A whitepaper series · Three papers · Four authors

The Intelligence‑Centred Enterprise

An operating model for the intelligence age. A diagnostic framework for CIOs, CFOs, and Chief AI Officers.

This is not a technology problem. It is an operating model problem that AI has made impossible to ignore.

Download the papers or start with the diagnosis

Two papers · Free · No form required

70%
of transformation programmes fail to meet their objectives.
McKinsey, Transformation Practice
39%
of technology capability spend is wasted on capabilities never absorbed.
ADAPT, Australian CFO data
95%
of enterprise generative AI pilots deliver no measurable P&L impact.
MIT NANDA · State of AI in Business, 2025

Each component is healthy. The system is broken.

From Paper 02 — From Operating Model to Operating Reality
01

Does this describe your organisation?

Five predictable ways the metabolic loop breaks. Each maps to a specific failure mode. Each produces recognisable symptoms. Each reinforces the others.

When Sense breaks

Loss of Situational Awareness

Leaders making decisions based on quarterly reports while the world moves weekly. Customer complaints surface months after the root cause began. Competitive moves discovered through news articles, not internal intelligence. A frontline team spots a pattern that could save millions, but the signal dies three management layers before it reaches anyone with authority to act on it.

Cost: Strategic decisions made on information that was already obsolete when it reached the boardroom.

When Reason breaks

Decision Quality Collapse

When was the last time your executive committee made a decision that surprised anyone in the room? Decision quality collapse does not feel like a crisis. It feels like stability. The committee believes it is making sound decisions because it follows rigorous process — but rigorous process applied to incomplete information, filtered through politics, produces decisions that are internally consistent but externally wrong.

Cost: Slow, low-quality decisions compound. Within two years, the decision architecture reflects the past, not the present.

When Act breaks

Execution Bottleneck

Initiatives succeed in controlled environments. Leadership celebrates. The initiative then enters the broader organisation, where it collides with legacy processes, competing priorities, and the frozen middle: the layer where innovation goes to be absorbed, diluted, and quietly deprioritised.

Illustrative case · Financial services A firm invested $22 million in AI over eighteen months. Fourteen initiatives launched. Three reached production. The technology in all fourteen was sound. The operating model could only absorb three.

Cost: Stranded investment. Talent attrition: your best people leave because they cannot get things done.

When Learn breaks

Organisational Amnesia

The same integration mistake happens in three consecutive AI deployments. A risk identified and documented in one programme reappears in the next because the knowledge was never transferred. Governance checks compliance but never asks the question that matters: what did we learn? Knowledge stays trapped in teams and individuals. When those individuals leave, the knowledge leaves with them.

Cost: Every repeated mistake has a direct cost. The compounding cost is worse — competitors that have built the learning infrastructure accelerate while you repeat.

When no one owns the loop

Fragmented Intelligence

The data team delivers excellent insights. The strategy team makes reasonable decisions. The project teams execute competently. The governance function monitors risk diligently. And yet the enterprise outcome is dysfunction. The Chief AI Officer was hired to solve this — but she has no operating model to work from, no governance framework designed for it, and no delivery discipline that connects the layers.

Illustrative case · Asia-Pacific transport authority An AI programme combined real-time traffic prediction with an agentic response layer. Prediction was delivered and accuracy exceeded the business case. The response automation never reached production: legal counsel could not resolve who carried liability when an automated response plan contributed to an incident. Data sat with engineering, decision rights sat with operations, and liability sat unowned. The programme was scaled back to prediction-only.

Cost: The difference between an organisation that learns and one that merely accumulates experience.

Cases are illustrative composites drawn from engagement patterns. Figures and sectors are anonymised.

02

The metabolic loop

At the heart of ICE is a simple loop borrowed from biology: Sense, Reason, Act, Learn. Most organisations over-rotate on Act. They are very busy, but not very intelligent. ICE treats the entire loop as the enterprise's metabolism — the way it converts information into improved performance.

