---
name: "assess-workflow-value"
description: "Check whether the workflow reduces human effort after review, correction and maintenance, then decide whether to keep, revise or stop it."
version: 1
updated: "2026-11-04"
requires_human_review: true
tooling_summary: "No MCP is needed with supplied measurements and workflow context. Optional scoped read access can help establish case counts, dates and effort evidence."
model_guidance_reviewed: "2026-11-02"
model_guidance_catalog: "https://jasminejames.dev/leading-deeper/model-guidance.json"
source: "https://jasminejames.dev/leading-deeper/assess-workflow-value"
---

# Check whether the workflow is helping

Check whether the workflow reduces human effort after review, correction and maintenance, then decide whether to keep, revise or stop it.

New companion exercise for the talk resources; not a historical result or the original internal prompt. Adapt this to your context before use.

## Use as an agent skill

[Check whether the automation is helping](https://raw.githubusercontent.com/jasminelsjames/leading-deeper-ai-skills/e1d2e65d3922c5e376fc0c6e8cfbb2f85722b1a7/skills/assess-workflow-value/SKILL.md): Compare preparation, review, corrections and upkeep before deciding whether to keep an automation. Reading the skill does not grant account access or permission to execute changes.

## Start here: one first run

Check the effort behind one workflow.

Use the AI tool approved for your work with supplied material. Wording and layout will vary by model; look for the evidence, uncertainty and review decisions shown here. Connections and recurring runs depend on your setup.

All examples use invented sources and results. They are illustrations, not reports from Jasmine’s workplace or measured outcomes.

## Fictional inputs

**Comparable work | Fictional week:** Five completed drafts per approach, with the same acceptance requirements assumed. No failed or unfinished cases in this illustration. Effort is active person-minutes, not agent runtime or elapsed time.

**Effort records | Invented measurements:** Manual work: 25 minutes per draft including preparation and checks, with zero separate baseline maintenance explicitly assumed. Assisted work: 7 minutes per draft including preparation, review and corrections, plus 10 minutes of maintenance for the week.

**Unknowns and request:** Setup effort, monetary cost and use of attention are unknown. Equal quality is assumed for the example, not independently verified. Return a decision draft and one next measurement; make no workflow or staffing changes.

## Workflow

1. Human — Supply a fair comparison: Name the completed task, period and quality requirements. Provide human effort for both approaches and label assumptions, estimates and missing values.
2. AI — Prepare a decision draft: Explain what the effort comparison can establish, showing a traceable calculation. Separate the recurring difference from unknown setup, cost, quality and impact.
3. Human — Choose the next measurement: Check comparability and arithmetic, then decide whether the evidence supports keeping, revising or stopping. Unknown effort stays unknown rather than becoming zero.

## Illustrative output

Workflow assessment | Five comparable drafts

**Recurring human effort:** Manual: 125 active person-minutes. Assisted: 45 active person-minutes including maintenance. Difference: 80 minutes before setup. [Supplied fictional effort records]

**Quality and unknowns:** Equal acceptance requirements are assumed for the illustration, not independently validated evidence. Setup effort and monetary cost are unknown.

**Recommendation:** Revise the assessment: collect setup effort and real comparable case measurements before claiming an all-in benefit.

**Next evidence:** Have the workflow owner record active preparation, review, corrections and maintenance for the next comparable completed cases; retain failures and exclusions.

## What to look for across models

- Comparable cases, units and a traceable calculation
- Assumptions, missing evidence and setup treatment
- Quality, monetary cost and use of attention kept distinct from effort
- A reviewed keep, revise or stop proposal and one next measurement

## Human review

- Were the cases comparable, and did you count all human effort once, including unsuccessful attempts in your real comparison?
- Do the known results support the decision, or do you first need setup, quality or other missing evidence?

## MCP and tool requirements

No MCP is needed with supplied measurements and workflow context. Optional scoped read access can help establish case counts, dates and effort evidence.

