Dominic Chiappe · People, capability & transformation

Thinking about how organisations perform in an AI-enabled world

AyEye Opinion · Exploration paper ·

Where in an organisation does integrated workforce management belong and how do we make it work?

Integrated workforce management belongs wherever an organisation can reconcile delivery, capability development and fair treatment with authority to act.

Large turquoise and deep-blue breaking waves fill the scene with white splashes beside dark angular rocks, leaving a small calm pool next to ochre sand.
Original artwork: Gathering Possibility. Chiappe × OpenAI.

Executive view

Integrated workforce management belongs wherever an organisation can reconcile delivery, capability development and fair treatment with authority to act. The evidence does not identify a universally superior department. Our provisional judgement favours supported delegation: decisions close to work with capable managers, accessible specialists and shared assurance. That is stronger as a design proposition than as a proven organisational model.

AI could make routine people and work capabilities easier to access inside daily operations. It does not follow that specialist expertise becomes unnecessary or that individual time savings become enterprise growth. Transformation requires redesign, investment and choices about how released capacity is used.

For 2027–2030 we examine three plausible configurations: specialist-led augmentation, embedded shared services and more autonomous local orchestration. They can coexist. Their relative value depends on managerial capability, operational interdependence, reliable information and the cost of oversight. The decision is not whether to embrace every new capability. It is which changes are worth making under which conditions.

The tide does not choose the wave

Watching the surf in St Ives prompted an observation: however large the waves, the tide still comes in with relative predictability. Even then, we choose which wave to ride.

That is a useful starting point for thinking about AI, not a law of organisational change. Technology creates pressure and possibilities. Economics, institutions and deliberate choices determine how they become practice. We do not need to ride most waves. We need to build the capability to recognise and ride the one worth committing to. That means preparing for significant opportunity while judging readiness and value rather than excitement alone.

An organisation may be running an established service, manufacturing products or developing something new. Who decides how employees, contractors and technology contribute? Who releases people from current commitments? Who develops tomorrow’s expertise while meeting today’s demands? How do we transform without interrupting the business?

Here, integrated workforce management means coordinating capabilities to deliver work across organisational boundaries. It includes planning, allocation, development, transformation and accountability. It does not mean treating people as interchangeable resources or pretending that an AI agent is an employee.

What the research can tell us

This is a selective evidence-led analysis rather than a systematic review. It combines workforce surveys, HR implementation research, transformation cases and field studies of AI assistance. Designs range from randomised interventions to observational accounts. We distinguish their findings from our recommendations. None directly compares all four ownership models for an integrated employee, contractor and AI workforce. The future configurations below are scenarios rather than forecasts with assigned probabilities.

The coordination problem predates the current generation of AI. A 2022 MIT Sloan Management Review and Deloitte study surveyed 4,078 leaders and managers. Some 74% considered effective management of external contributors critical to success. Only 30% felt sufficiently prepared for greater reliance on them. Its executive interviews explored responsibilities spanning HR, procurement, finance, IT and business leadership. This is descriptive research conducted with a consultancy partner. It identifies a management gap rather than proving a preferred reporting line. Research summary

Integration also has limits. In that study, PlanOmatic’s chief executive described unsuccessful efforts to draw contractors into the company’s culture. They had other clients and their own professional identities. Coordination need not require everyone to belong in the same way. Research summary

Our interpretation is that integration should make contributions fit together while respecting different relationships. Uniform treatment could undermine the flexibility people value.

Four plausible organisational homes

The following comparison is our organisational analysis rather than a finding established by the studies. These patterns can overlap: an HR-led or operations-led arrangement could also distribute decisions across business units.

HR-led ownership connects workforce planning with skills, development, employee experience and mobility. It is attractive where building capability is the central challenge. It needs operational influence and a remit beyond employees. Otherwise it can become an excellent view of people with limited authority over their work.

Operations-led ownership places allocation close to demand, delivery and service quality. It may suit organisations where capacity decisions are frequent and tightly connected. Its challenge is protecting learning, wellbeing and future capability when immediate output dominates.

Business-unit ownership with shared standards keeps decisions close to customers and specialised work. Different units can adapt their approach. The costs can include duplicated systems, competition for scarce expertise and restricted mobility unless enterprise rules have teeth.

