Public course

Cost, efficiency and AI

An operating-loop workshop for executives

This course does not teach you to use AI. It teaches you to sign off on it.

  • Where the value is
  • How strong the evidence is
  • Who owns it
  • How to scale without losing control
74
slides in the main deck
9
acts in the arc
67
verified research sources
Discuss a session View the full deck

WHY NOW

Adopting is not the same as earning

The question was never whether to invest. It is why the investment has not paid. This course rebuilds the part that goes missing in between.

  • 88% → 13%

    High adoption, low realisation

    88% of organisations worldwide now use AI in at least one business function, yet only 13% report significant enterprise-level value. The gap is not technical. It is an operating-loop gap.

    Stanford HAI, AI Index 2026; Accenture. Three surveys with different samples and definitions — not a single funnel.

  • 5×

    AI widens gaps, it does not close them

    Companies most exposed to AI grew revenue per employee by 33.5%; the strongest fifth of that group reached 163%. Same technology, a fivefold spread.

    PwC, Global AI Jobs Barometer 2026. Redrawn from the published data.

  • +55% / −19%

    One tool, two directions

    In a controlled experiment, developers with AI assistance finished a set task about 55% faster. In another study, experienced developers were 19% slower — and believed they had been faster. The difference lies in task boundaries, quality control and how work is divided.

    GitHub controlled experiment; METR / Becker et al., 2025 (a result the team could not later reproduce).

SYLLABUS

Nine acts

Each act carries one hook, one body of evidence, one method and one output. The three peaks fall in diagnosis, accounting and governance.

  1. AlarmThe window is closing

    Policy has set the timetable and competitors have doubled their budgets. 94% of organisations keep investing even without a short-term return.

    slides P4–P9

  2. RealityUsing is not earning

    The three causes of the value gap, plus a six-dimension maturity self-assessment: which tier are you in.

    slides P10–P17

  3. DiagnosisWhere money leaks and process jams

    If you cannot state a unit cost, you cannot claim a saving. One line of spend carries four different prices; a standard process usually has dozens of real variants.

    slides P18–P28

  4. SelectionStop spreading thin

    Twenty simultaneous pilots is the same as none. A value-by-feasibility matrix, and a written list of what you will not do.

    slides P29–P34

  5. CapabilityWhat AI can actually do

    Live demonstrations of retrieval, assistive and agentic modes, with risk rising at each step; and the four technical forks an executive has to settle.

    slides P35–P43

  6. EvidenceWho has actually done it

    International benchmarks alongside Chinese practice, every figure graded for evidence strength, plus a checklist you can take away.

    slides P44–P50

  7. AccountingSaved hours are not profit

    Time only becomes money through one of five exit conditions. Three ROI figures — executives should only act on the last two.

    slides P51–P58

  8. GovernanceThe line you do not cross

    Measured hallucination ranges across leading models, the collapse that follows a contaminated premise, twelve risk categories and a procurement questionnaire.

    slides P59–P67

  9. ExecutionNinety days, five gates

    Initiation, prototype, pilot, financial confirmation, scale-up. The fourth gate is signed by finance; nothing replicates across departments before the fifth.

    slides P68–P72

OUTCOMES

What participants leave with

Not a deck — three completed worksheets and four cards that fit in a pocket. We review the results together ninety days later.

Three worksheets, completed in the room

  1. Scenario portfolio

    Three to five priority scenarios, each signed by a business owner, alongside a written list of what was cut and why.

  2. Value calculation

    Baseline, benefit route, five layers of indicators and three ROI figures per scenario, with space for a finance signature.

  3. Ninety-day plan

    Each of the five gates carries a date and an owner; the fourth names the finance signatory explicitly.

Four pocket cards

  • Eight questions for any claim

    Ask every vendor in turn: what is the baseline, who measured it, was there a control group, how large was the sample.

  • Eight questions for procurement

    Which model underneath, where the training data came from, whether your data trains it, and whether you can export and switch.

  • Five exit conditions

    For saved time to reach the income statement, at least one of them has to hold.

  • Executive glossary

    Technical terms explained in operating language, and the five claims most likely to mislead a buyer.

FORMATS

Four delivery formats

One method, cut to the audience, the clock and the industry. Every version draws on the same traceable source base.

