Alyve

Research companion · October 2026

AI value comes from four organisational choices, not from more AI.

Almost every organisation now uses AI. Only 6% report significant value and at least 5% of EBIT from it (McKinsey, 2026). What the 6% do differently:

Research synthesis by Mark Cameron, CEO, Alyve

  1. 01Concentrate on a few outcomes
  2. 02Redesign the work before automating it
  3. 03Measure results in dollars
  4. 04Build a learning system

The question for your team

How long does a full turn of our learning loop take: weeks or quarters?

Read the evidenceField guide

The gap

More organisations are getting some value. The top is no more crowded than it was.

Read with careThese figures come from one survey but measure different things, and should not be read as a single conversion funnel. Together they show the gap between use, individual benefit and enterprise economics.

Illustrative field. Not drawn to any single study's sample.

The middle is moving. Significant value remains concentrated: McKinsey's high-performer share has not grown in two years.

Next: 01 Concentrate on a few outcomes

01Concentrate

Put the money behind a few outcomes. Let everyone else learn.

The organisations creating value don't choose between speed and depth. They fund a small number of outcomes with milestones and money, and give everyone else the tools and permission to experiment and pass on what works.

How to choose the few

Competitive strategy is the filter.

Choosing the few is a strategy question before it's a technology one. We work through five decisions in order. An outcome earns concentrated investment when it changes how the organisation wins, not because it's easy to automate.

Strategy: where and why

  1. 01

    AI North Star

    Where is AI taking us, and what must it never replace?

  2. 02

    The filter

    Competitive strategy

    Which decisions must we make faster and better than anyone else?

Operating model: how we make it work

  1. 03

    People

    How do we work, and who decides what, so every piece of work teaches the next one?

  2. 04

    Process

    Which work gets redesigned, handed off or stopped?

  3. 05

    Architecture and capability

    What architecture and capability does all of that require?

Our framing: the order in which we work through these decisions with leadership teams.

Start at step 5 and the value stays small. Start at step 1 and the technology decision gets much easier. When all five hold, learning velocity becomes enterprise value.

Read through this lens, C.H. Robinson's quotes falling from 20 minutes to 31 seconds (company-reported) aren't only an efficiency story. Quote speed is part of how a freight broker competes, so the saving sits on a decision that decides who wins the work.

Our framing

Pointing the same way: BCG 2026 (opens in a new tab) (B), organisations strong on both strategic clarity and applied AI generate 5x the AI value of those weak on both; McKinsey 2026 (opens in a new tab) (B), the leading-value cohort is 3.3x as likely to intend fundamental transformation.

Learn widelyGo deep on a fewEnterprise valueCheap, shallow experiments.What works is shared across.Funded outcomes draw on what the shallow wells find.
Breadth feeds depth: experiments everywhere, money behind a few, and a channel between them.

Our framing

For your next team meeting

Which one or two outcomes carry a financial milestone and a date this quarter?

Next: 02 Redesign the work

02Redesign

Redesign the work before you automate it.

Nearly three quarters of the leading-value cohort fundamentally redesign workflows because of AI, against about a quarter of others (McKinsey, 2026). Automating an unchanged process tends to move the bottleneck rather than remove it.
Specialist quote20 minutes
AI agent quote31 seconds
Time to deliver a freight quote, company-reported

How far redesign reaches: from task to business model (appendix)

For your next team meeting

Where are we redesigning work rather than inserting AI into existing processes?

Next: 03 Measure results in dollars

03Measure

Capacity is not cash. Measure the result in dollars.

AI reliably frees up time for individuals. It becomes value only when the organisation decides where that time goes and tracks what changes in dollars.

01 · Capacity created

  • Time saved
  • Faster drafting
  • Fewer emails
  • Quicker analysis

02 · Conversion layer

Where does the released capacity go?

