By the end of this lesson
- Distinguish system performance, improvement in work and cash benefits.
- Calculate net time changes by role, and examine full costs and critical assumptions.
- Recommend continuation, narrowing, expansion or exit, with conditions that would change your judgment.
Before you begin
- Bring lesson 9’s eligible-task counts, completion records and workload changes, plus lesson 10’s incidents and repair plan.
- Read this lesson’s Beichen value scenario. Its annual volumes and scenarios are fictional teaching inputs, not estimates derived from the twenty introductory JSON records or real performance results.
Read the Beichen case and introductory data
Start with what the next investment is for
After a pilot, help the owner choose the next step rather than justify money already spent. Begin with a specific recommendation: retain day-shift scope while testing a critical assumption, narrow to useful and supportable tasks, expand after new evidence is available, or stop. State the resources required and when the decision will be reviewed.
Follow the steps from a suitable draft to actual use, faster or more reliable human completion, improvement in the wider workflow, and possibly lower spending or higher output. Each step needs evidence. A good draft does not speed approval by itself. An engineer saving five minutes of lookup may simply wait five minutes longer for a part.
GOV.UK’s benefits guidance distinguishes kinds of benefit and recommends updating cost and benefit estimates as research progresses. Treat the value case as a revisable judgment. Better user experience, quality and access can be worthwhile without being converted into cash. Explain who benefits, how and what supports the claim.
Unpack the assumptions behind the eight-million claim
The exercise value page claims twelve minutes saved per task, multiplied by 200,000 annual tasks, gives 40,000 hours; at RMB 200 per hour, that is RMB 8 million. The arithmetic is correct. The financial conclusion is not established: eligibility, actual use, unchanged quality, added review, maintenance and a realizable budget reduction remain unproven.
Check the source and scope of the annual volume, tasks entering and completing the new workflow, how the twelve minutes were measured, extra review and rework, and what happened to released time. An arbitrary fifty-percent discount does not resolve these unknowns. The twenty synthetic records help select questions; they cannot establish annual volume or returns.
For every parameter, record its source, period, inclusions, exclusions and the person who can provide better evidence. Completion comes from task records. Neither lesson 9’s login rate nor this scenario’s acceptance rate can substitute for it. Accepted suggestions may still leave work waiting for approvals or parts.
Calculate the net change and who does the work
For a worked calculation, assume 100 eligible tasks, of which 35 use the assistant and reach completion. Each completed task saves the engineer five minutes and adds four minutes of expert review. Assume other tasks add no review and quality and rework do not change. These are illustrative assumptions, not consequences established by the supplied acceptance rate. Engineers save 175 minutes, specialists add 140, and total work falls by 35 minutes.
Change added review to six minutes while holding those assumptions constant: specialists add 210 minutes, so total work rises by 35 minutes. If all 100 opportunities actually require review, count all 100—not just the 35 completions. State these conditions so someone can check the calculation rather than accept a headline about one minute saved.
Then examine staffing: are specialists already backlogged, can engineers use released time for other work, has quality improved, and who handles withdrawn knowledge and incidents? Adding minutes describes a total but does not make roles interchangeable. A workflow that saves frontline time while overloading specialists may still be unusable.
| Illustrative 35 completed tasks | Four extra review minutes each | Six extra review minutes each |
|---|---|---|
| Engineer time released | 35 × 5 = 175 minutes | 35 × 5 = 175 minutes |
| Additional specialist time | 35 × 4 = 140 minutes | 35 × 6 = 210 minutes |
| Total work-time change | 35 minutes less | 35 minutes more |
| Cash benefit established? | No | No |
Include build, operating and exit costs
Build costs include discovery, design, integration, initial evaluation and training. Operating costs include models and infrastructure, knowledge maintenance, review, support, permissions, reevaluation and incident handling. Exit requires unfinished-task transfer, access revocation, data handling and supplier closure. Where a price is unknown, show the work and quotation needed rather than assume zero.
Separate costs already incurred from additional costs relevant to the next decision. Past development spending does not justify continuing merely to avoid waste. A lifecycle comparison also should not hide build costs. Show past spending, future increments and the whole lifecycle separately, avoiding double counting one-off expense and depreciation. GOV.UK’s cost-per-transaction guide helps examine support across channels; this course’s lifecycle checklist is not a replacement for its particular public-sector accounting rules.
Keep effort and cash in different columns. Three review hours are workload; buying three additional outsourced hours may create a cash expense. Existing staff time has an opportunity cost—the other work forgone—but does not automatically reduce payroll. Explain allocations of shared support costs and retain uncertainty about rework and incident frequency and impact.
Match the claim to the evidence
“Waiting fell during the trial” is an observation. “The decline coincided with launch” is an association. “The assistant may have contributed after checking other changes” requires additional analysis. “The assistant caused a particular reduction” needs a design able to rule out major alternative explanations. A stronger verb does not upgrade the evidence.
