Chapter 4: Your MPV AI Opportunity Framework

🎯 Chapter Objectives

Chapters 1–3 built the skill. Chapter 4 turns it into your contribution to MPV's AI roadmap. By the end of the theory you will know how to spot, rank, and pitch AI opportunities — and you will have starter ideas for every major MPV department to seed your thinking.

Apply the 6 Opportunity-Spotting Signals to find AI candidates in your real work.
Recognise the anti-signals — places AI is the wrong tool — and the moon-shot trap (HBR's hardest lesson).
Rank opportunities on the 5-dimension /10 scoring grid, with bands (Pilot / Revisit / Drop).
Browse top-5 starter ideas for Finance · Procurement · HR · Factory Ops · Distribution.
Understand what it means to be an MPV AI Seeder in Phase 2 and beyond.
Use the 90-day Phase 2 development plan structure to map your own journey.

📌 Chapter 4 Learning Snapshot

Spot

6 signals: Repetition · Time Sink · Reading · Translation · Patterns · Communication at Scale.

Rank

5-dim /10 grid. 8–10 pilot · 5–7 revisit · 0–4 drop.

Plan

Starter ideas per department + 90-day development plan + Phase 2 selection criteria.

4.1 The Opportunity-Spotting Checklist — Six Signals

Most employees do not know what AI can help with because they have never been asked the right questions. The six signals below are the single most useful tool you will take away from this program. The more signals a task hits, the stronger the candidate.

🔁

1. Repetition

You or someone in your team does this every day, every week, or every month. The shape rarely changes. Example: Weekly QA defect summary. Monthly variance commentary. Daily shift handover.

⏱️

2. Time Sink

The task takes more than 15 minutes per occurrence and the output is usually similar. Example: The 90-minute Monday QA report. The 60-minute weekly tender-summary email.

📖

3. Reading & Summarising

Most of the work is reading long documents and producing a short summary. Example: 40-page hospital tender → 1-page MPV response. Supplier audit reports → 3-paragraph QA brief.

🌐

4. Translation / Bilingual

The task crosses Vietnamese ↔ English (or Chinese / Korean for supplier work). Example: Translate Chinese resin technical spec to Vietnamese. English MOH tender → Vietnamese internal brief.

🔍

5. Pattern Detection

You stare at numbers trying to spot what changed, why, or what's coming. Example: Defect rate by shift/line/operator. Demand forecast variances. Supplier on-time delivery trends.

📢

6. Communication at Scale

Same conversations, many times. Example: Hospital delay notifications. Supplier follow-up emails. Internal HR FAQ ("how many leave days do I have?"). Onboarding answers.

The simple test: Walk through your last week. Highlight everything you did that hit 2+ signals above. Those are your AI candidates. Most employees find 3–8 candidates in 10 minutes once they know what to look for.

Quick Recap — The 6 Signals

Repetition Time sink Reading & summarising Translation Pattern detection Communication at scale

4.2 Anti-Signals — Where AI Is NOT the Answer

Equally important: knowing when AI does not belong. Forcing AI into the wrong places is the fastest way to burn credibility and lose budget. Five anti-signals.

Final safety / clinical decisions
Humans must decide. Per Golden Rule #1 + Vietnam AI Law Article 4 (human-centric).

Highly personal / interpersonal
Difficult employee conversations, customer recovery calls, condolences. Voice matters.

A 10-minute phone call is optimal
Some misunderstandings dissolve in one phone call. AI cannot replace that.

Tiny-volume workflows
A task done 1–2 times per year. Setup cost exceeds savings.

Data sensitivity makes safe patterns impractical
If every use needs heavy redaction AND blocks the enterprise plan path, skip it for Phase 1.

The Moon-Shot Trap — HBR's Hardest Lesson

Harvard Business Review's January–February 2018 article "Artificial Intelligence for the Real World" (Davenport & Ronanki) studied 152 cognitive-AI projects across 250 companies. The headline finding has aged exceptionally well: moon shots fail far more often than low-hanging fruit succeeds.