  • Workflow redesign
  • Leadership decisions
  • Capacity redeployment
  • Ownership
  • Incentives
  • Measurement

Leakage when the layer is missing

  • Absorbed capacity
  • Unchanged workflow
  • No owner
  • No metric movement
  • Faster average outputSpeed without a quality standard moves the bottleneck to the reviewer. Leaders receive more work, and more of it regresses to the average.Our framing

03 · Value converted

  • Throughput
  • Service
  • Revenue
  • Cost
  • Risk
  • Innovation
  • Decision quality

The conversion layer and leakage labels are our framing of the evidence, not a measured causal model.

For your next team meeting

Where do AI results appear in our monthly business reviews, in dollars rather than training numbers?

One more question: What is our rule for stopping something that isn't working?

Next: 04 Build a learning system

04Learn

Culture is too vague. Build a learning system.

The leaders treat learning as infrastructure: time to experiment, shared standards for judging output, and a way to pass on what works. How fast that loop turns decides how fast capability compounds.
  • Learning loop turns every fortnight
  • Learning loop turns every quarter
1.01.52.02.53.006121824Months2.8x1.2x
Same improvement per turn, different loop length. Over two years the gap is more than double.

Illustrative model, not data. Assumes a 2% improvement per turn.

The question for your team

How long does a full turn of our learning loop take: weeks or quarters?

  1. 01

    Sense

    • Cross-functional learning
    • Shared context

  2. 02

    Reason

    • Shared evaluation standards
    • Technical and governance support
    • Shared context

  3. 03

    Act

    • Experimentation time
    • Technical and governance support
    • Recognition and incentives

  4. 04

    Learn

    • Experimentation time
    • Shared evaluation standards
    • Cross-functional learning
    • Recognition and incentives
  5. The loop then returns to Sense. Shared context: Templates, standards, values and an agreed AI constitution that every tool draws on, so experiments start from the organisation's knowledge rather than the model's defaults.

The mapping is our synthesis.

MIT Sloan · two-year field study · 2026

B

3

organisation-wide solutions after more than 80% of participants disengaged

141

with structured support for experimentation, evaluation, technical help, risk screening and recognition

Two-organisation, two-year field study (working paper): an academic medical centre and a corporate law firm. In the law firm, more than 80% of domain experts dropped out of AI innovation efforts.

Source: MIT Sloan 2026 (opens in a new tab)

Next: Proof

Proof

In the three cases that worked, the work changed before the technology did.

The surveys show the pattern. These cases show the decisions behind it. Three created measurable value; one is a cautionary tale. Figures are company-reported or come from executive interviews, and have not been independently audited.
  1. DCompany-reported case

    C.H. Robinson

    Redesign the work before you automate it, and make it safe to report red.
    Sector
    Logistics and freight brokerage, United States. A 120-year-old company whose revenue fell about 34% from 2022 in a freight downturn.
    What they concentrated on
    The routine, high-volume steps in a shipment's life: quoting, order entry, appointment setting and tracking.
    How they operate differently
    Teams mapped workflows using Lean methods and removed tasks that added no value before automating what remained, with hundreds of AI agents built mostly in-house by around 450 engineers who know the shipping business. Cross-functional groups of engineers, operations experts, finance and legal are questioned Socratically to decide what gets built. Failure mode and effects analysis runs before deployment. Status reports allow only red or green, and the company celebrates a red: whoever reports one gets the organisation's help to recover.
    What they measured and report
    A 45% productivity uplift since 2022. Quotes that took specialists 20 minutes now take 31 seconds. Double-digit earnings per share growth since 2023 despite the revenue fall. Staff moved to higher-value work such as tariff advice, and natural turnover of 11 to 14% a year wasn't backfilled in automated areas, so headcount no longer tracks volume there. The company is still hiring for supply chain consulting and small-business sales.
    The lesson
    Redesign the work before you automate it, and make it safe to report red.
    Sources

    Fortune, Eye on AI, 14 July 2026 (opens in a new tab)

    C.H. Robinson Q1 2026 earnings summary (opens in a new tab)

  2. DCompany-reported case

    JPMorgan Chase

    Let everyone learn, put the investment behind a few journeys, and don't bank the hours.
  3. DCompany-reported case

    Moderna

    The question of who does what work needs one owner across people and technology, whatever the org chart says.
  4. DCompany-reported caseCritical · cautionary case

    Klarna

    When cost is the only measure, quality leaks out unseen until customers notice. In our framing, this is the Efficiency Trap.