Later in Beichen’s scenario, review actually adds six minutes and the parts supplier has also changed. Update costs immediately, then examine whether waiting improved partly because of the supplier change. Without a comparison or sufficient records, report the observation and uncertain contribution, and propose evidence that distinguishes the explanations.
Keep task scope, equipment, shifts, source versions and calculation methods comparable across periods. Show excluded and unfinished work separately. Evaluate risk alongside value: faster work while wrong versions still affect high-temperature tasks cannot override lesson 8’s stop conditions.
Find the assumption that changes the recommendation
Sensitivity analysis changes important assumptions to see how the conclusion moves. The Beichen exercise supplies annual eligible volumes of 60,000, 90,000 and 140,000; acceptance of 20%, 35% and 55%; and per-use operator savings of three, five and twelve minutes. These scenarios explore uncertainty. None is an observed outcome, and acceptance is not completion.
First compare operator savings with added review per use. Under the illustrative assumptions that every use completes, only those tasks need review and no other work is added, four review minutes give net changes of −1, 1 and 8 minutes. Six review minutes change them to −3, −1 and 6. Positive values mean less time; negatives mean more. Maintenance, quality and cash realization are absent, so this is not an annual return forecast.
This identifies questions more urgent than multiplying annual volume: actual review burden, the gap between acceptance and completion, and whether released time can be reused. If continuation works only under optimistic assumptions, narrow scope and test the critical parameters. Ask business and operations owners to agree the cost of collection and review point instead of simply asking for “more data.”
| Teaching scenario | Operator minutes released | Net with four-minute review | Net with six-minute review |
|---|---|---|---|
| Conservative inputs | 3 minutes | −1 minute | −3 minutes |
| Middle inputs | 5 minutes | 1 minute | −1 minute |
| Optimistic inputs | 12 minutes | 8 minutes | 6 minutes |
Finish with a clear recommendation, not a large total
A one-page note can cover the decision requested, permitted scope, three important outcome and workload findings, one adverse fact, distinct options, your recommendation and reversal conditions, resources and review date. State the recommendation before its justification so the owner knows what this meeting must answer.
Common errors are omitting specialist workload, converting salary-weighted minutes into cash, double-counting waiting within resolution time, or retaining the invalid eight-million headline. If expansion to South China is requested, list fresh equipment, permission and staffing checks. Existing day-shift value cannot guarantee another region. Investment approval does not mean the new scope has passed operational evaluation.
Practice the method
Rewrite Beichen’s value page as a note that lets an owner choose the next step. Paper or a spreadsheet is sufficient; no real budget or production data is needed.
- 01
Unpack the eight-million formula. Mark volume, use, completion, time saved, review and cash realization as known, assumed or unsupported.
- 02
Recalculate the illustration with 100 opportunities and 35 completions using four and six review minutes. Explain how the formula changes if every task needs review.
- 03
List build, ongoing and exit costs. Separate cash and effort, including extra specialist, knowledge-maintenance and support work.
- 04
Add the supplier change as an alternative explanation. Write one claim supported now and another not yet supported.
- 05
Compare retaining scope, narrowing for verification and stopping. State a recommendation and reversal condition. Treat South China expansion as a separate discovery decision; leave real approval to authorized owners.
Identify the missing inputs in a value estimate
Compare three acceptance assumptions and two review durations, separating conditional per-task arithmetic, effective completions and the costs of all attempts.
Acceptance is not effective completion. Missing completion and review counts stay blank, so these assumptions cannot establish annual net value.
Read the field guide, work through the calculation or walkthrough, then check the guide's review notes. Open the downloaded files in a spreadsheet or text editor; no code is required.
Check your understanding
Write your answer before opening the explanation, then check what you might have missed.
If the eight-million arithmetic is correct, why is it not an established benefit?
Eligibility, completion, quality, added cost and realization remain unverified. A correct expression does not establish its inputs or a cash reduction.
Of 100 opportunities, 35 complete; each saves five minutes and adds six review minutes. What is the net change?
Assuming only those 35 add review and no other work is added, total work increases by 35 minutes. If other tasks need review, recalculate; the increase may be larger.
If the supplier changed at launch too, can you attribute the lower waiting time to AI?
Coincidence alone is insufficient. Report the observation, retain the alternative explanation and seek records or a comparison design that distinguishes causes.
Your record and next step
A sourced value case with full costs and an investment recommendation tested against changes in key assumptions.
Avoid double-counting benefits or hiding added work and worse-performing groups. Leave the investment decision to an authorized owner.
Take the recommendation, ongoing costs and open issues to lesson 12. Whether continuing or exiting, verify that the receiving team can operate and responsibilities are owned; pilot funding and the original team are not permanent support assumptions.