🖼️ The MD Anderson Story

HBR Case Study — MD Anderson Cancer Center

The moon shot

$62 million AI cancer-diagnosis system intended to be the hospital's flagship AI. Put on hold without ever being used on a patient.

The wins

Same hospital. Small AI projects: hotel recommendations for patients' families · identifying who needs billing help · IT support automation. Quietly successful — improved patient satisfaction, finances, freed nurse-manager time.

The lesson

Useful beats impressive. Compound small wins. Build the muscle. Then go bigger.

Apply to MPV: Do not start Phase 2 with "AI for predictive maintenance across every machine" or "AI-transform the factory". Start with one repetitive admin task (Category ①), one report you'd love to see (Category ②), or one question employees ask all the time (Category ③). Win small. Compound. Earn the right to bigger bets in Phase 3.

Quick Recap — Anti-Signals + Moon-Shots

Avoid safety/clinical decisions Avoid interpersonal judgement Avoid tiny-volume tasks MD Anderson — $62M moon-shot failed Start small, compound

4.3 How to Rank What You Find — The 5-Dimension Scoring Grid

You'll generate more candidates than you can pilot in Phase 2. Score them ruthlessly. Each dimension is 0–2; total /10.

Dimension	0 points	1 point	2 points
Frequency	A few times per year	Weekly	Daily or many times per day
Time Spent	Under 10 minutes	10–30 minutes per occurrence	Over 30 minutes per occurrence
Data Sensitivity	Highly sensitive (needs enterprise AI)	Some sensitive (needs desensitisation)	Mostly non-sensitive (public AI is fine)
Output Structured	Output shape always different	Output partly templated	Output fits a clear template
Good-Known	Hard to judge correctness	Spot-checkable	Easy to recognise a correct answer

🟢 8–10 → Pilot

Strong candidate. Pilot in 2–4 weeks. Most likely to deliver visible time saving fast.

🟡 5–7 → Revisit

Refine the task definition, change the scope, or wait for Phase 2 enterprise AI tools.

🔴 0–4 → Drop

AI is unlikely to help here. Recommend process redesign or keep human-only.

Workers consistently underestimate how much time their "small" tasks take. If you score Time Spent as 0 or 1, time yourself for one week before locking in the score. Many "10-minute tasks" turn out to be 30-minute tasks once timed.

Quick Recap — Scoring Grid

5 dimensions × 0–2 = /10 🟢 8–10 pilot · 🟡 5–7 revisit · 🔴 0–4 drop Time yourself before locking scores

4.4 Starter Ideas — Top 5 Opportunities per MPV Department

To prime your thinking before Lab 4, here are 25 starter ideas across five MPV departments. These are starting points, not the answer — your own context will surface better ones. Each is tagged with its HBR category.

💰 Finance & Accounting — Top 5 ▼

Monthly variance commentary draft ① — given P&L variance vs budget, AI drafts the explanation; CFO reviews. Saves ~3 hr/month.
Invoice/PO classification ① — incoming POs read, key fields extracted, routed to right approver. Saves ~30 min/day.
Supplier-contract review ① — long supplier contracts summarised; non-standard clauses flagged. Saves ~2 hr/contract.
Cashflow scenario narratives ② — given two scenarios, AI writes the executive summary. Treasury-team time reclaimed.
Tax filing cover letters & correspondence summaries ① — never the form itself; just the surrounding prose. ~2 hr/month.

📦 Procurement & Planning — Top 5 ▼

RFQ comparison ① — quotes from 3+ suppliers compared on price, lead time, terms, risks; recommend shortlist. ~90 min/RFQ.
Demand-forecast commentary ② — forecast vs actual variance, AI explains the gap before the planning meeting.
Inventory exception draft ① — weekly stockout/aging report drafted from raw data; planner reviews.
Supplier audit prep ② — 12 months of supplier's complaints, CAPAs, on-time delivery summarised for audit visit.
Logistics email triage ① — inbound emails classified (delay / claim / invoice / query) and replies drafted.