The ICE metabolic loop A circular diagram showing four connected stages — Sense at top, Reason at right, Act at bottom, Learn at left — with arrows flowing clockwise between them. The loop represents how an enterprise converts information into improved performance. Sense Reason Act Learn The Metabolic Loop
  • Sense

    Detect weak signals from customers, markets, systems, operations. Aim: early awareness, not perfect certainty.

  • Reason

    Interpret signals using data, context, ethics, diverse perspectives. Algorithms inform; humans make meaning.

  • Act

    Empower teams closest to the signal to act with speed and safety, within clear guardrails.

  • Learn

    Capture feedback, embed new patterns into processes and policies. Without learning, sensing becomes noise.

03

Where to focus first

The metabolic loop breaks because four layers that should work together do not. Each layer, when missing or misaligned, creates a specific constraint. The fourth has no established discipline behind it — and that is the gap this series diagnoses.

Layer 01

Leadership and Governance

What boards and executive committees must change. Governance designed for capital expenditure approval cannot evaluate systems whose behaviour changes after deployment. The shift: from "did we follow the plan?" to "what did we learn?"

Layer 02

Enterprise Architecture

How organisational structures must evolve. Cross-functional pods organised around value streams, not functional silos. A transition architecture that bridges legacy and target state simultaneously.

Layer 03

Infrastructure and Platforms

What the technology stack must deliver. Real-time data pipelines that feed the Sense function. Observability platforms that monitor system health continuously. Compute architecture that scales without bankrupting the organisation.

Layer 04 · The missing one

Delivery Discipline

The discipline for taking AI systems from pilot to reliable production at enterprise scale. Does not exist as a defined body of knowledge, professional practice with shared standards, or something an organisation can hire for, train for, or certify against.

The dimension organisations have invested in least

The dimension you have invested in least is the one that determines whether the other three produce results.

Paper 02, Section 6 · The pattern you will see
04

What it looks like at the team level

The unit of delivery in an Intelligence-Centred Enterprise is the Pod — a small, durable team fusing human intent with machine horsepower, structurally redesigned around the metabolic loop. Three roles, two human, one machine.

Human · The Strategist

Product Lead

Sets the destination and the guardrails.

Defines commander's intent, not task assignments. Focus is on intent, ethics, and direction. Owns the question of whether the outcome aligns with strategy and values.

"Is the outcome aligned with our strategy and values?"

Machine · The Engine

Agent Swarm

Autonomous agents running the loop 24/7.

The Scout monitors signals (Sense). The Analyst models probabilities (Reason). The Auditor checks compliance before outputs reach humans. Runs continuously while humans focus on high-leverage decisions.

Continuous. Auditable. Always learning.

Human · The Tuner

Systems Architect

Designs the factory, not the product.

Does not do the work; ensures the system that does the work keeps working. Owns the question of whether agents are learning from the right data and operating within correct parameters.

"Are our agents learning fast enough, and is the data quality sufficient?"

05

The journey ahead

No large organisation is truly AI-native today. The advantage goes to those who can move through the stages deliberately. Three stages, each defined by what changes in the operating model and in governance.

Stage 01

AI-Enabled

Today's Default

AI attached to existing processes: copilots, chatbots, recommendation engines. Productivity gains appear, but the operating model is mostly unchanged.

  • Operating Model AI bolted onto old processes. Existing workflows with copilots attached.
  • Governance Compliance-first oversight. AI governance is reactive and policy-driven.
Stage 02

Transitional ICE

The Hybrid

Cross-functional pods organised around value streams. Work redesigned around Sense, Reason, Act, Learn. Governance and metrics start to reflect learning velocity.

  • Operating Model Pods around value streams. Teams running the loop around problems like claims resolution or customer onboarding.
  • Governance Learning-based oversight with guardrails. Boards shift from "did we follow the plan?" to "did we learn fast enough, and safely?"
Stage 03

AI-Native

Direction, not destination

Strategy, operating model and culture anchored in intelligence as the primary organising principle. The enterprise keeps maturing as the technology evolves.

