The Architecture of an Effective HR Function: Roles, Policies, Processes, and Accountability

An organization chart shows HR as a box with a director and a few teams beneath. That captures one element of the function's design and leaves out most of what determines whether it works. A working HR function has at least five elements. Decision rights settle who may set or change people rules, such as pay ranges, hiring approvals, and dismissals. Delivery roles settle who does the work, and in most organizations line managers carry out more day-to-day HR work than HR staff do. Policies are the organization's commitments: what employees are entitled to and what managers must do. Processes are the designed sequence of steps, forms, approvals, and system workflows through which each policy is carried out. Accountability settles who is answerable, and to whom, for whether the intended outcome actually happens.

These elements fail independently. A sound policy can run through a process nobody owns; a well-designed process can be run by managers who lack the authority to follow it; a clear structure can sit on decision rights never written down, so every difficult case goes to the managing director. The most common design error is to treat delivery roles as the whole architecture, so each reorganization redraws the boxes and leaves the other four elements untouched.

The five-element framework and this table are this article's own teaching tool, not a model taken from a single source.

The best-known answer to the delivery-role question is the model called, depending on who is describing it, the three-pillar, three-box, or "three-legged stool" model. A shared service centre handles high-volume transactions and routine queries: contracts, payroll changes, records, first-line policy advice, often through a portal and a contact centre. Centres of expertise (sometimes "of excellence") employ specialists who design policies and programmes in fields such as reward, talent, learning, and employee relations. HR business partners work alongside senior line managers on the people side of business decisions. Article 14 examines the business partner role in depth; here it matters as one component among several.

The structure is routinely called "the Ulrich model," and the label is only partly accurate. Dave Ulrich's 1990s work on shared services and HR roles supplied much of its logic, but, as Article 14 shows, he describes himself as having observed and written about the structure rather than created it. His fullest account of HR organization, with Jon Younger and Wayne Brockbank (2008), does not use the stool image. It describes five roles: service centres, corporate HR, centres of expertise, embedded HR, and "operational executors," who take on the casework and operational tasks (disciplinary cases, recruitment logistics, programme delivery) that otherwise fall between the other roles. Its first premise is the one most often lost in adoption: "HR organization should be structurally aligned with the organization structure of the business." By 2018 Ulrich was telling a conference that the stool had been "a good idea" at the time, but "in 2018, it's not a good idea because the world has changed."

The model is better read as one solution that suited large, diversified companies in a particular period than as a design standard. Service centres, the 2008 article notes, "emerged in the late 1990s as HR leaders (and other functional organizations such as purchasing) realized that many administrative tasks are more efficiently performed in a centralized, standardized way." That logic needs two conditions: enough transaction volume to make standardization pay, and enough variety among business units to need partners who understand each one. Where neither holds, the structure adds handoffs and little else.

Evidence on adoption comes mostly from case studies and practitioner surveys, and it points in a consistent direction. Lars Boglind, Freddy Hällstén, and Pia Thilander (2011) studied seven Swedish organizations through 192 interviews. All seven "adopted the HR transformation as a standard blueprint," with management consultants playing a leading role; service centres were set up, "the local HR staffs were reduced radically, and the remaining role, the HR business partners, took on lesser importance." The organizations then adapted the blueprint in varied ways, some of them innovative. The authors' conclusion is a warning about imitation: "It is not easy to imitate a theoretical model or a 'best practice' model without taking the translation process into consideration." Fang Lee Cooke's (2006) study of a multinational's UK service centre reached a related verdict: the "financial and emotional cost of moving to a shared services model can outweigh the tangible cost savings predicted by firms."

Survey evidence places the friction at the joins. A CIPD survey of senior HR professionals, as reported by Ulrich, Younger and Brockbank (2008), found that introducing shared services had produced "boundary problems (identified by 55% of respondents), gaps in service provision (42%) and communication difficulties (37%)," and that "the difficulty of separating out transactional work (45%) was even more of an issue." These are HR professionals' own perceptions, not independent measures of service quality, but the problems they named lie between units rather than within any one.

