The 70 Percent CFOs Are Underinvesting In
BCG's 10/20/70 framework, places only 30% of AI value in the technology and data layers. The remaining 70% sits in people and process: talent, change management, role design, and ownership. Yet most AI implementation work in finance focuses on the 30%.
A recent HBR article by Guy Champniss reports significant psychological debt accumulating in employees as a result of AI adoption: measurable effects on motivation, autonomy, and competency that translate directly into reduced AI use, lower task complexity, and higher avoidance behaviour.
Furthermore, Wilson and Daugherty's seminal Harvard Business Review research on collaborative intelligence, combined with Accenture's 2025 survey of 14,000 workers across 12 countries, find organisations that systematically address these mechanisms achieve 5x higher workforce engagement, 4x faster skill development, and 2.8 percentage points higher revenue growth than peers. That makes role design a capital allocation question CFOs cannot delegate.
Negative effects on employees may proof a major barrier to AI adoption. In this article I discuss how to counter these.
Psychological Debt & AI Avoidance
Self-Determination Theory establishes that motivation depends on three conditions: autonomy, competence, and relatedness. The Job Characteristics Model identifies skill variety, task identity, task significance, autonomy, and feedback as the dimensions that predict engagement. Both predict that work redesigned in ways that erode these conditions will fail on engagement regardless of technical capability.
Recent research published in Scientific Reports by Przegalinska and colleagues confirms the mechanism in AI collaboration specifically. Across 3,562 participants in four experimental designs, AI collaboration improved immediate task performance but the gains did not transfer to subsequent independent work. More critically, transitioning from AI-assisted to solo work produced significant decreases in intrinsic motivation, even as workers reported greater perceived control. AI was completing tasks rather than building human capabilities.
Guy Champniss operationalises this for AI implementation specifically. Drawing on a survey of 1,200 employees across 10 sectors, the research identifies six dimensions of psychological debt that AI adoption can accumulate: cognitive debt (loss of decision-making capability through over-reliance), autonomy debt (sense of losing control over how work is done), competency debt (declining confidence in own abilities relative to AI), relatedness debt (decreased social interaction), credibility debt (concern about how peers view AI use), and professional identity debt (mismatch between AI use and group identity).
The findings are concrete. Psychological debt predicts AI avoidance behaviour: workers with high debt scores were significantly more likely to use AI rarely or not at all (debt score 60), compared to those using AI multiple times daily (36). High-debt workers also used AI for simpler tasks (debt 46) than those using it for complex strategic work (35). Workers in their first five years of employment carried the highest psychological debt (54) compared to those with over 20 years experience (40), suggesting the developmental stage where AI integration is most damaging is also where it is most prevalent.
The implication is direct: how AI is integrated into roles determines whether workers engage with it productively or accumulate psychological debt that undermines adoption. The investment is the same. The outcomes diverge based on design choices.
How to Design Human-AI Collaboration
Four task design principles, synthesised from this body of research and operationalised for finance in Chapter 5 of the CFO AI Playbook, address these mechanisms directly.
Preserve meaningful decision-making authority. AI provides analysis and recommendations; humans retain final authority on judgement-relevant decisions. In budget variance analysis, AI identifies unusual variances and potential causes, but FP&A analysts determine which require action. In credit risk assessment, AI scores applications but credit managers review borderline cases and make final approvals. ING's AI Principles in Practice illustrates how this gets operationalised: product teams document how human judgement is preserved before any model is deployed, through discussion and collaboration with the people who will use it. ING also provides "explainability", plain-language descriptions of how models work and "nutrition labels" showing data sources, model limitations, and known risks. This addresses autonomy debt directly: workers retain agency because the role design makes their judgement structurally central, not nominally retained.
Design for capability building, not task completion. AI shows reasoning, assumptions, and key drivers so workers learn patterns rather than consume outputs. In financial forecasting, AI displays sensitivities and assumptions so analysts can challenge predictions, rather than producing forecasts as black-box outputs. J.P. Morgan operationalises this by positioning AI as an insights provider rather than a decision-maker, insights are useless unless they fit into and support employee-generated arguments or hypotheses. The cognitive friction is deliberate: requiring employees to develop hypotheses or generate arguments before turning to AI ensures higher-order cognitive functions stay active. This addresses cognitive debt and competency debt by ensuring AI use builds expertise rather than substituting for it.
Design for progression, not flux. AI permanently handles routine, high-volume tasks while humans move to higher-value work. The critical principle is minimising situations where employees must return to routine work. Switching between higher-value work and routine tasks intensifies the boredom contrast that Przegalinska's research documented.
Manage the progression journey. Even with progressive role design, employees need support during the transition. This means framing the move away from routine tasks as career advancement, structuring progression through increasing complexity, and building team depth so routine work is covered without requiring workers to step backwards. Microsoft's Copilot Champs Community illustrates how this can be operationalised through peer-to-peer support, a grassroots initiative where employees explore role-relevant AI use in trusted environments rather than receiving top-down training. The peer dynamic addresses competency debt and credibility debt simultaneously: people learn from colleagues at their level, and AI use becomes a shared norm rather than an isolated exposure.
AI ROI & Role Design
The 5x engagement gap and 2.8 percentage points of revenue growth in the Wilson & Daugherty and Accenture findings represent the difference between organisations that treat role design as part of AI implementation and organisations that do not. The investment in AI tools is roughly the same. The returns diverge based on whether the work the tools integrate into has been designed in ways that work with the psychological mechanisms or against them.
This reframes the question CFOs should be asking. Not "what AI tools should we deploy?" but "how are we designing the work AI integrates into, and what return are we generating on that design choice?" Role design is not an HR concern about employee engagement. It is the variable with the largest measurable effect on whether the AI investment generates returns. Organisations capturing multiples of the available return are the ones that have made role design a strategic discipline alongside technology selection and governance.
APG's Man Machine Balance
APG, Europe's largest pension fund managing €577 billion, has operationalised these principles through its Man-Machine Balance philosophy. AI permanently handles analyst report differentiation and pattern recognition for SDG-investment identification. Senior investment professionals focus on decision-making, relationship management, and strategic positioning without switching back to routine tasks. Implementation was driven through Digi Tribes, hackathons, and a Digi Competition where winning ideas are guaranteed implementation, ensuring employees drive innovation rather than passively receive new tools. The cultural framing positions AI as opportunity rather than threat.
The structural choice is what matters. APG did not just deploy AI tools, it designed roles around the psychological mechanisms that determine whether AI adoption sustains or fails. Most finance functions have not.