Which of the four choices each case illustrates

  • C.H. Robinson

    • Concentrate
    • Redesign
    • Measure
    • Learn

  • JPMorgan Chase

    • Concentrate
    • Redesign
    • Measure
    • Learn

  • Moderna

    • Concentrate
    • Redesign
    • Learn

  • Klarna

    • Redesign
    • Gap: measured cost only

Mapping is our synthesis.

Next: Australia

Australia

Australian organisations are strong on control and light on redesign

PwC's 2026 study includes Australian enterprises. They lead the global sample on AI security and trail AI leaders on most measures of redesign.

  • Redesigned workflows around AI

    Australia7%
    AI leaders56%

  • Incentives that encourage experimenting with AI

    Australia13%
    AI leaders59%

  • Using AI for business model transformation

    Australia3%
    AI leaders59%

  • Cross-sector collaboration

    Australia10%
    AI leaders50%

  • Robust protections for data, models and infrastructureAustralia ahead

    Australia73%
    AI leaders69%

6.8 months

average time from pilot launch to value

Source for all figures: PwC Australia, 20 April 2026 (opens in a new tab)

Our synthesis

Australia has built the governance foundation. The gap is in what gets built on top of it. In our framing, this is where the Control Trap starts: controls ready before the work has changed.

For your next team meeting

Which of these numbers would our organisation recognise?

Next: Your move

Your move

Ten questions that show whether you're wide or deep.

These are questions, not a diagnosis. They are designed for leadership teams at any level to answer together.
  1. 01

    Which enterprise outcomes do we want AI to materially change over the next 24 months?

  2. 02

    Which three to five end-to-end value streams deserve disproportionate attention?

  3. 03

    Where are we redesigning work rather than inserting AI into existing processes?

  4. 04

    When AI releases capacity, where exactly will that capacity be redeployed?

  5. 05

    Who owns the economic and operational outcome for each major AI investment?

  6. 06

    What capabilities must leaders and teams build to work differently with AI?

  7. 07

    What are we learning that makes the next deployment faster, safer and more valuable?

  8. 08

    What are the explicit criteria to stop, scale or redirect an AI investment?

  9. 09

    How long does a full turn of our learning loop take: weeks or quarters?

  10. 10

    If AI does the first-draft work our people learned on, how will we build the next generation's judgement?

Pattern one

Wide not deep

The answers mostly describe tools, pilots, use cases and adoption metrics. Activity is widespread but largely unconnected.

Pattern two

Building enterprise AI capability

The answers describe strategy, workflow redesign, capability, learning, financial outcomes and capital reallocation, connected into one system.

Most organisations sit somewhere between the two. The useful question is direction of travel.

A 45-minute session for a leadership team

  1. 01

    Ten minutes alone

    Each person picks the two questions above they think the team answers worst, and writes a one-line honest answer to each.

  2. 02

    Fifteen minutes together

    Compare answers. Where they differ, ask why before deciding who's right.

  3. 03

    Ten minutes

    Choose one outcome to concentrate on and one learning mechanism to strengthen. Name an owner and a date for each.

  4. 04

    Ten minutes

    Agree what you'd need to see in the next monthly business review to know it's working.

Your next step

Find your weakest step.

Pick the step people in your organisation would answer most differently if you asked them one by one. That's usually your weakest. Below is the next move, and the question to put to your team.

Next move

List the five decisions that most decide whether you win. Pick the one where faster, better judgement would matter most, and make that your first concentrated outcome.

The question

Which decision must we make faster and better than anyone else?

Back to choosing the few

Our framing. Nothing you select is stored or sent anywhere.

Appendix

Appendix: models, evidence and sources

Supporting models and the full evidence base behind the argument above.

A1Appendix · Seven choices in detail

The leaders make a coherent set of organisational choices.