Joint enterprise ownership brings the functions together under an accountable executive. This can reconcile competing priorities across the organisation. It can also create a committee that discusses everything while deciding little. The executive’s authority, budget and escalation rights matter more than the meeting’s title.

These are alternatives to test rather than a ranking. Each could support embedded capabilities or leave them concentrated in specialists. The choice of lead function and the distribution of everyday capability are separate questions. None of the studies discussed here establishes the right organisational owner.

Which arrangement looks most effective?

Our best current judgement favours supported delegation: decisions close to the work with capable managers, accessible specialists and shared assurance. Confidence is moderate in those ingredients but low in any claim that a particular department should always lead. We have not found a direct outcome comparison of the four models across a combined employee, contractor and AI workforce.

Outcome evidence cautions against equating devolution with effectiveness. In a study of 99 Jordanian financial firms, routine HR devolution had no statistically significant independent association with turnover or financial performance in the adjusted models. Strategic HR involvement did. The study linked a 2008 HR-director survey with earlier financial records so it cannot establish that HR involvement caused better performance. Darwish and Singh, 2013

A Cranet-based analysis also argues for strategic HR integration accompanied by a prominent specialist role relative to line managers. Its observational modelling does not establish the best reporting line but it challenges the idea that embedding capability makes specialist influence dispensable. Dany and colleagues, 2008

Coordination evidence offers another clue. A nine-hospital study associated stronger communication, shared goals and mutual respect across care providers with better service outcomes. It was observational rather than a test of reporting lines. The implication we draw is to strengthen coordination where work is interdependent rather than assume that moving a box on the organisation chart will achieve it. Gittell and colleagues, 2000

Employee outcomes can diverge from financial results. A longitudinal case study found growing HR departmental importance and strong financial performance alongside declining commitment and morale. This is a warning about what to measure rather than proof that devolution caused the decline. Hope Hailey and colleagues, 2005

The following recommendations are conditional interpretations of that evidence and the cases below. Their strength lies in the design logic rather than a proven ranking of organisational homes.

Where managerial capability or consistency is weak: start with stronger HR-led standards, coaching and review before devolving consequential people decisions. Keep operational leaders responsible for delivery. Evidence supports attention to implementation capability more directly than it supports HR holding the overall lead permanently.

Where work is tightly interdependent: an operations-led arrangement with embedded HR expertise or a joint enterprise lead is a plausible choice. Shared capacity decisions and escalation should span the whole workflow. The case for coordination is stronger than the evidence distinguishing which of those two leads is better.

Where units face genuinely different local needs: business-unit discretion within common standards is a reasonable hypothesis when managers are capable and outcomes are visible. Test whether local gains create costs elsewhere or unequal treatment. The evidence reviewed does not establish a size threshold or show that federation is generally superior.

Where failures can have serious consequences: retain specialist approval and independent challenge for the relevant decisions even if everyday allocation is local. Regulation, worker representation and established agreements should shape those boundaries. This is a governance recommendation rather than a demonstrated finding that more centralisation produces better outcomes.

For a smaller organisation, these responsibilities may be held by a few people rather than several departments. For a larger one, the same labels may conceal many different arrangements. Judge actual decision rights, support and results rather than organisational size alone.

Give ownership something concrete to own

Our provisional hypothesis is that integrated workforce management needs a clear owner of the overall outcome alongside explicitly distributed decision rights.

For an established service, a manufacturing operation or a new venture, business leaders might own demand and delivery. Operations might allocate available capacity. HR would shape development, progression and employee arrangements. Procurement would manage external sourcing and supplier commitments. Finance would test the economics. IT and security would establish technical access and controls with the relevant risk specialists.

The difficult decisions sit between those responsibilities. Can a manager retain someone whose skills are needed elsewhere? Who funds training when another unit receives the benefit? Who pays for checking an agent’s output? Who can pause work when quality deteriorates?

Our proposed starting point is a regular demand-and-capacity review for a defined part of the operation. Record the work required, available skills, existing commitments and constraints. Give each allocation decision a named decision-maker and a deadline. Escalate unresolved conflicts to the accountable lead. Review delivery, workload and capability development together rather than treating utilisation as the goal.