  • Operating Model AI-native value creation as default. Every process designed from first principles with intelligence as the primary constraint.
  • Governance Continuous governance as a learning system. Intelligence informs risk, compliance, and board decisions in real time.
06

Where do you stand?

Whichever role you sit in, this matters. Same diagnosis, three different angles. Take the four-question readiness check below — the full 20-question version is in Paper 02.

For the CIO

Use the readiness assessment to identify your binding constraint. The constraint is rarely what you are building. It is how you are governing, coordinating, and sustaining what you build.

For the CFO

The 39% wasted spend is not a technology problem. It is an operating model problem. Returns will not improve until governance changes — not until you spend more on tools.

For the Chief AI Officer

This gives you language for the gap you have been experiencing since you took the role. The discipline your role requires does not yet exist as codified knowledge.

L1 — Leadership and Governance

Our board reviews AI portfolio health, not just AI project status.

Not startedEmbedded
L2 — Enterprise Architecture

We have cross-functional pods organised around value streams, not functions.

Not startedEmbedded
L3 — Infrastructure and Platforms

Our data infrastructure supports real-time signal detection, not just batch reporting.

Not startedEmbedded
L4 — Delivery Discipline

Someone is explicitly accountable for AI systems in production, not just for AI projects.

Not startedEmbedded
—

    Get the full 20-question version in Paper 02
    07

    How a CFO measures it

    Traditional metrics measure production. ICE measures adaptation. Four metrics that translate intelligence into financial performance — one for each stage of the metabolic loop, plus one for the system as a whole.

    Sense → Reason

    Signal-to-Decision Latency

    Lsd = Tact − Tdetect

    Time from signal detection to decision execution. A proxy for risk exposure and missed opportunities. The faster you close this gap, the less value leaks out before it can be acted on.

    Cycle Time → Signal-to-Decision Latency
    Act → Learn

    Experimentation Yield

    Ey = Codified Changes / Total Experiments

    Percentage of experiments that result in codified operating model changes. A measure of return on exploration. Zero errors usually means zero innovation; this metric tells you whether the learning is actually compounding.

    Error Rate (Sigma) → Experimentation Yield
    Human + Machine System

    Cognitive Leverage Ratio

    Clev = Algorithmic Decisions / Human Interventions

    Ratio of algorithmic decisions to human interventions. Measures the scalability of your operating model. The right question is no longer cost per hour. It is decisions of consequence per person, per quarter.

    FTE Cost → Cognitive Leverage
    Learn → Sense

    Knowledge Refresh Rate

    Kdep = Knowledge Decay / Total Knowledge Base

    Refresh rate of business rules, prompts, and models. Software is a sunk cost; the intelligence inside it is a living asset that depreciates if it is not actively maintained. Treat models like inventory, not like fixed assets.

    Software CAPEX → Knowledge Refresh Rate
    08

    The evidence

    The argument is not a hunch. Filter the peer-reviewed and field-experimental research behind each claim in the series, then open any study for the method, the numbers, and the source. The pattern is consistent: value comes from augmentation and learning, not from bolting tools onto old processes.

    +14% avg
    NBER · Peer-reviewed

    Generative AI lifted support-agent productivity 14% on average.

    Backs augmentation over automation: AI made people more productive rather than replacing them, and the gains were uneven across skill.

    DesignStaggered field deployment across 5,179 customer-support agents; productivity measured as issues resolved per hour.
    Key finding+14% on average, +34% for novice and low-skilled workers, near-zero for the most experienced. AI appeared to disseminate the best workers' tacit knowledge.

    Brynjolfsson, E., Li, D. & Raymond, L. (2023). Generative AI at Work. NBER Working Paper 31161. nber.org/papers/w31161

    +34% novices
    NBER · Peer-reviewed

    Least-experienced workers gained 34%; experts barely moved.

    The signature ICE insight: AI lifts the floor, not the ceiling. It compresses the experience curve, which is a workforce and capability question, not a tooling one.