### Read optional effort and case evidence

- Needed when: Only when approved measurements or records must be retrieved rather than supplied.
- MCP or tool examples: Approved local CSV, spreadsheet or timing-log tools; An approved reporting or issue-tracker connector or MCP; An approved calendar connector or export for context only
- Tools must support: Read the selected timing measures, completed-case records and quality findings; Check definitions, case counts, dates and coverage; Record provenance and distinguish active effort from calendar or system elapsed time
- Access needed: Read access to the named records and reporting window. Calendar duration or task state alone is not an effort measure; no edits, messaging, administration or tracking setup is required.
- Without the connection: Supply selected timing notes, case outcomes and known overhead in a small table, or leave unknowns explicit and plan the next measurement.

Provide an approved aggregate table, timing notes or excerpts with dates, units, scope and provenance. Keep missing values unknown and propose one small manual measurement.

## Model starting point

Reviewed November 2, 2026.

Start with GPT-6 Luna (gpt-6-luna). A starting point for organizing a small supplied evidence table and explaining deterministic calculations. Check the arithmetic with a calculator or code and review the assumptions. Start with low reasoning effort.

Move up to GPT-6.1 Sol (gpt-6.1-sol) when: The cases are hard to compare, records conflict or the decision needs a deeper examination. A stronger model cannot recover unmeasured effort.

### Alternatives

- Claude Haiku 5.5 (claude-haiku-5-5): An alternative for structuring supplied evidence in an approved Claude host. Compare review effort and complete workflow cost on your own cases.

### Human responsibility

The owner decides whether to keep, revise or stop. A model cannot establish savings from missing measurements or infer that released effort produced customer value or growth.

Suggested fit based on official provider documentation, not our own speed or quality benchmarks. Test on representative examples before relying on a model.

A model does not connect your tools or grant access. MCP support, available tools and model selection depend on the agent host and your organization's approved setup.

### API price reference

USD per one million input and output tokens, using standard first-party API pricing. These are not subscription prices or total workflow costs. Tool calls, reasoning, retries, caching and different tokenizers affect the final bill.

### GPT-6 Luna

Provider: OpenAI
API model ID: gpt-6-luna
Context: 1.05 million tokens
Standard API: $0.10 input / $0.50 output per 1M tokens.
Model documentation: https://developers.openai.com/api/docs/models/gpt-6-luna
Pricing: https://developers.openai.com/api/docs/pricing

### GPT-6.1 Sol

Provider: OpenAI
API model ID: gpt-6.1-sol
Context: 1.05 million tokens
Standard API: $2 input / $10 output per 1M tokens.
Model documentation: https://developers.openai.com/api/docs/models/gpt-6.1-sol
Pricing: https://developers.openai.com/api/docs/pricing

### Claude Haiku 5.5

Provider: Anthropic
API model ID: claude-haiku-5-5
Context: 1 million tokens
Standard API: $0.10 input / $0.50 output per 1M tokens. For requests with prompts up to 100,000 tokens.
Standard API: $0.50 input / $2.5 output per 1M tokens. For requests with prompts over 100,000 tokens; the higher rates apply to that request.
Model documentation: https://platform.claude.com/docs/en/models/haiku-5-5/overview
Pricing: https://platform.claude.com/docs/en/about-claude/pricing

Machine-readable catalog: https://jasminejames.dev/leading-deeper/model-guidance.json

## Context to supply

- Workflow step, intended outcome, owner, people involved and protected judgment or development
- Definition of a comparable completed case, acceptance checks and reporting period
- Manual baseline and assisted human effort, including remaining preparation, review, correction, interventions and handoffs
- Recurring maintenance and one-time setup effort, with case counts or an explicit amortization horizon where known
- Known tool costs, quality findings and evidence of what available attention was used for
- Source, date, unit, coverage and measured or estimated status for each input; unknowns may remain unknown