Seven choices recur across the evidence. They are associations observed in surveys and field studies, not proven causes, and they appear to reinforce one another. Value comes from coherence, not from one practice in isolation. Each sits under one of the four choices in the main argument.

Select a choice to see its evidence

  • 1 of 7 · Sits under: 01 Concentrate

    McKinsey 2026: organisations reaching its leading-value threshold are 3.3x more likely than others to intend fundamental business transformation, and most pursue growth and/or innovation alongside efficiency.

    McKinsey 2026 (opens in a new tab)

    BCG Applied AI Index 2026: organisations strong in strategic clarity and applied AI generate 5x as much AI value as those weak on both.

    BCG 2026 (opens in a new tab)

    Enterprise implication

    Fund against strategic outcomes and value pools, not a catalogue of AI ideas.

A2Appendix · Judgement and the talent pipeline

Where will the next generation of judgement come from?

Experts become more valuable when AI is in the work, because only they can ask the sharp question and tell a good output from a plausible one. But expertise has usually been built through years of entry-level work, and that's the work AI is taking first.

Who climbs to the top in ten years?

Senior specialistsExperiencedprofessionalsEntry-level workAI now does first
Expertise has usually been built on the bottom rungs. Early-career employment in the most AI-exposed occupations is down 16% relative to peers (Stanford Digital Economy Lab, 2025).

Our framing

Stanford Digital Economy Lab · 2025

B

16%

relative decline in employment, ages 22 to 25, most AI-exposed occupations

Brynjolfsson, Chandar and Chen, "Canaries in the Coal Mine?", November 2025 update.

Early-career workers aged 22 to 25 in the most AI-exposed occupations saw a 16% relative decline in employment, after controlling for firm-level shocks. Experienced workers in the same occupations were stable or grew. The declines are concentrated where AI automates work rather than augments it.

An earlier August 2025 version reported 13%. US payroll data, observational.

Source: Stanford Digital Economy Lab 2025 (opens in a new tab)

The tension

Option one

Judgement first: limit AI in the early months of a role so people learn the work before they delegate it.

Option two

Accelerated apprenticeship: use AI to expose people to more cases sooner, with experts reviewing the reasoning, not only the output.

We don't think the evidence settles this yet. Most organisations will need both, by role.

A3Appendix · Value stream

The unit of transformation is getting bigger.

Each step widens the scope of what changes, and the consequence for the enterprise.Our framing

  1. 01 · One person, one activity

    Task

    Do an existing task faster.

    Example: Draft, summarise, retrieve.

  2. 02 · A team and its process

    Workflow

    Redesign how work moves end to end.

    Example: Remove queues, hand-offs and rework.

  3. 03 · The judgement layer

    Decision

    Change how evidence is assembled, options are evaluated and decisions are made.

    Example: Pricing, demand, maintenance, resource allocation.

  4. 04 · Several functions, one outcome

    Value stream

    Improve an outcome across multiple functions.

    Example: Order-to-cash, quote-to-delivery, asset uptime.

  5. 05 · The whole enterprise

    Business model

    Use AI to create new value pools, products, services or economics.

    Example: New offers, new pricing, new economics.

This is not an argument against automation. Routine automation can create substantial value. The distinction is between isolated task acceleration and redesign of strategically important work.

A4Appendix · Leadership system

AI value is a leadership system.

No single function can create enterprise AI value alone. Select a domain to see what it owns and what it depends on.

  1. 01 · CEO / Executive team / Strategy

    Strategy

    Set the enterprise outcomes, strategic hypotheses and value pools AI is meant to change.

    Depends on: Work, Technology, Economics

  2. 02 · COO / Business leaders / Transformation

    Work

    Redesign end-to-end workflows, decision rights, hand-offs and human-AI roles.

    Depends on: Strategy, Technology, People

  3. 03 · CIO / CTO / CDO / AI leaders

    Technology

    Build the data, architecture, platforms, controls and reusable capabilities needed to scale.

    Depends on: Work, Economics

  4. 04 · CHRO / Business leaders / Change

    People

    Build capability, trust, incentives, learning loops and new ways of working.