Why does a service centre so often settle into purely administrative work, when the model's purpose was to free the rest of HR for work that adds value? Much of the answer lies in how the centre is justified and measured. It is usually funded on projected savings and run on cost per transaction, resolution time, and accuracy. Those measures reward closing a query, not solving the problem behind it: a manager whose team has high absence can receive a correct answer about the sick-leave rule without anyone asking why absence is high. Article 11 traced how HR's claim to legitimacy moved from administering employment toward creating organizational value. A service centre judged only on transaction cost reverses that move for all the work passing through it. Abandoning standardization, which genuinely reduces errors in pay and records, would be the wrong response. A better one is to give each end-to-end process, such as hiring or leaving, a single owner across every unit it passes through, and to measure the outcome of the whole process (a vacancy filled by someone still there a year later, say) rather than each unit's step.

The second half of the architecture concerns what the function commits to and how commitments are carried out. Three terms are often used interchangeably and should be kept apart. A policy is a commitment, such as an entitlement to annual leave or a rule that no dismissal proceeds without a hearing. A process is the designed route by which the policy is carried out: the form, the approval chain, the deadline, the system workflow. A practice is what actually happens, as employees experience it. A leave policy can be generous and the application process sensible while the practice in one department is that leave is refused whenever the supervisor is short of staff.

The most widely cited model of this gap comes from Patrick Wright and Lisa Nishii (2007 working paper; published as a book chapter in 2013). It runs from intended HR practices, designed by decision-makers to produce the employee responses they want, to actual practices, then to perceived practices, employee reactions, and organizational performance. The reasoning is direct: "not all intended HR practices are actually implemented, and those that are may often be implemented in ways that differ from the initial intention." Because practices "must usually be implemented by multiple individuals (supervisors, interviewers, trainers, etc.) who will not be uniform in their implementation efforts," actual practice varies with the implementer, and each employee then interprets what he or she receives. One methodological point has practical force: surveying a single HR representative tends to capture intended practices, while surveying employees and line managers is more likely to capture actual ones. An HR function that audits its policies by asking HR measures only the first link in its own chain.

Empirical work supports the distinction. Shaista Khilji and Xiaoyun Wang (2006), using 195 interviews, 508 questionnaires, and company documents from Pakistan's banking sector, contrasted the reports of managers and non-managers inside and outside HR. They found that "implemented HRM may be substantially different from intended HRM," and that consistent implementation was associated with higher employee satisfaction with HRM, which in turn was positively related to organizational performance. Lisa Nishii, David Lepak, and Benjamin Schneider (2008) showed that even identical practices are read differently. In a supermarket chain where one document set HR policy for all employees, staff who believed practices existed to improve service quality and employee well-being reported more commitment and satisfaction than those who believed they existed to cut costs or exploit employees. Both studies are cross-sectional, so the direction of influence is open: satisfied employees may also read their employer's motives more charitably.

Suppose a grievance policy promises a hearing within ten working days, and in one division complaints routinely wait three months. Four parties can be held responsible, and each answer is partly right.

• HR, because it owns the policy and designed the process. But HR rarely controls the managers who must act, and holding it to account for compliance it cannot command tends to produce policing rather than change.
• The line manager, because the manager is the implementer. John Purcell and Sue Hutchinson (2007), surveying employees in 12 companies chosen as "excellent," found that both the quality of front-line leadership behaviour and satisfaction with HR practices had "a strong effect on employee attitudes," and noted that these managers "are increasingly charged with the implementation of many HR practices." A manager, however, can only answer for what he or she has the time, skill, and authority to do.
• Senior leadership, because it sets the targets, staffing levels, and incentives that make a policy easy or costly to follow. A supervisor who refuses leave because the line is understaffed is responding to a staffing decision taken above.
• A process owner, where one exists: a named person answerable for the whole route from policy to outcome.