Employee and contractor perspectives belong in that discussion. A technically efficient allocation can still create unreasonable workloads or remove meaningful opportunities to learn. Agreement about how concerns are raised should accompany agreement about who assigns work.

Transforming while the business keeps running

The following is our proposed approach to transformation rather than a sequence validated by the research cited here. Its pace and safeguards would need to fit the work, particularly where physical processes or safety-critical responsibilities are involved.

Start with where work actually happens. Map the contribution of employees, contractors and existing technology alongside informal handovers, waiting time, exceptions and rework. Establish a baseline for quality, capacity, workload and continuity. The organisation chart alone will not show all of this.

Then ask what should change before deciding what to automate. Could a handover disappear? Is the task still needed? Would different access to expertise solve the problem? Redesign may involve new roles and decision rights as much as new tools. Specify who can authorise actions, accept outputs and intervene when something goes wrong.

Development and redeployment belong in that design. If people-management responsibilities move closer to teams, specify the coaching, specialist access and protected time that will make them usable. Test what remains centrally assured as carefully as what becomes local. Identify which capabilities people will need, provide time to learn and agree how changed responsibilities affect their work. A transition plan should also account for the capacity needed to maintain the current operation while people learn the new one.

A bounded trial could test the proposed arrangement against the baseline with suitable comparison where feasible. Agree quality and continuity thresholds, a named person who can stop the trial and a workable fallback before it starts. Expand only when benefits hold after accounting for checking and transition costs without unacceptable effects on people or delivery. Compare results across teams and examine worker feedback alongside managerial reports. Specify the acceptable differences before reviewing the trial rather than selecting favourable measures afterwards. If benefits fail to hold or agreed safeguards are breached, revise or stop. A successful local trial would still leave questions about wider dependencies and scale.

What actual transformation tells us

Cochlear’s move towards lean manufacturing offers a concrete example of partnership rather than simple delegation. Researchers described HR staff working within manufacturing while team leaders took greater responsibility for coaching and skills assessment. Employees could approach HR outside their supervisory relationship. Yet some operators reported favouritism in assessments. The company subsequently added shift supervisors to coach team leaders. This qualitative case suggests that embedding responsibilities required support and correction rather than the removal of specialist oversight. A single purposively selected company cannot establish comparative superiority or isolate the cause of its production gains. Gollan and colleagues, 2015

Stronger causal evidence for supported operational change comes from a randomised experiment at 20 plants within 17 Indian textile firms. Intensive implementation support for a package of management practices improved first-year productivity by 17%. The package included quality and inventory practices so the experiment does not isolate workforce governance or establish whether HR or operations should lead. Its small selected sample and substantial free consulting also limit transferability. Bloom and colleagues, 2013

A nine-year follow-up found that about half the adopted practices had been dropped although a significant treatment–control gap remained. Managerial turnover and lack of director time were frequently cited reasons. Those explanations were not independently randomised. Our inference is that capability needs renewal and continuing attention rather than a one-off implementation programme. Follow-up, 2020

AI assistance is not yet organisational transformation

A peer-reviewed study of 5,172 customer-support workers found a 15% increase in issues resolved per hour. Benefits were greater for less-experienced workers with evidence consistent with learning. Some quality measures declined among the most skilled workers. This was assistance in one firm rather than autonomous workforce management. Generative AI at Work, 2025

A larger randomised field study tested Microsoft 365 Copilot access among 7,137 knowledge workers in 66 firms. During weeks 13–24, the intention-to-treat estimate was 1.4 fewer hours a week using email. Researchers did not detect changes in the quantity or composition of observed Office tasks from individual provision. The study used product telemetry but did not directly measure work quality or enterprise productivity. Microsoft researchers participated. The result separates time savings from work redesign rather than proving redesign would succeed. Shifting Work Patterns with Generative AI

An OECD survey of more than 5,000 SMEs in seven countries offers another distinction. Among generative AI users, 83% reported no change in overall staff need while about 14% reported reduced reliance on external contractors and 6% reported increased reliance. These are managerial reports from late 2024 rather than causal employment estimates. Our inference is that workforce effects may include changes in purchased services and internal capacity rather than simply employee headcount. OECD, 2025

Nor should we assume that a human–AI combination always beats either alone. A meta-analysis of 106 experimental studies found improvement over humans alone on average but worse performance than the better of humans or AI alone. The studies were published between January 2020 and June 2023 and mostly concerned decision tasks. They are not a verdict on current agents. They challenge us to design the division of work rather than assume collaboration creates value. Nature Human Behaviour, 2024

Human accountability remains necessary. Effective review requires expertise, time and usable evidence rather than a nominal approval step.