## Prompt

```text
Assess one AI-assisted workflow and return a decision draft. This is a new companion exercise, not evidence that the talk's examples produced any particular saving.

INPUTS
Workflow, intended outcome and owner: [fill in]
People involved and judgment or growth to protect: [fill in]
Completed-case definition and acceptance checks: [fill in]
Reporting period and comparable manual/assisted cases: [fill in]
Active human effort and its components: [supply measurements or bounded estimates]
Maintenance, setup, supported case counts and any proposed amortization horizon: [fill in or mark unknown]
Known costs, quality findings and use of attention: [fill in or mark unknown]
Approved evidence and its provenance: [supply records, excerpts or scoped read locations]

COMPARE
Compare like-for-like completed cases with similar scope, complexity and quality requirements. Make volume, staffing and case-mix differences visible. Label each input measured, recalled, estimated or assumed, with its source, date, unit and coverage. Missing evidence remains unknown; it is never zero by default.

Use active person-minutes, including preparation, review, correction, interventions and relevant handoffs. Count each person's effort once. If a review total already includes corrections, do not add them again. Do not infer effort from message counts, assistance percentages, scheduled calendar duration, generated outputs or elapsed agent runtime. Include unsuccessful attempts and rework within the effort window. Do not compare only successful assisted cases with the full manual baseline; show unfinished or excluded cases and limit conclusions when a fair completed-case comparison is unavailable.

Allocate recurring maintenance across the cases it supports. Amortize setup only when its effort and an explicit supported or proposed case horizon are known; distinguish projection from observation. Never charge setup twice. If setup is unknown or excluded, label the result a recurring-effort comparison before setup, not an all-in saving.

In minutes: baseline effort/case = baseline case effort + allocated baseline overhead; assisted effort/case = assisted human effort + allocated assisted overhead. Net recurring reduction/case = baseline effort/case minus assisted effort/case. All-in reduction/case also subtracts allocated setup. A period total multiplies the relevant per-case result by a supported comparable case count. Preserve negative results. Use actual totals or suitable summaries when case sizes differ. If inputs are bounded estimates, calculate conservative bounds and state their provenance. Convert minutes to hours only after computing the scoped total. Do not extrapolate beyond the supplied frequency and period without labeling the assumptions.

VALUE
Keep effort, monetary cost, quality and use of attention separate. Known API/tool/hosting charges are not human time. Do not monetize effort without a supplied rate and stated assumptions. Review acceptance checks, errors, coverage and rework; AI's self-assessment is not independent quality evidence. Reduced effort does not establish customer value, growth or wellbeing. Distinguish what the owner intended to make room for from what actually happened. Routine work for one person may be useful practice for another; protect judgment, relationships, learning, ownership and reviewer capacity.

OUTPUT
Return a concise conclusion, calculation or comparison, provenance, quality/cost findings and observed use of attention. Separate supported observations, bounded estimates and unknowns; state setup treatment and material exclusions. Propose keep, revise or stop against the intended outcome and acceptance conditions. When evidence is insufficient, propose a bounded revision or measurement rather than declaring success. Name one small next measurement, owner, scope and review point.

STOP
Work with supplied data; no MCP is compulsory. Optional external evidence needs scoped read access to the agreed accounts, measures and period. Return the draft here unless a private destination is authorized. Do not change workflows, assign people, publish results, contact colleagues or create tracking or scheduled automation.
```

## What a person reviews

- Are the cases and quality requirements comparable, and are every input, exclusion and estimate traceable?
- Does human effort include preparation, review, correction and overhead exactly once, with unknown setup kept distinct?
- Does the conclusion separate effort, cost, quality and actual use of attention while protecting judgment and person-relative growth?
- Is the proposed keep, revise or stop decision proportionate to the evidence?

## Evidence to check

- Use active human effort for comparable completed cases; retain unsuccessful attempts, rework and case-mix or coverage limits.
- Check maintenance allocation and setup treatment; a before-setup comparison does not establish an all-in benefit.
- Compare quality and known costs for the same scope. Record what available attention supported without assuming a customer, growth or wellbeing result.
- If evidence is missing, take one small prospective measurement rather than deriving hours from messages, assistance rates or calendar time.

## Result and value

Compare preparation, review, corrections and upkeep before deciding whether to keep an automation.

**Human judgment:** You decide whether the evidence supports keeping, changing or stopping the automation, and where any released capacity should go.

**Value to check:** Compare similar tasks and periods, count all human effort, record quality tradeoffs and label estimates separately from observed results.

## Connection to the talk

A new measurement companion

**Practice:** Use a repeatable Softball IQ task, such as a debrief or content review, as a proposed measurement exercise.

**Work:** Apply the same comparison to a workplace workflow, including preparation, review, correction and the outcome people need.

## Fictional example

[Read the example JSON](https://jasminejames.dev/leading-deeper/assess-workflow-value-example.json). This is invented material, not a company report.