The research reviewed above suggests that the gap is a property of the system more than the fault of any one party. David Bowen and Cheri Ostroff (2004) argued that HR practices act as messages, and that a "strong" HR system is one in which "individuals share a common interpretation of what behaviors are expected and rewarded." They grouped nine features under three headings: distinctiveness (visibility, understandability, legitimacy of authority, relevance), consistency (instrumentality, validity, consistent HRM messages), and consensus (agreement among the principal HR decision-makers, and fairness). A grievance policy that senior managers visibly ignore fails on consistency and consensus whatever its wording. The design implication is to assign accountability across the whole chain: HR for processes managers can realistically follow and for measuring practice rather than paper; line managers for enactment, once they have the means; senior leaders for the conditions; and one owner for each end-to-end process, the assignment most often missing and the one the others depend on.

Two objections deserve weight. The first is that the "strength" idea is better developed in theory than in evidence. A review by Helen Hewett and colleagues (2018) concluded that it "has not been comprehensively tested as it was intended," that the two main questionnaires failed to support some of the proposed features, and that empirical research "has been relatively limited." It is best used as a checklist for finding inconsistency, not as an established law. The second objection is that not every gap between policy and practice is a failure. Wright and Nishii treat variation among implementers as normal, and some of it is judgment applied to cases the policy did not foresee. A supervisor who grants unpaid leave outside the rules to a worker whose parent is dying may be closer to the organization's intentions than the policy is. The design question is which variation to remove (inconsistency that is unfair or unlawful) and which to permit (discretion within stated limits). That is a question of decision rights.

Every HR function must decide which people decisions are made centrally and which are left to units and line managers. Centralization buys consistency, legal control, economies of scale, and comparability, so the same job is paid on the same basis across sites. Devolution buys speed, local knowledge, and ownership: a decision made by the person who must live with it is more likely to be carried out. The costs mirror the benefits. Centralized systems are slow and blind to local conditions; devolved ones drift into inconsistency that employees experience as unfairness. Ulrich's principle that HR structure should follow business structure answers the question for the function as a whole. Within any structure, it still has to be answered decision by decision.

The splits are this article's illustration, not research findings. The right allocation depends on the legal environment, the organization's size, and how similar its units are.

A common and costly mistake in HR design is to borrow the structure of a much larger organization. The three-pillar model presupposes scale, and most workplaces do not have it. In Britain's 2011 Workplace Employment Relations Study, "Most workplaces with 500 or more employees have someone with an HR title (85%), but this is the case for 9% of those with 5 to 9 employees," and in 78 percent of workplaces employment relations remained "part of the wider duties of the owner or a general manager" (van Wanrooy et al., 2014). In small organizations HR is a set of tasks done by managers, often informally, with an outside accountant or lawyer covering payroll and compliance. At that size this is not a defective HR function but the appropriate one.

What matters early is less the structure than the employment model founders carry in their heads. The Stanford Project on Emerging Companies followed nearly 200 Silicon Valley technology start-ups from the mid-1990s and classified founders' "blueprints" by three questions: why employees are attached to the firm, how they are selected, and how their work is coordinated and controlled (Baron & Hannan, 2002). Five types emerged, which the researchers called star, engineering, commitment, bureaucracy, and autocracy. Blueprints shaped when HR arrived: "Commitment and Star firms tended to be the fastest to bring in HR expertise," while "In young Autocracy companies, often the 'HR' function is handled by the boss' secretary, who processes payroll." They also shaped later administrative weight. The researchers' model predicted 27.4 managers and administrators per 50 other employees in firms with bureaucracy-model founders, against 10.3 for commitment-model founders.

The most striking findings concern change. Changing the blueprint "significantly raises turnover, especially among the employees who have been with the enterprise the longest," and firms that changed it "were roughly 2.3 times as likely to fail subsequently as similar firms that had retained a stable blueprint." This is a real dilemma for a growing company: the informal model that suited 40 people may not suit 400, yet changing it is risky. The evidence needs careful reading. It comes from technology start-ups in one region and era; founders chose their blueprints rather than being assigned them; and a struggling firm may change its blueprint because it is already in trouble, which would inflate the apparent effect. The findings support treating redesign as costly, to be done deliberately and rarely, not as something to avoid.