Could technology make an embedded service more plausible?

Organisational ownership need not dictate one application for everything. Workforce plans, authoritative records, execution tools and everyday interfaces may sit in different places.

Our proposed test of openness goes beyond the existence of an API. Can systems exchange information with consistent meanings? Can permissions follow the right person or agent? Can decisions be traced? Can a component be replaced without losing essential history? Are prices, quotas or contractual restrictions making nominally open connections impractical?

Consider a virtual process agent helping a manager fill a short-term capacity gap. Within permitted data, it might retrieve availability, suggest people with relevant skills, check documented rules, prepare an approval request and follow up agreed actions. These are illustrative possibilities rather than a claim about how much HR work can be automated. Fixed rule checks can use deterministic automation. Skills suggestions are decision support. Taking an action on someone’s behalf requires explicit delegated authority with exceptions routed to a named reviewer.

Our conjecture is that such services could make routine information and coordination available inside everyday work while specialists concentrate on difficult judgements, employee protection and organisational development. Larger datasets could improve the evidence available but volume alone cannot establish accuracy, fairness or relevance. An emerging interface should make advice easier to obtain without obscuring who decides or who can intervene.

Some building blocks already exist. The MCP authorization specification describes scoped access to protected servers for supported implementations. W3C’s established PROV data model represents where information came from and the activities and actors involved. These provide mechanisms for controlled access and traceability rather than evidence that an integrated workforce service improves organisational outcomes. Neither guarantees accurate skills data or sound decisions. MCP specification, W3C provenance model

Transparency should mean that an affected person can understand the relevant information, see who authorised a decision and challenge it. It should not mean exposing everyone’s personal data. Interoperability also requires shared meanings for skills, capacity and commitments rather than simply connected APIs.

There are reasons for caution: outdated skills profiles, opaque recommendations, excessive surveillance, inappropriate access and additional checking work could make the service worse than existing arrangements. A smooth interface could conceal fragmented responsibility. Our test would be whether people gain usable guidance and contestable decisions while accountable specialists remain reachable. Technical plausibility is a reason to experiment carefully, not proof that embedded management has become effective.

The operational test should be equally concrete. Carry the baseline and safeguards from the trial into ongoing reviews. Measure accepted outcomes, delays, rework, supervision and learning. Then ask what additional capacity makes possible: more customers served, improved quality, a new offering or greater resilience. Account for integration, checking, rework, training and transition costs. Track whether released capacity is actually used to deliver those benefits. Faster task completion alone does not demonstrate growth.

Three plausible configurations for 2027–2030

This horizon is an analytical planning window, not a promised timetable. We assume continuing availability of AI assistance but not a particular rate of technical progress or autonomous reliability. Adoption also depends on affordability, permissions, credible data and institutional acceptance. These configurations can coexist within one organisation. An HR-led function might support all three for different decisions.

1. Specialist-led augmentation

Specialists retain consequential decisions while managers and employees use assistance for information, preparation and routine administration. This is plausible where rules are complex, data are fragmented or managerial capacity is weak. The advantage is consistency and accessible expertise. The downside is a specialist bottleneck hidden behind a faster interface.

Watch for: shorter end-to-end resolution times without longer specialist queues, repeat enquiries or poorer employee access. Measure whether preparation actually reduces specialist effort.

Revise the approach if: local decisions remain unnecessarily slow despite sound guidance and capable managers. Selective delegation may then be better than adding more central processing. Conversely, unresolved errors argue against expanding autonomy.

2. Embedded shared workforce services

Managers and teams use common services for capacity planning, skills advice and approvals within enterprise standards. Specialists design policies, support exceptions and provide independent challenge. This is our leading conditional hypothesis where work crosses units and the organisation can maintain common information and protected management time. It is not a research-established winner.