    Depends on: Work, Strategy

  5. 05 · CFO / Finance / Executive team

    Economics

    Measure value, allocate capital, decide what to stop or scale, and convert capacity into financial or operating outcomes.

    Depends on: Strategy, Technology

    48% of CFOs say finance tracks AI-driven value and reallocates capital; 8% lead enterprise-wide AI value goals with automated investment triggers. IBM 2026 (opens in a new tab)

Five-lens model: our framing.

Gartner CIO survey · 2026 · separate study

13%

of CIOs report their enterprise has realised significant or transformational value from AI. Enterprise value remains scarce, even as activity grows.

Source: Gartner 2026 (opens in a new tab)

A6Appendix · Evidence explorer

The evidence base, open to inspection.

Methodology disclosure

Association is not causation. Most enterprise ROI studies are observational surveys. Experimental and longitudinal evidence is labelled separately. Different studies define "value" differently, so figures should not be combined as if they share one methodology.

A experimental or longitudinal field evidence. B large-sample survey or multi-project analysis. C expert synthesis or heuristic. D company-reported case or executive interview.

Filter by strength or theme

29 of 29 entries

  • McKinsey

    2026

    BSurvey evidence

    AI use is close to universal among surveyed organisations.

    89% regular use in at least one function

    Method

    The state of AI in 2026. Global survey, August 2026, self-reported. Themes: Value.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2026

    BSurvey evidence

    Most respondents report improved individual productivity from AI use.

    80% report productivity gains

    Method

    The state of AI in 2026. Global executive and manager survey, self-reported. Themes: Value, Workforce.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2026

    BSurvey evidence

    A minority report any positive enterprise EBIT contribution from AI.

    37% report any EBIT contribution, essentially unchanged from 2025

    Method

    The state of AI in 2026. Global executive survey, self-reported EBIT attribution. Themes: Value, Measurement.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2026

    BSurvey evidence

    A leading-value cohort attributes at least 5% of EBIT to AI and reports significant enterprise value.

    6% reached this value threshold, flat since 2025

    Method

    The state of AI in 2026. Cohort McKinsey defines from survey responses (it calls this group high performers); n = 92 of 1,521 AI-using respondents. Themes: Value.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2026

    BSurvey evidence

    The leading-value cohort is more likely to intend fundamental transformation and pursue growth alongside efficiency.

    3.3x more likely to intend transformation

    Method

    The state of AI in 2026. Cohort comparison, observational; high-performer cohort n = 92. Themes: Strategy.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2026

    BSurvey evidence

    Nearly three quarters of the leading-value cohort fundamentally redesign workflows, versus about a quarter of others.

    ~75% vs ~25%

    Method

    The state of AI in 2026. Cohort comparison, observational; high-performer cohort n = 92. Themes: Workflow.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2026

    BSurvey evidence

    The leading-value cohort is more likely to have defined processes to measure AI impact.

    2x as likely

    Method

    The state of AI in 2026. Cohort comparison, observational; high-performer cohort n = 92. Themes: Measurement.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • McKinsey

    2025

    BSurvey evidence

    Among organisational attributes tested, workflow redesign showed one of the strongest associations with reported EBIT impact from generative AI.

    Workflow redesign most associated with EBIT impact

    Method

    The State of AI 2025. Regression across survey respondents, associative. Themes: Workflow, Value.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • BCG

    2024

    BSurvey evidence

    Few companies create substantial AI value; leaders concentrate investment on fewer, higher-priority opportunities.

    4% substantial value; 22% some value

    Method

    Where's the Value in AI?. Survey of ~1,000 senior executives across 59 countries. Themes: Value, Strategy.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • BCG

    2026

    BSurvey evidence

    Organisations strong in both strategic clarity and applied AI generate far more AI value than those weak on both.

    5x AI value

    Method

    Applied AI Index 2026: The Formula for Agentic AI Value. Composite index across surveyed organisations. Themes: Strategy, Value.