    Why it mattersTreated agents with two months' tenure performed as well as untreated agents with far more experience — months of learning compressed, not years.
    ImplicationValue concentrates where capability is being built. This is the case for capability graphs and continuous reskilling over headcount reduction.

    Brynjolfsson, E., Li, D. & Raymond, L. (2023). Generative AI at Work. NBER Working Paper 31161.

    +40% quality
    Harvard / BCG · Field experiment

    Consultants using AI worked 25% faster and produced 40%+ higher quality.

    Backs augmentation on complex knowledge work — but only inside AI's capability frontier. Outside it, AI made experts measurably worse.

    DesignPre-registered randomised experiment with 758 BCG consultants across 18 realistic tasks; conditions: no AI, GPT-4, GPT-4 with prompting guidance.
    The jagged frontierInside the frontier: +12.2% tasks completed, 25.1% faster, 40%+ higher quality. Outside it: AI users were ~19% less likely to be correct — the case for human-in-the-lead judgement.

    Dell'Acqua, F. et al. (2023). Navigating the Jagged Technological Frontier. Harvard Business School Working Paper 24-013.

    55.8% faster
    GitHub / MSR · Controlled trial

    Developers with an AI assistant finished the task 55.8% faster.

    Backs decision and delivery velocity: compressing cycle time is the visible edge of learning velocity — the only metric ICE treats as defensible.

    DesignControlled experiment; developers implemented an HTTP server in JavaScript. Treated group: 71 minutes vs 161 for control (p = 0.0017).
    HeterogeneityLargest gains among less-experienced developers — echoing the floor-lifting pattern seen elsewhere.

    Peng, S., Kalliamvakou, E., Cihon, P. & Demirer, M. (2023). The Impact of AI on Developer Productivity: Evidence from GitHub Copilot. arXiv:2302.06590.

    95% stall
    MIT NANDA · Industry study

    95% of enterprise generative AI pilots deliver no measurable P&L impact.

    Backs the central thesis: embedding beats piloting. The barrier is the learning gap and integration into workflows, not model quality.

    Design150 leader interviews, a 350-employee survey, and analysis of 300 public AI deployments.
    Key findingDespite $30–40bn in enterprise spend, only ~5% of pilots reach material value. The divide is high adoption, low transformation — an operating model failure, not a tooling gap.

    MIT NANDA initiative (2025). The GenAI Divide: State of AI in Business 2025.

    39% wasted
    ADAPT · CFO data

    CFOs estimate 39% of technology capability spend is effectively wasted.

    Backs embedding beats piloting: capability is bought but never absorbed into how work happens. More tooling will not move the number.

    ContextAustralian CIO/CFO data: only ~47% of key processes run in real time, ~50% of staff have trusted real-time data, and ~40% of mission-critical apps still depend on legacy platforms.
    ImplicationThese constraints make learning velocity structurally impossible, regardless of how many AI tools are deployed.

    ADAPT, Australian CIO & CFO research (cited in Paper 01).

    EPOCH shift
    MIT Sloan · Working paper

    New work is shifting toward human-intensive capabilities AI complements.

    Backs the human edge: Empathy, Presence, Opinion, Creativity, Hope. The roles that grow are those where humans set meaning and direction.

    FrameworkEPOCH captures the human capabilities that complement rather than compete with AI.
    Key findingTasks emerging in 2024 score higher on EPOCH than pre-existing ones; high-EPOCH occupations saw stronger employment growth (2015–2023) with favourable projections to 2034.

    Loaiza, I. & Rigobón, R. (2025). The EPOCH of AI: Human–Machine Complementarities at Work. MIT Sloan Working Paper (SSRN 5028371).

    4+ functions
    McKinsey · Global survey

    Top AI performers embed AI across four or more business functions.

    Backs embedding beats piloting: economic outperformance correlates with pervasive intelligence, not isolated pilots in one or two functions.

    Definition"AI high performers" attribute at least 20% of EBIT to their AI use.
    PatternHigh performers are far more likely to have embedded AI across four or more functions; lower performers restrict it to one or two isolated pilots.