  • Full-day workshop

    6–8 hours · executive team

    All nine acts with two hands-on workshops. Ends with three completed worksheets and signed commitments.

  • Ninety-minute briefing

    90 minutes · board or executive breakfast

    The three peaks only: the value gap, the accounting discipline, and hallucination and governance. Suited to building consensus before approval.

  • Role-specific sessions

    Half day · single function

    One version each for the chief executive, finance, operations, technology and people functions, with questions and cases replaced throughout.

  • Industry editions

    By client sector

    Manufacturing, financial services, energy and state-owned enterprises, and retail and professional services — cases, data and framing all swapped.

EVIDENCE

Every figure carries its grade

This is what separates the course from most AI training: we do not only give you numbers, we tell you how much decision weight each one can carry.

  1. A

    Regulatory filings, financial statements, controlled experiments

    Strong enough for a board resolution.

    Cost savings stated in a regulatory disclosure; completion-time differences in a randomised trial.

  2. B

    Company disclosures and government case collections

    Strong enough for a business case and an implementation hypothesis.

    Self-reported efficiency gains; industry cases published by a ministry.

  3. C

    Media and vendor accounts

    Good for a hypothesis only — verify before you act.

    Effect figures in press coverage; customer stories supplied by a vendor.

Every figure is introduced with its grade. Where a definition is unusual, or where the original could not be verified locally, that is stated in the room. The method itself is part of the course.

SOURCES

Where the course comes from

Everything is distilled from 67 verified public sources, and every conclusion traces back to a specific place in the original report. This is the base the course stands on.

Research and evidence base

  • Chinese policy and industry

    4 sources

    The State Council's AI Plus initiative, the NDRC's policy commentary, and CAICT's industry research and large-model deployment roadmap.

  • Global benchmarks

    9 sources

    Stanford HAI AI Index, BCG AI Radar and its high-performer study, Deloitte, Accenture, PwC and the World Economic Forum.

  • Cost and process tooling

    5 sources

    FinOps Toolkit, OpenCost, the FOCUS cost specification and sample data, Cloud Custodian, and PM4Py for process mining.

  • Governance and risk frameworks

    3 sources

    The NIST generative AI risk profile, and CAICT's AI safety governance and cloud computing blue books.

  • Enterprise cases

    10 sources

    International benchmarks and Chinese practice in equal measure, across finance, manufacturing, energy, construction and knowledge work — each graded for evidence.

  • Deck and visualisation

    A dozen or so

    Mermaid, Marp, Apache ECharts and AntV, together with the open datasets published by the research institutions above.

The distilled library

Those 67 sources are worked into nine modules — 50 documents and roughly 150,000 characters. Narrative, data, cases, method and compliance each sit in their own place.

  • 3docs

    Narrative spine

    Core claim, the nine-act arc, opening script

    Teaching material
  • 8docs

    Data

    Value gap, global trends, Chinese policy, productivity evidence, talent, agents, value tiers, hallucination findings

    Public
  • 4docs

    Case library

    Case index, international benchmarks, Chinese practice, evidence grading and verification checklist

    Public
  • 12docs

    Methods

    Seven-step loop, cost transparency, process diagnosis, scenario ranking, indicator tree, ROI model, governance gates, deployment roadmap, technology selection, risk register, cost definitions, cloud architecture

    Public
  • 7docs

    Course design

    Full-day and ninety-minute plans, role-specific packs, classroom worksheets, maturity self-assessment

    Public
  • 4docs

    Deck production

    Slide-by-slide script, chart sources, phrasing and pacing, visual standards

    Teaching material
  • 3docs

    Sources and compliance

    Source register, figure traceability table, licensing limits

    Public
  • 4docs

    Industry editions

    Manufacturing, financial services, energy and state-owned enterprises, retail and professional services

    Public
  • 4docs

    Reference cards

    Glossary, books and tools, audio, video and further reading

    Public

Modules marked as teaching material contain instructor phrasing, slide scripts and staging decisions. They ship with a booked session and are not published here. The research base, the method and the evidence standard are fully public.

Every figure has a traceability record in the library: which report, which chapter, when it was retrieved, and whether the original text was verified locally. The few citations we could not verify at source are marked as such.

Run this method against your own numbers

Sessions are tailored by industry and by role. Tell us your size, your sector and the problem you most want solved, and we will propose a format and a length.

Discuss a session