Flagship course
ML Spend Attribution Lab
A hands-on path for finance and platform pairs who need a defensible map from model activity to ledger lines.
What you will leave with
A living tagging contract
Agreed fields for project, model family, environment, and owner — written so engineers can automate them.
Shared-pool allocation logic
Documented rules for idle capacity, reserved instances, and multi-tenant clusters.
An auditor packet
Export templates that explain exclusions, estimates, and confidence levels.
Informational pricing
HK$ 8,900 / seat
Cohort seats include live labs, office hours, and workbook access for six months. Pricing on this page is informational only — no checkout is available on the site.
Duration: 5 weeks · ~6 hours / week
Format: Live remote with optional Wan Chai study meetups
Modules
Inventory the opaque bill
Parse cloud exports, identify ML-shaped spend, and mark what cannot yet be attributed.
Design the tagging spine
Choose fields that survive org reshuffles and map cleanly into a financial auditing app for machine learning cost attribution.
Allocate shared GPU pools
Practice mediation scripts when research and production share the same hardware.
Inference and experiment nuance
Separate training spikes from steady serving, including canary and A/B traffic.
Publish the audit pack
Assemble narratives, appendices, and open questions for finance leadership.
Instructor
Clara Ng
Former FinOps lead for a regional AI platform team in Hong Kong. Clara now coaches finance-engineering pairs on attribution systems that hold up under audit scrutiny.
A real limitation
This lab does not connect to your production billing APIs or generate live invoices. You practice on anonymized datasets and take the method home — implementation tooling remains your team's choice.
Learner notes
Module 03 was the turning point. We finally wrote down how idle A100 hours get split between research and customer inference — imperfect, but signed off by both leads.
Dense for finance-only attendees in week two. Pairing with an engineer made the tagging exercises workable; solo, I would have needed more prep time.
FAQ
Do I need cloud engineering experience?
Comfort reading cloud cost exports helps. Deep Kubernetes skill is not required, but someone on your team should be able to apply the tagging contract after class.
Is software included?
No product license is sold here. We teach methods you can apply inside the tooling you already use or plan to buy.
What is the main limitation of the program?
We cannot certify that your internal controls meet a specific regulator's expectation. We teach documentation habits; your legal and audit partners remain the authority.
Can teams enroll together?
Yes — finance and platform pairs get the most from paired exercises. Ask us about duo seats when you contact the team.