    McKinsey & Company (2023–2024). The State of AI.

    No studies match that filter.

    The counterintuitive finding

    AI lifts the floor, not the ceiling.

    Across the studies, the same pattern repeats: the largest gains go to the least-experienced. Experts barely move. The strategic question is therefore about capability and learning velocity — who can climb the curve fastest — not about replacing people.

    If AI compresses years of experience into months, your advantage is no longer who you hire — it is how fast your organisation learns.

    Source: Brynjolfsson, Li & Raymond (2023), Generative AI at Work — issues resolved per hour, by skill quintile.

    09

    Read the full series

    A trilogy building toward a single argument: enterprise AI fails at the operating model layer, not the technology layer. Three papers, one framework, four authors.

    Paper 01 — Vision

    Leading the AI‑Centred Enterprise of the Future

    How intelligence‑centred enterprises thrive in the living economy.

    Introduces the Intelligence-Centred Enterprise. Three pillars, five building blocks, the metabolic loop, the CFO's learning velocity dashboard, and the leadership manifesto for the intelligence age.

    • The metabolic loop
    • 5 building blocks
    • CFO dashboard
    • Leadership manifesto
    Download PDF 24 pp · 6.6 MB
    Paper 02 — Diagnostic

    From Operating Model to Operating Reality

    Why the intelligence‑centred enterprise stalls, and where to focus first.

    The five failure modes. The four layers. The category error. The full 20-question readiness assessment, plus case studies from financial services, transport, and infrastructure.

    • 5 failure modes
    • 4 layers
    • 20-question check
    • 3 case studies
    Download PDF 39 pp · 6.1 MB
    Paper 03 — Implementation

    Building the Delivery Discipline

    How to build and run the missing layer.

    Specific structures, roles, governance patterns, and delivery methodology for the discipline that does not yet exist anywhere in the industry.

    • Pod operating patterns
    • Governance templates
    • Delivery methodology
    Coming late 2026 Notify me

    Get Paper 03 on release

    Papers 01 and 02 are above — download them now, no form required. Leave an email and we'll send Paper 03 the day it ships, plus the occasional framework update from the authors. No marketing, no list-sharing.

    No spam · Unsubscribe anytime
    From diagnosis to operating reality

    The framework names the gap. Closing it is the work.

    Most teams know which layer is weakest within minutes of taking the check. The harder question is what to change first, and in what order. That is the conversation the authors have with leadership teams.

    Take the readiness check Talk to the authors
    10

    The authors

    Four practitioners building this body of work in parallel with their day jobs. Each brings a distinct vantage point: enterprise advisory, delivery systems, infrastructure, and architecture.

    Mark Cameron, CEO Alyve

    Mark Cameron

    CEO, Alyve

    27 years advising government and enterprise on practical AI adoption. Forbes contributor on AI strategy and organisational change. Teaches Digital and AI Transformation Leadership at Deakin University Executive MBA.

    LinkedIn
    Vijayan Seenisamy, Creator of AI ROF

    Vijayan Seenisamy

    Creator, AI ROF™

    Creator of the AI Role Operating Framework. Author of The AI Delivery Manager Blueprint and The Pilot Trap. 20 years across Accenture, Telstra, Deloitte, Bupa, and Woolworths Group.

    LinkedIn
    Vinod Bijlani, AI Practice Leader HPE

    Vinod Bijlani

    AI Practice Leader, HPE

    22 years in data science and solution architecture. Holds 25 patents in AI/ML and IoT. Contributor to the World Economic Forum and Forbes Technology Council. Architect of smart city and intelligent transportation systems globally.

    LinkedIn
    Sharma Madiraju, Enterprise AI Strategist

    Sharma Madiraju

    Enterprise AI Strategist

    Among the region's leading CIOs. Deep expertise in enterprise technology strategy and architecture across financial services, defence, retail, logistics, and professional services. Engineers high-velocity decision systems from legacy estates.

    LinkedIn
    Download the papers