# 5 Synthetic Workflow Scenarios **Status:** Public-safe synthetic examples **Task ID:** REVIEW-001D / SCENARIO-002 **Date:** 2026-07-08 This file shows five synthetic workflow scenarios. They are not real customer cases. The goal is to make the prototype easier to understand by showing the same pattern repeatedly: ```text messy input → safe small test → visible output → human check → reusable capability ``` --- ## How to read these scenarios Each scenario uses only synthetic or public-style sample content. The examples are intentionally small so a reviewer can see what the workflow is trying to prove without needing private data, production access, or a real customer case. Start with the comparison table, then read the full scenarios only if you want more detail. --- ## One-page comparison table | Scenario | Pain point | Safe first test | Human check | Reusable capability | |---|---|---|---|---| | FAQ grouping | Repeated questions are mixed together and hard to prioritize. | Group synthetic questions into topic labels only. | Check whether topic labels are understandable and correctly grouped. | FAQ Topic Capability | | Appointment notes | Free-text messages contain time, person, and follow-up details. | Extract appointment fields for review, without updating a real schedule. | Check whether fields match the original message before any real change. | Appointment-Service Capability | | Daily work notes | Daily notes mix delays, checklist items, follow-ups, and review tasks. | Group notes into simple operational categories. | Check whether grouping is useful and whether owners are needed. | Daily Note Capability | | Small-test value check | A broad AI idea may jump to sensitive data or customer-facing automation too early. | Turn the vague idea into one small synthetic-data experiment. | Check whether the test avoids sensitive data and external automation. | Small-Test Value Capability | | Quality gate review | A capability draft may sound useful but lack reusable structure. | Check whether the draft has examples, language coverage, and review rules. | Check whether the capability is simple, visible, and not overclaiming. | Capability Pack Quality Gate | --- ## Scenario 1: FAQ grouping ### Before A team has repeated simple questions, but the questions are mixed together and nobody knows which topics should be turned into FAQ entries first. ### Synthetic input ```text Can I change my appointment time? Where can I find the refund policy? Do you support weekend pickup? Can I update my phone number? ``` ### Safe first test Group similar questions into simple topic labels without answering customers directly. ### Example output ```text Appointment changes: - Can I change my appointment time? Policy questions: - Where can I find the refund policy? Pickup availability: - Do you support weekend pickup? Account/contact updates: - Can I update my phone number? ``` ### Human check A human reviewer confirms whether the topic labels are understandable and whether any question is placed in the wrong group. ### Capability shown ```text FAQ Topic Capability ``` ### Reusable when This capability can be reused when a team has repeated questions or messages and wants to group them before writing answers, automating replies, or touching real customer data. --- ## Scenario 2: Appointment notes ### Before Short appointment messages contain useful details, but date, time, person, and follow-up notes are mixed inside free text. ### Synthetic input ```text Alex wants to move the Tuesday 3pm visit to Friday morning. Please call back before noon. ``` ### Safe first test Extract appointment fields for human review. Do not update the real schedule automatically. ### Example output ```text Person: Alex Original time: Tuesday 3pm Requested change: Friday morning Follow-up: Call back before noon Automation allowed: No Human review required: Yes ``` ### Human check A human reviewer checks whether the extracted fields match the original message before any real schedule is changed. ### Capability shown ```text Appointment-Service Capability ``` ### Reusable when This capability can be reused when a workflow contains short free-text service notes that should be converted into reviewable fields before any real schedule, CRM, or production system is changed. --- ## Scenario 3: Daily work notes ### Before A small team keeps daily notes, but delays, material issues, follow-ups, and review items are mixed together. ### Synthetic input ```text Paint delivery was late. Front desk checklist done. Need to call supplier tomorrow. Review photos before closing. ``` ### Safe first test Group notes into simple operational categories. ### Example output ```text Delay: - Paint delivery was late. Checklist: - Front desk checklist done. Follow-up: - Need to call supplier tomorrow. Review: - Review photos before closing. ``` ### Human check A human reviewer checks whether the grouping is useful and whether any item needs a responsible owner. ### Capability shown ```text Daily Note Capability ``` ### Reusable when This capability can be reused when a team has messy daily notes and wants to turn them into reviewable categories before assigning owners, building SOPs, or connecting task tools. --- ## Scenario 4: Small-test value check ### Before A team has a vague AI idea, but the first step is too broad and may touch sensitive data too early. ### Synthetic input ```text We want to use AI to improve customer service. ``` ### Safe first test Turn the vague idea into one small, reviewable, non-production experiment. ### Example output ```text Unsafe first step: - Let AI answer real customers automatically. Safer first test: - Use 20 synthetic customer-service questions. - Ask AI to group the questions by topic. - Do not send answers to customers. - Have a human review the topic groups. Reason: - This tests usefulness without exposing real customer data or customer-facing output. ``` ### Human check A human reviewer confirms whether the test avoids sensitive data, avoids customer-facing automation, and produces a result that can be checked manually. ### Capability shown ```text Small-Test Value Capability ``` ### Reusable when This capability can be reused when an AI idea is too broad and needs to be reduced into a safe first experiment before real data, customers, or production systems are involved. --- ## Scenario 5: Quality gate review ### Before A drafted capability pack may sound useful, but it is not clear whether it is reusable, explainable, and reviewable. Plain-language version: ```text The idea sounds good, but we cannot tell whether another person can reuse it safely. ``` ### Synthetic input ```text Capability draft: Customer Message Sorting Capability Purpose: Sort customer messages into topics. Example layer: missing Traditional Chinese explanation: missing English explanation: present Human review rule: unclear ``` ### Safe first test Check whether the capability draft has enough structure to be reused safely. ### Example output ```text Value card: Present Example layer: Missing Traditional Chinese version: Missing English version: Present Human review rule: Unclear Gate result: Needs revision Next safe patch: - Add a small synthetic example. - Add Traditional Chinese explanation. - Define what a human must review before use. ``` ### Human check A human reviewer confirms whether the capability is simple enough, has a visible example, and does not imply production automation. ### Capability shown ```text Capability Pack Quality Gate ``` ### Reusable when This capability can be reused when a workflow ability sounds useful but still needs a visible example, language support, human review rule, and public-safe boundary before being shared or reused. --- ## Boundary ```text Synthetic examples only. No real client data. No private company data. No production workflow. No live automation. No claim that these are real case studies. ```