    Index and survey analysis

    Open source ↗ (opens in a new tab)

  • BCG

    2026

    BSurvey evidence

    Almost half of organisations now generate value from AI, but only a small group generate significant value.

    Almost 50% generate value; 7.5% future-built; 41% scaling

    Method

    Applied AI Index 2026: The Formula for Agentic AI Value. Survey of 1,300+ CxOs. Themes: Value.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • PwC

    2026

    BSurvey evidence

    AI returns are concentrated in a leading cohort.

    Top 20% capture 74% of returns

    Method

    2026 AI Performance Study. 1,217 senior executives across 25 sectors. Themes: Value.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • PwC

    2026

    BSurvey evidence

    Most AI leaders have redesigned workflows around AI; few Australian enterprises have.

    56% of AI leaders vs 7% of Australian enterprises

    Method

    2026 AI Performance Study (Australian findings). Cohort comparison, observational. Themes: Workflow.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • PwC

    2026

    BSurvey evidence

    Leaders are more likely to have Responsible AI frameworks and cross-functional governance; employees are more likely to trust outputs.

    1.7x frameworks; 1.5x boards; 2.1x trust

    Method

    2026 AI Performance Study. Cohort comparison, observational. Themes: Governance.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • PwC Australia

    2026

    BSurvey evidence

    Australian enterprises lead the global sample on AI security and trail AI leaders on most measures of redesign.

    7% vs 56% redesigned workflows; 73% vs 69% robust protections; 6.8 months pilot to value

    Method

    PwC AI Performance Study: Australian companies lead on AI security. Australian subsample of PwC's 2026 global study, published 20 April 2026. Themes: Workflow, Governance, Value.

    Large-sample survey

    Open source ↗ (opens in a new tab)

  • NBER

    2025

    AField evidence

    Individual AI provision reduced email time but no corresponding shift in quantity or composition of work was detected.

    ~2 fewer hours a week on email

    Method

    Shifting Work Patterns with Generative AI. 7,137 knowledge workers across 66 firms. Themes: Academic, Workforce, Value.

    Note: Run by Microsoft Research on Microsoft 365 Copilot across 66 large firms. Firm-specific differences were the strongest predictor of how much workers used the tool.

    Randomised field experiment

    Open source ↗ (opens in a new tab)Second version ↗ (opens in a new tab)

  • MIT Sloan

    2026

    BSurvey evidence

    Structured support for experimentation produced far more organisation-wide solutions than unstructured adoption.

    141 vs 3 organisation-wide solutions

    Method

    Why some organisations turn AI experiments into business value. Two-organisation, two-year field study (working paper). Themes: Academic, Learning.

    Note: An academic medical centre and a corporate law firm. In the law firm, more than 80% of domain experts dropped out of AI innovation efforts.

    Longitudinal field study

    Open source ↗ (opens in a new tab)

  • Accenture

    2025

    BSurvey evidence

    Executive buy-in and leadership understanding are associated with higher reported ROI.

    2.5x ROI; 6x leader understanding

    Method

    Making Reinvention Real with GenAI. 2,000+ client projects; 3,000+ C-suite executives. Themes: Leadership.

    Note: Draws on Accenture's own client projects.

    Multi-project empirical analysis

    Open source ↗ (opens in a new tab)

  • World Economic Forum

    2025

    BSurvey evidence

    Skills and leadership vision are leading barriers to AI adoption.

    50% skills; 43% leadership vision

    Method

    Future of Jobs Report 2025. Global employer survey. Themes: Workforce, Leadership.

    Large-sample employer survey

    Open source ↗ (opens in a new tab)

  • Microsoft

    2026

    BSurvey evidence

    Frontier professionals more often refine processes together, share mistakes and document repeatable human-agent workflows.

    Higher frequency of shared learning behaviours

    Method

    Work Trend Index 2026. Multi-country worker survey. Themes: Learning, Workforce.

    Large-sample survey plus telemetry

    Open source ↗ (opens in a new tab)

  • EY-Parthenon

    2026

    BSurvey evidence

    CEOs see AI as a productivity driver but many struggle to convert gains into financial outcomes.