👥 Human Resources — Top 5 ▼

CV first-pass screening ① — CVs summarised against job description; recruiter decides. Saves ~5 min/CV × 100 CVs.
Onboarding FAQ chatbot ③ — answers common new-joiner questions in Vietnamese ("where do I find...", "how do I claim...").
Training session summaries ① — trainer notes converted into clean takeaway documents.
Policy translation VN ↔ EN ① — new policies drafted in both languages; HR reviews.
Engagement-survey theme clustering ② — open-ended responses grouped into top themes for management.

🏭 Factory Operations — Top 5 ▼

Shift handover summary ① — shift notes converted to structured handover doc. Saves ~20 min × 3 shifts/day.
Defect cluster description ① — defects grouped by type/line/cause; QA report drafted.
SOP draft helper ① — engineer bullet notes converted to draft SOP for review.
Safety incident first draft ① — incident narrative drafted from reporter's voice notes.
5S audit summary ② — 5S photos and observations consolidated into manager report.

🚚 Distribution & Customer — Top 5 ▼

Customer reply drafts ① — hospital and distributor replies in correct tone; sales reviews.
Order confirmation translation ① — Vietnamese ↔ English for export confirmations.
Complaint classification ① — complaints tagged (defect / delivery / packaging / billing) for routing.
Tender document scan ① — key requirements extracted from long hospital tenders. Saves ~60 min/tender.
Customer-meeting prep notes ② — customer's last 6 months of interactions summarised for sales call prep.

Pattern: Notice how most are Category ① — Process Automation. That matches HBR's 47% finding. Cognitive Insight (②) appears where MPV has accumulated data. Cognitive Engagement (③) appears in HR/customer-facing — these are the higher-risk, slower-rollout cases that need Chapter 3's governance rigour.

Quick Recap — Starter Ideas

25 ideas across 5 departments Mostly Category ① Use as primer, not the answer

4.5 What It Means to Be an MPV AI Seeder

Phase 2 isn't a course you sit through. It's a role you take on. The Phase 2 employees become MPV's first cohort of AI Seeders. The role has four components.

🌱

Plants seeds

Spreads CRAFT prompting, Do-Not-Paste discipline, and AI literacy across colleagues by demonstration — not by sending policy emails.

🛠️

Builds no-code tools

Constructs small AI-enabled workflows in Power Automate / Microsoft Copilot Studio / MISA AMIS workflow — turns one-off prompts into reusable team tools.

🧑‍🏫

Coaches colleagues

Helps three other teammates write their first CRAFT prompts. Replaces "this is hard" with "look — like this".

🛡️

Enforces the 5 Rules

Speaks up when a colleague is about to paste something they shouldn't. Logs incidents. Closes the loop with the DPO when needed.

Phase 2 Selection Criteria — Who Gets In

Completed Phase 1 capstone
Generated Chapter 4 Capstone Report with 3 HBR-tagged opportunities and 90-day plan.

Manager nomination
Line manager confirms time will be protected (~4 hr/week for 3 months).

One strong (8–10/10) opportunity
A real Category ① or ② quick-win identified in their department.

Willingness to coach
Commits to coaching at least 3 other colleagues in Phase 2 months 2–3.

Phase 2 Deep-Dive — What Actually Happens

Weeks 1–2 — Common Training (8 hours)

All Phase 2 Seeders together: advanced CRAFT, hands-on Microsoft Copilot Studio, prompt-library construction, governance refresher.

Weeks 3–4 — Departmental Workshop (4 hours/department)

Each department: Seeders apply CRAFT to their top 8/10 opportunity. Build the actual prompt library. Set up logging.

Months 2–3 — Ship & Iterate

Seeders deploy their top opportunity, measure time saved, coach 3 colleagues, score outputs weekly. Quanskill coach available 1 hr/week.

