process minutes minus AI-assisted full-process minutes (AI-assisted must include input prep, waiting, review, correction, and rework)
Monthly capacity releasedMinutes saved per item x monthly item volume / 60 x actual adoption rate
Where the capacity wentWhat the released time was actually used for: backlog work, new business, less overtime, or no confirmed destination yet
Actual cash benefitOnly count benefit that is attributable, realized, and not double-counted (e.g., an actual reduction in overtime pay, or actual avoided hiring)
Incremental cost for the same periodOngoing cost of tool licenses, operations, review, exception handling, and data preparation

Warning

Common mistake: multiplying the estimated released capacity directly by an hourly rate and writing the result into the investment recommendation as an already-realized cash benefit. Capacity value is not the same as cash savings; financial assumptions and payback periods must be labeled as estimates with sensitivity ranges, and never presented as a guaranteed return.

Key takeaway: Capacity being freed up does not mean money was actually saved; the value ledger has to say clearly where the freed capacity ended up.

Resource allocation across projects

60 min

Draw resource allocation as a map: scale, sustain, reduce, exit

Once every project has value evidence and cost figures, the next step is to lay out the portfolio and decide how to allocate resources. Not every project deserves more investment, and not every mediocre project should be stopped immediately. Sorting projects into four allocation categories keeps the conversation focused on what to do next, instead of re-litigating whether each project is good.

Four resource-allocation categories
CategoryMeaningTypical action
ScaleStrong value evidence, manageable cost, high strategic alignmentIncrease budget, staffing, or expand the scope of applicability
SustainStable value with limited further growth, reasonable operating costKeep current investment level, revisit periodically for a scale opportunity
ReduceValue is insufficient to justify the current level of investment, but some value remainsNarrow the scope, lower operating cost, or reassign people to another project
ExitWeak value evidence, cost now exceeds benefit, or strategy no longer needs itFormally stop it and record the rationale; released people and budget move to other projects

Info

Exit is not a record of failure; it is a normal outcome of resource allocation. A stop decision backed by clear evidence that a project is no longer worth continuing is a completely different quality of decision from a vague stop that nobody is willing to actually make; the former is responsible portfolio management, the latter is the real problem.
  1. Step 1
    Sort every project listed in Module 1 Slide 1 into one of the four categories, based on its value-ledger result.
  2. Step 2
    For every project in Reduce or Exit, state explicitly what people and budget are being released and which project or use they are being redirected to.
  3. Step 3
    Bring this allocation map to the sponsor and governance owner for review, alongside the value ledger, as the basis for the investment recommendation.

Key takeaway: Resource allocation is not about scoring each project; it is about deciding where the money and the people go next.

Example: building the portfolio and value ledger

60 min

Example (SYN-CS-001): the customer-service knowledge assistant's portfolio and value ledger

The following uses the shared synthetic case SYN-CS-001 (the example company "Demo Service Company A") to walk through a simplified portfolio and value ledger. Demo Service Company A's customer-service knowledge assistant has completed its IL 301 pilot and received limited-adoption authorization.

Demo Service Company A (synthetic example): a snapshot of three AI projects' portfolio
ProjectStageValue evidenceRecommended allocation
Customer-service knowledge assistantLimited adoption (customer-service department)30 baseline records and 40 evaluation cases show shorter reply times, at a 70% actual adoption rateScale: evaluate expanding to other departments after completing the department-difference analysis
Internal document search toolPilot finished, adoption decision pendingOnly an internal demo exists; no control group or adoption-rate dataReduce: gather proper evaluation evidence first, pause further investment
Meeting-notes summarizerStuck in prototype for over six monthsNo baseline or evaluation record at allExit: no evidence base; recommend stopping and releasing the assigned people
Customer-service knowledge assistant value ledger (synthetic figures, illustrating the calculation)
FieldValue / explanation
Minutes saved per item8 minutes (baseline average 15 minutes minus AI-assisted average 7 minutes, including review and correction)
Monthly capacity releasedAbout 46 hours (8 minutes x 2,000 monthly items / 60 x 70% actual adoption rate)
Where the capacity wentAbout 30 hours went to working through previously delayed backlog cases; the remaining 16 hours have no confirmed destination yet and are logged as an open item
Actual cash benefitAn estimated NT$0 for this period: no evidence yet of reduced headcount or avoided hiring; only the improvement in backlog throughput is recorded
Incremental cost for the same periodAI tool licensing, reviewer time, and data-governance operating cost (synthetic assumption, see cost_assumptions.csv)

Info

Notice this example deliberately records the actual cash benefit as an estimated NT$0, rather than simply multiplying 46 hours by an hourly rate. That is the adequate wording this module requires: capacity released and cash benefit must be stated separately, and no benefit may be claimed as realized without evidence it was actually cashed in.
Check yourself

A project's value ledger states 'releases 100 hours of capacity per month, equivalent to saving NT$50,000 in labor cost per month,' without explaining where those 100 hours actually went. What is the biggest problem with this ledger?

A project's value ledger states 'releases 100 hours of capacity per month, equivalent to saving NT$50,000 in labor cost per month,' without explaining where those 100 hours actually went. What is the biggest problem with this ledger?

Key takeaway: An adequate portfolio analysis honestly separates which projects deserve to scale from what the released capacity actually turned into, rather than blending the two into one number.

Save as PDF

What next?