    50% cite AI; 23% struggle to convert; only 16% have clear, real-time visibility of AI ROI

    Method

    CEO Outlook 2026. Global CEO survey. Themes: Value, Measurement.

    Large-sample CEO survey

    Open source ↗ (opens in a new tab)

  • IBM

    2026

    BSurvey evidence

    About half of finance functions track AI-driven value and reallocate capital, but very few lead enterprise-wide AI value goals with automated investment triggers.

    48% track and reallocate; 8% lead with automated triggers

    Method

    CFO study 2026. Survey of 1,500 CFOs. Themes: Measurement, Leadership.

    Large-sample C-suite survey

    Open source ↗ (opens in a new tab)

  • Gartner

    2026

    BSurvey evidence

    Few CIOs report significant or transformational AI value; vanguard organisations apply stronger economic discipline, including explicit stopping rules.

    13% significant value; 31% have pilot stopping rules; 17% retire low-value AI in production; 2.6x vanguard likelihood with exit rules

    Method

    CIO survey on AI value 2026. Global CIO survey. Themes: Governance, Measurement, Value.

    Large-sample CIO survey

    Open source ↗ (opens in a new tab)

  • Editorial synthesis

    2026

    CSynthesis

    Hours saved become ROI only when the organisation decides where released capacity goes.

    Heuristic

    Method

    Capacity conversion heuristic. Synthesis across sources listed here. Themes: Value, Measurement.

    Expert synthesis

    Our synthesis, no single source

  • Stanford Digital Economy Lab

    2025

    BSurvey evidence

    Early-career workers aged 22 to 25 in the most AI-exposed occupations saw a relative decline in employment, while experienced workers in the same occupations were stable or grew.

    16% relative employment decline, ages 22 to 25

    Method

    Canaries in the Coal Mine? (November 2025 update). US payroll data, controlling for firm-level shocks; an August 2025 version reported 13%. Themes: Workforce, Academic.

    Observational payroll analysis

    Open source ↗ (opens in a new tab)

  • C.H. Robinson

    2026

    DCompany-reported case

    Lean workflow redesign before automation, hundreds of in-house AI agents and red or green status reporting accompanied a large productivity uplift during a freight downturn.

    45% productivity uplift since 2022 (company-reported)

    Method

    Fortune Eye on AI interview; Q1 2026 earnings summary. Executive interview and company earnings materials, not independently audited. Themes: Cases, Workflow, Value.

    Company-reported case

    Open source ↗ (opens in a new tab)

  • JPMorgan Chase

    2025

    DCompany-reported case

    A two-pillar strategy: top-down redesign of priority journeys and bottom-up opt-in self-service use, with financial benefit tracked per initiative.

    Gross benefit credited to AI growing about 30 to 40% a year (company-reported)

    Method

    Derek Waldron on building an AI-first bank culture (McKinsey). Interview with the chief analytics officer, not independently audited. Themes: Cases, Strategy, Measurement, Learning.

    Executive interview

    Open source ↗ (opens in a new tab)

  • Moderna

    2025

    DCompany-reported case

    Merged people and technology functions so one plan covers the work itself, with roles created, removed and redesigned as a result.

    Merged people and technology functions (structural change)

    Method

    Why Moderna merged HR and IT (UNLEASH). Interview with the Chief People and Digital Technology Officer; no published financial result. Themes: Cases, Workforce, Leadership.

    Executive interview

    Open source ↗ (opens in a new tab)

  • Klarna

    2025

    DCompany-reported case

    Cautionary case: an AI assistant cut customer service cost, but the company later said cost had dominated evaluation, quality fell, and it began recruiting human agents again.

    Reversed after quality fell (company-reported)

    Method

    Klarna and the return of human agents (Fortune). Company press release and CEO statements reported by the press. Themes: Cases, Measurement, Workflow.

    Company-reported case

    Open source ↗ (opens in a new tab)

A7Appendix · Sources

Sources

Research synthesis prepared October 2026. Sources use different samples, definitions and methodologies. Claims are stated within the limits of the underlying evidence.