Skip to content

Instantly share code, notes, and snippets.

@VinACE
Created January 28, 2026 11:39
Show Gist options
  • Select an option

  • Save VinACE/d7234b0463d53e64400499ca7406c25d to your computer and use it in GitHub Desktop.

Select an option

Save VinACE/d7234b0463d53e64400499ca7406c25d to your computer and use it in GitHub Desktop.
AI Delivery Head Cadence examples
Your Core Role (Big Picture)
As AI Practice Head, you are NOT the delivery manager and NOT the architect.
👉 Your real role is to act as the bridge between:
Business outcomes
Technical execution
Delivery governance
Think of yourself as the “Outcome Owner”.
If the project fails, it won’t be because the model didn’t work —
it will be because expectations, priorities, or communication broke.
2️⃣ Where You Should Concentrate Most (80/20 Rule)
🔹 A. Business Alignment & Scope Control (VERY IMPORTANT)
This project has high AI risk if scope is not tightly controlled.
You must ensure absolute clarity on:
For UC1 (Safety / PPE):
What counts as helmet compliance?
Partial face visible → violation or not?
False positives tolerance? (e.g., 90% accuracy acceptable?)
Alert fatigue rules (how many alerts per hour?)
For UC2 (Email Automation):
Which email intents are IN scope?
What happens if extraction confidence < threshold?
SLA ownership — AI or humans?
CRM API limitations (create vs update only?)
📢 What YOU communicate:
“This is a pilot, not an enterprise-wide rollout. Success = defined accuracy + workflow adoption.”
🔹 B. Outcome-Driven Delivery (Not Task-Driven)
Your PM will track:
Timelines
Tasks
Milestones
👉 You track outcomes.
Create success metrics like:
UC1:
PPE detection accuracy ≥ X%
Reduction in manual CCTV review
Average alert response time
UC2:
% emails auto-ticketed
Reduction in manual effort
First response time improvement
📢 What YOU communicate:
“We are not delivering models, we are delivering measurable operational impact.”
🔹 C. Risk & Dependency Management (Silent Killer Area)
This project depends heavily on customer-side readiness.
Your focus:
Camera angles & lighting (UC1)
SOP clarity & SME availability (UC1 + UC2)
CRM API stability (UC2)
Sample data quality (emails, videos)
🚩 Typical risks you must surface early:
“Cameras exist” ≠ “Cameras usable for AI”
SOPs are tribal knowledge, not documented
CRM APIs undocumented or unstable
📢 What YOU communicate:
“AI accuracy depends on real-world data quality — delays here directly delay outcomes.”
3️⃣ Your Role vs Technical Team vs PM (Clear Separation)
🧠 You (Practice Head – AI)
Own vision, scope, success definition
Validate architecture direction, not code
Decide trade-offs (accuracy vs speed vs cost)
Escalate business risks
Handle executive communication
🛠️ Technical Team
Model training & fine-tuning
Integration (RTSP, CRM APIs)
Dashboards & alerts
Performance optimization
📅 Project Manager
Sprint planning
Tracking milestones
Dependency follow-ups
Status reporting
👉 If you start doing PM work → you’re underutilized
👉 If you start coding → you’re misused
4️⃣ What You Should Communicate at Each Phase
🔹 Phase 1: Kickoff
Your message:
Clear scope boundaries
Pilot success criteria
Roles & escalation paths
“No scope creep without business approval”
🔹 Phase 2: Development
Your message:
“Accuracy will improve iteratively”
“Early demos > perfect final demo”
“Feedback loop is mandatory”
You should attend demos, not standups.
🔹 Phase 3: UAT & Go-Live
Your message:
“This is a human + AI system”
Define fallback rules
Confirm ownership post go-live
Align on hypercare expectations
5️⃣ One Line That Defines Your Job
“My job is to ensure AI delivers business value predictably, safely, and at scale — not just technically, but operationally.”
If you want, next I can:
Create a 1-page AI Practice Head responsibility checklist
Draft a steering committee update template
Help you define AI success KPIs for CXOs
Or map this into an OKR structure for your AI practice
Just tell me where you want to go next 🚀
6️⃣ Your Weekly Operating Rhythm (What YOU should actually do)
Instead of ad-hoc involvement, run the project on a fixed cadence.
🗓️ Weekly – AI Practice Head Cadence
1. 30-min Outcome Review (with PM + Tech Lead)
Focus ONLY on:
Are we still aligned to business success metrics?
What assumption broke this week?
Any accuracy / data / dependency risks?
❌ Not:
Task-level updates
Jira stories
2. Demo Review (once every 2 weeks)
Insist on:
Live video feed demos (UC1)
Real emails → tickets (UC2)
📢 Your standard line:
“Show me real data, not test data.”
3. Stakeholder Touchpoint (Fortnightly)
You speak to:
Ops Head (Safety)
Customer Support Head
IT / Security
Purpose:
Validate adoption
Remove friction
Reset expectations
7️⃣ What You Must STOP Doing (Very Important)
Many Practice Heads fail because they do this 👇
❌ Micromanaging Accuracy
92% vs 94% doesn’t matter early
Consistency & explainability matter more
❌ Letting Scope Expand “Casually”
“Can you also detect gloves?”
“Can it auto-close tickets?”
🚨 This is how AI pilots die.
Your response:
“Noted for Phase-2. Let’s stabilize Phase-1 first.”
❌ Overpromising AI Intelligence
Never say:
“AI will prevent all incidents”
“Emails will be fully automated”
Say instead:
“AI will significantly reduce manual effort and improve response speed.”
8️⃣ Steering Committee: What YOU Should Present
You should present only 5 slides, not 50.
📊 Slide 1: Business Objective
Why this pilot exists
What problem it solves
📊 Slide 2: Current Status (RAG)
Green / Amber / Red
One-line reason
📊 Slide 3: Measured Outcomes
Accuracy
Automation %
Time saved
📊 Slide 4: Risks & Decisions Needed
Data gaps
SOP clarity
Go/No-Go inputs
📊 Slide 5: Scale Recommendation
Rollout?
Pause & refine?
Extend scope?
📢 Your closing line:
“Based on pilot evidence, my recommendation is…”
9️⃣ How You Evaluate Success (Real AI Leadership Test)
The project is successful IF:
✅ Business users trust alerts
✅ People actually use dashboards
✅ Manual effort reduces measurably
✅ CXO understands value without technical explanation
The project is NOT successful just because:
Model accuracy is high
Project finished on time
🔟 Your AI Practice Head “North Star”
Always ask yourself:
“If this scales to 10 plants and 100x volume, will it still work operationally?”
If yes → green light
If no → pause & redesign
🔥 Final Takeaway
Technical teams build AI.
Project managers deliver AI.
You make AI usable, scalable, and defensible.
That’s leadership.
If you want next, I can:
Draft a Practice Head playbook (PDF-style)
Create a risk register specific to AI pilots
Help you define Phase-2 roadmap
Prepare CXO-level talking points
Tell me which one you want next 👌
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment