An organisation's balance sheet records what it owns and owes, but much of what allows it to earn is not recorded there. The knowledge, judgement, relationships and tacit skill held by its people produce the income on which every projection depends. Yet the cost of keeping that capability current is treated, in most accounting and budgeting systems, as an expense of the period in which it is incurred. This section argues that development is better understood as a provision for future productive capacity, and considers what follows when an organisation adopts artificial intelligence under conditions of rapid change.
The argument has an individual counterpart. A household can fund its retirement carefully and still be exposed, because the plan assumes the earner will remain employable and makes no provision for maintaining the capability that produces the income. The organisational version has the same structure at larger scale. An enterprise can have sound strategy, adequate capital and well-chosen technology, and still be exposed if nothing has been set aside to keep its people able to use what it has bought.
Capability as an asset that depreciates
Economists have treated education and training as investment since Becker (1962), who showed that spending on skills raises productive capacity and future earnings. The practical implication is that money spent to preserve or build capability is not automatically discretionary. It may be closer to maintenance of a productive asset than to a benefit offered to staff.
Unlike a plant or a portfolio, this asset depreciates through relevance. The World Economic Forum (2025), drawing on a survey of more than 1,000 employers, reports that employers expect around 39% of workers' existing skills to change or become outdated between 2025 and 2030. This is an expectation and not a measured loss of jobs, but it indicates the scale of adjustment that organisations themselves anticipate. Autor (2015) offers a useful way to read it. Technology does not only replace human labour. It also complements it, which alters the relative value of different tasks and skills. The more common risk for an organisation is therefore not that its people become redundant but that its capability drifts. Experienced staff remain knowledgeable while the mix of technical skill, judgement, communication and technological fluency that the environment rewards moves away from the mix the organisation holds. For the purposes of this book, organisational drift can be defined as the gradual divergence between the capabilities an organisation holds and those its environment rewards.
Why development is deferred
If the case is so intuitive, the question is why development is so often the first line to be cut. Behavioural science offers three overlapping explanations. The first is present bias. Laibson (1997) described how people discount the future steeply when a choice involves an immediate cost and a delayed benefit, and development fits that pattern closely, since the cost is certain and immediate while the benefit is diffuse and arrives later. The second is mental accounting. Thaler (1999) showed that people sort money into mental categories with different rules attached, and training budgets commonly sit in a discretionary category that is treated as expendable when pressure rises. The third is structural. Because training is expensed, the capability it builds is invisible on the balance sheet, so the loss that follows from underinvestment is also invisible until it surfaces as an operational failure.
Consider a hypothetical engineering services firm in a lean year. The training budget is reduced by a third, and nothing noticeable happens for eighteen months. By the time the consequences appear, as slower onboarding, thinner succession cover and a team that cannot absorb a new analytical tool, the people who made the decision have moved on and the link between cause and effect is difficult to reconstruct. The behavioural lesson is that the feedback loop is too long and too weak for intuition to correct the decision, which is precisely why it needs a structural safeguard.
The economic case for the employer to invest
A standard objection comes from Becker himself. If a skill is general, meaning that it is valuable to other employers as well, then a rational firm should not pay for it, because the employee can take it elsewhere. Acemoglu and Pischke (1999) challenged this in their analysis of training in imperfect labour markets. Where wage structures are compressed and mobility is costly, firms capture part of the return on general training and therefore have reason to fund it. In a market with scarce skills and real frictions in moving between employers, the poaching argument is weaker than it first appears. It does not disappear, but it has to be weighed against the cost of an unmaintained workforce and not treated as a reason to avoid development altogether.
The resource-based view of the firm points in the same direction from a different starting point. Barney (1991) argued that sustained advantage comes from resources that are valuable, rare and difficult to imitate. Technology that any competitor can purchase rarely meets that standard. A workforce whose capability is continuously renewed and embedded in the organisation's routines comes closer to it, because the accumulated judgement is difficult to copy.
Development as the conversion mechanism for AI adoption
This is where the argument connects most directly to the central concern of this book. Artificial intelligence tools can be bought, licensed and deployed quickly. The capability to use them with judgement cannot. Where the tools are widely available, advantage shifts to how well people integrate them into decisions, and that is a behavioural question as much as a technical one. Insufficient development tends to produce two opposite failures. In the first, people under-use the technology because they do not trust it or do not understand where it helps. In the second, they rely on it too readily. Parasuraman and Manzey (2010) reviewed the evidence on complacency and automation bias and found that both can occur even in trained and experienced users, which suggests that exposure alone does not protect against them and that deliberate development of calibrated trust is required.
In a safety-critical or high-reliability setting the stakes of that calibration are higher, and the cost of an uncorrected failure is not recovered by a later training course. Read through the ARIF-BARI lens, development is one of the levers by which behavioural adoption risk can be reduced, because it acts on the gap between having a tool and using it well. That is a proposition for empirical testing and not an established finding, and it should be presented as such.
The psychological contract and retention
Rousseau (1989) described the psychological contract as an individual's belief about the terms of the exchange between themselves and their employer, much of which is implied and not written. Perceived investment in an employee's development is commonly part of that implied exchange. When an organisation asks people to absorb AI-driven change in how they work, the implicit offer is that it will help them remain valuable. Withdrawing support while demanding adaptation can be experienced as a breach, with predictable consequences for commitment.
There is also a selection effect worth considering, although it is an inference and not a tested result. The employees with the strongest outside options are usually the first to leave when development stops, which means the organisation may lose capability disproportionately at the point where it most needs to retain it.
What the evidence does and does not show
The financial claim deserves the most careful handling. Huselid (1995) reported associations between high-performance work practices, which included training, and lower turnover, higher productivity and better corporate financial performance. Edmans (2011) found that firms ranked as good employers earned higher long-run stock returns. Both studies are influential, and both are observational. Well-run firms may simply be those that can afford to invest in people, so reverse causality and omitted variables remain live explanations. The Edmans measure also captures employee satisfaction broadly and not development specifically. Evidence on the return from any individual programme is mixed and depends heavily on whether learning transfers to the job, which is a behavioural problem and not a budgeting one.
The defensible position is therefore a probabilistic one. Development is a necessary condition for maintaining capability, the theoretical case for treating it as provision is strong, and the expected financial benefit is positive but variable and difficult to attribute. It should not be presented as a guaranteed multiple on spend.
The South African context
South African employers already operate within a compliance framing for skills. Employers above the payroll threshold pay the Skills Development Levy, and the skills development element of the B-BBEE scorecard rewards defined categories of spend. These mechanisms have value, but they carry a behavioural risk. When development is managed to achieve scorecard points, spend can follow what is recognised and not what the organisation's capability map requires. The OECD (2025) identifies persistent skills mismatches and barriers to employment in South Africa, which raises the importance of getting the allocation right and not merely meeting the minimum.
A provision architecture
If development is a provision, it can be structured like one. A practical approach separates three purposes. The first is operating development, which keeps people proficient in their present roles. The second is transition development, which prepares people and teams for changes already foreseeable, such as the introduction of a new system or a restructured process. The third is a capability reserve, which is held deliberately for shifts that cannot yet be specified and is ring-fenced so that it is not the first thing released when short-term pressure arrives. The reserve mirrors, at organisational level, the separation between retirement capital, household continuity and career adjustment funds proposed for the individual.
Governance matters as much as allocation. A board that reviews capital expenditure and cash reserves has reason to review a capability map as well, identifying the skill clusters on which the organisation's earning depends, how concentrated they are in a few individuals, and how long it would take to rebuild them. A capability stress test can then be applied in the way a financial stress test is applied. If the three most critical skill clusters lost a quarter of their holders, or became materially less relevant, what would restoration cost and how long would it take? Measurement should lean on leading indicators such as capability coverage, time to proficiency, transfer to the job and concentration of critical knowledge, since training hours and attendance record activity and not capability.
Limits and trade-offs
The approach has limits that should be stated plainly. Capability maps date quickly, and an over-specified reserve can become a rigid commitment that is itself a source of drift. Ring-fencing reduces flexibility, which is a real cost in volatile conditions. Development that is not accompanied by role redesign and managerial support often fails to transfer, so a larger budget does not by itself produce a more capable organisation. And an organisation can over-invest, funding development in areas where the return is low. The case is not for spending more indiscriminately. It is for making the trade-off visible and deliberate, so that the decision to defer development is taken with its future cost in view.
Reflection prompts
Which capabilities does your organisation's earning depend on most, and who holds them? If a quarter of those people left, or the skills became less relevant within three years, what would restoration cost?
When development budgets were last reduced, what was the stated reason, and what did you observe eighteen months later? What would have made the delayed consequence visible earlier?
Where in your AI adoption plan is development treated as a cost of rollout, and where is it treated as the mechanism through which the investment converts into value?
If your development spend is shaped mainly by compliance requirements, what would a spend shaped by a capability map look like, and where would the two differ most?
Glossary entries
Capability provision. A deliberate allocation of resources to maintain and renew the knowledge, skill and judgement on which an organisation's future earning depends, treated as maintenance of a productive asset and not as discretionary expenditure.
Organisational drift. The gradual divergence between the capabilities an organisation holds and those its environment rewards, which may occur without any single event that signals it.
Capability reserve. A ring-fenced portion of the development allocation held for shifts in capability requirements that cannot yet be specified.
References
Acemoglu, D., & Pischke, J.-S. (1999). Beyond Becker: Training in imperfect labour markets. The Economic Journal, 109(453), F112 to F142.
Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. Journal of Economic Perspectives, 29(3), 3 to 30.
Barney, J. (1991). Firm resources and sustained competitive advantage. Journal of Management, 17(1), 99 to 120.
Becker, G. S. (1962). Investment in human capital: A theoretical analysis. Journal of Political Economy, 70(5, Part 2), 9 to 49.
Edmans, A. (2011). Does the stock market fully value intangibles? Employee satisfaction and equity prices. Journal of Financial Economics, 101(3), 621 to 640.
Huselid, M. A. (1995). The impact of human resource management practices on turnover, productivity, and corporate financial performance. Academy of Management Journal, 38(3), 635 to 672.
Laibson, D. (1997). Golden eggs and hyperbolic discounting. The Quarterly Journal of Economics, 112(2), 443 to 478.
Organisation for Economic Co-operation and Development. (2025). OECD economic surveys: South Africa 2025. OECD Publishing.
Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381 to 410.
Rousseau, D. M. (1989). Psychological and implied contracts in organizations. Employee Responsibilities and Rights Journal, 2(2), 121 to 139.
Thaler, R. H. (1999). Mental accounting matters. Journal of Behavioral Decision Making, 12(3), 183 to 206.
World Economic Forum. (2025). The future of jobs report 2025. World Economic Forum.