The potential gain is local responsiveness without abandoning shared capability. The risks are hidden managerial workload, biased allocation and a central service that offers advice without owning failures.

Watch for: actual cross-unit mobility, fewer unresolved allocation conflicts, consistent treatment and credible routes to challenge decisions. Compare checking time with the coordination work removed.

Revise the approach if: disparities, escalation delays or development losses persist after credible support. Return affected decisions to specialists or narrow the service rather than interpreting every failure as resistance to change.

3. More autonomous local orchestration

Business units delegate bounded actions to process agents while retaining enterprise controls for consequential commitments. This could be attractive where workflows are modular, rules are explicit and mistakes are detectable and reversible. It is less plausible for poorly specified judgements or tightly coupled operations where a local action creates distant consequences.

Its potential is faster coordination and experimentation. Its downside is duplicated capability, opaque commitments, access failures and fragmented accountability. Technical reliability would be necessary but insufficient: someone must own the combined workflow.

Watch for: accepted outcomes after review and rework, exception frequency, recovery time and the full cost of maintaining permissions and integrations. A high task-completion rate alone is weak evidence.

Revise the approach if: human checking absorbs the capacity released or failures become difficult to contain. Reduce delegated scope or return to decision assistance. Waiting for better evidence is a legitimate choice.

Across all three, review by decision type rather than declare one enterprise-wide destination. Routine preparation, work allocation, development and employment judgements need not share the same degree of delegation. Publish the boundaries internally so people know who can act, who can intervene and where accountability remains.

A wider evolution, with reasons for doubt

One possible evolution is personnel, then HR, then integrated workforce management and eventually more widely embedded people and work capabilities. We offer this as a hypothesis about changing emphasis rather than a universal history or an inevitable sequence. Administration, employee relations, strategic planning and everyday management can coexist throughout. Changing the department’s name tells us little about where capability actually sits.

The useful possibility is that work design, capacity planning and development become everyday management capabilities supported by specialists. This is different from transferring administrative tasks or expecting an overloaded team leader to become an expert in everything.

Research gives reasons for both interest and caution. A study of 174 line managers and 1,065 direct reports in two organisations found managerial ability the strongest predictor of effective HR implementation. Opportunity strengthened that relationship but the predicted moderating effect of motivation was not supported. These are survey findings rather than proof that devolving responsibility improves outcomes everywhere. Bos-Nehles and colleagues, 2013

A study of gender-equality reforms in Australian public services identified a less reassuring pattern: policies became disconnected from practice when communication, implementation and monitoring were inadequate. Its context is specific but it challenges any assumption that responsibility becomes effective merely by moving closer to the work. Williamson and colleagues, 2020

Nor is there one organisational destination. Research covering 2,335 organisations in 21 countries found that the distribution of HR decisions was associated with both organisational and national characteristics. It examined patterns of devolution rather than testing an evolutionary ladder or identifying a best model. Gooderham and colleagues, 2015

Our conjecture is therefore selective embedding with retained specialist strength. The comparative evidence remains incomplete, especially for contractors and AI agents working alongside employees. Employment expertise, fair standards, employee advocacy and independent challenge should not disappear into the delivery hierarchy. People need a credible route to raise concerns beyond their immediate manager. Some decisions require distance from local pressures as well as knowledge of the work.

The practical questions follow. Do managers have time and competence? Do their incentives reward development and fair treatment alongside output? Who checks consistency across teams? Who intervenes when local priorities undermine enterprise capability? An overloaded manager with additional HR duties would be a poor test of this hypothesis.

This hypothesis should be revised if wider delegation produces less consistent treatment, weaker learning or slower escalation even after credible support and resources are provided. Such findings would favour retaining or rebuilding specialist control over the affected decisions. Local responsiveness may be worth less than consistency in some settings. Other organisations may combine embedded everyday practices with stronger central assurance rather than move towards greater distribution.

The evolution worth exploring is how these capabilities become part of running and transforming the organisation. It is not a prediction that an HR department disappears or that one system absorbs another. We should judge possible futures by better work, fair treatment and clear accountability rather than by how convincingly they fit an evolutionary story.