End of Phase 2 — Handoff to Phase 3

Each Seeder presents: time saved, lessons learned, recommended scaling/next opportunity. Phase 3 picks 1–2 deeper builds.

Quick Recap — AI Seeder

Plants seeds Builds no-code tools Coaches 3 colleagues Enforces 5 Rules

4.6 The People Side of AI Adoption

Even the perfect AI tool fails if the humans around it don't change behaviour. Six principles MPV's first AI rollouts have learned the hard way.

1. Start with willing users

Sceptics convert by watching enthusiasts succeed, not by being lectured. The first 3 pilots should be the 3 employees who already want this.

2. Measure something visible

Hours saved per week. Drafts shipped per day. Pick a number you'll report. "It feels faster" doesn't survive a budget review.

3. Share small wins loudly

QA Manager went from 90 min → 20 min on Monday's defect report. Tell the whole company. Make the next 5 pilots inevitable.

4. Address fears directly

"Will AI replace my job?" Answer honestly: AI replaces tasks; people who use AI replace people who don't. MPV is in growth mode — upskill, don't shed.

5. Invest in training record

Track who has completed Phase 1, who's in Phase 2, who's certified to coach. This becomes ISO 13485 evidence too.

6. Give people time

4 hours/week protected time during Phase 2. Without protected time, AI projects die under daily-job pressure. This is a manager promise, not a hope.

Quick Recap — People Side

Start with willing users Measure visibly Share wins loudly Protect 4 hr/week

4.7 The 90-Day Development Plan — Your Path to Phase 2

Whatever your role, the 90-day arc from "finished Phase 1" to "ready for Phase 2 selection" looks the same.

Days 1–7 — Pick One Tool, Learn CRAFT Deeply

Subscribe to ChatGPT Plus or Claude Pro (or get added to the MPV enterprise account). Run your 5 CRAFT prompts from Chapter 2 Lab. Iterate each prompt until it returns a 90%-good output.

Days 8–30 — Build Your Prompt Library

Add 10 more CRAFT prompts for your actual recurring tasks. Save them in a shared note. Share at least 2 with a colleague and watch them use it.

Days 31–60 — Ship One Opportunity

Take your highest-scoring (8–10/10) opportunity from Chapter 2/4 ideation. Run it for 4 weeks. Measure hours saved per week. Score outputs using the Chapter 1 rubric. Log everything.

Days 61–90 — Teach Someone + Apply for Phase 2

Coach one colleague through CRAFT basics. Generate your Phase 1 Capstone Report (Chapter 4 Lab) with measured results, three HBR-tagged opportunities, and a Phase 2 application narrative. Submit.

The litmus test for Phase 2 readiness: You should be able to point to at least 4 hours/week of time saved, one colleague you coached, and three concrete HBR-tagged opportunities with scores and verdicts. If you have those, you're in. If not, do another 30 days.

📝 Chapter 4 Summary — What You Now Own

✅ 6 Opportunity-Spotting Signals: Repetition · Time Sink · Reading · Translation · Pattern Detection · Communication at Scale

✅ 5 Anti-Signals: clinical/safety · interpersonal · phone-call-optimal · tiny-volume · data-sensitivity-blocked

✅ HBR Moon-Shot Lesson: MD Anderson lost $62M on the big bet; same hospital won with small wins

✅ 5-Dimension Scoring Grid: Frequency · Time · Sensitivity · Structure · Good-known · → /10

✅ Bands: 🟢 8–10 Pilot · 🟡 5–7 Revisit · 🔴 0–4 Drop

✅ 25 starter ideas across Finance, Procurement, HR, Factory, Distribution — each tagged ① ② ③

✅ AI Seeder role: plants seeds · builds no-code · coaches 3 · enforces rules

✅ Phase 2 structure: 8h common + 4h dept + 2 months ship/iterate + Phase 3 handoff

✅ 90-day plan: pick tool → build library → ship one → teach & apply for Phase 2

🏁 Continue to the Capstone Lab →