The stages are this article's synthesis. The WERS and Stanford evidence shows how HR changes with size and founding model; it does not establish headcount thresholds.

The legal provisions and research findings below are real and dated. The company that follows them is fictional.

Bangladeshi employment law makes the distinction between policy and practice unusually concrete. Under section 3 of the Bangladesh Labour Act 2006, an establishment "may have its own service rules regulating employment of workers, but no such rules shall be less favourable to any worker than the provisions of this Chapter." The employer must submit the rules to the Inspector General of the Department of Inspection for Factories and Establishments (DIFE), who has 90 days to decide, and "No service rules mentioned in sub-section (2) shall be effective except with the approval of the Inspector General" (sections 3(2) and 3(3), unofficial English text as amended to 2018; government-owned establishments are exempt from the approval step). A company's core HR policy must therefore pass an external check of its content. Nothing comparable checks how the rules are practised.

The enforcement apparatus is thin relative to the workforce: according to the US Department of Labor, "In 2024, 442 labor inspectors conducted 34,008 worksite inspections," and it recommended raising the number to 1,861 for a labour force of about 74.5 million. Research inside workplaces shows how a legally required structure can exist and still be little used. Section 205 requires participation committees in establishments ordinarily employing at least 50 workers (Article 14 sets out their legal functions). A Better Work study of two garment factories (Afros, 2022), surveying 76 and 173 workers, found that nearly three-quarters viewed the committee positively, yet "only 4% of the workers in Factory A, and 8% in Factory B had directly taken any complaints to committee members." The committees' role had "narrowed down to communicating workers' grievances to the management," and the author identified "loopholes in the legal framework" because elections and proceedings take place "during working hours and under management control." Two factories cannot represent the country, but the pattern follows Wright and Nishii's chain: the structure is intended, formally present, and well regarded, while the practice is narrower than the design.

The Act has since been substantially amended, first by an ordinance approved in November 2025 and then by the Bangladesh Labour (Amendment) Bill, 2026, which Parliament passed on 9 April 2026 (The Business Standard, 9 April 2026). The reports consulted for this article do not indicate whether sections 3 or 205 were changed; check the gazetted text before relying on either.

What the exercise is designed to show. The visible symptom (inconsistent hubs) invites a structural answer, a new HR organization or a longer rulebook. The gap here lies in practice, decision rights, and the incentives line managers face, and those are the elements a response has to reach.

Part A asks how much differentiated HR structure an organization's situation justifies. Part B asks how far its policies are practised. Score them separately. Both are this article's teaching tools, drawn from the research discussed above; they have not been tested for reliability or validity.

Reading the results. A high Part A score with a low Part B score warns against reorganizing: the organization may be large enough for differentiated structure, but new boxes will not close a gap in practice. Low scores on both usually point to missing basics, such as written decision rights and trained managers, rather than missing HR staff. Following Wright and Nishii, Part B is most informative when HR staff, line managers, and employees complete it separately.

HR architecture is usually discussed as a question of boxes, and the evidence reviewed here points elsewhere. The documented problems of the three-pillar model arise at the joins between units; the documented gaps between policy and practice arise in the hands of line managers who never appear on the HR structure chart; and the best longitudinal evidence on growing firms suggests that changing the employment model is itself a risk. An effective HR function, on this reading, is one in which every commitment has a process, every process has an owner, and that owner can see practice rather than paper.

What remains unresolved is how far line managers can fairly be held accountable for HR outcomes when their own targets, staffing, and promotion depend on other things. Organizations often assign the accountability without the means, and the studies reviewed here do not tell us how much enactment improves when the means are supplied. That question can be tested inside a single organization, by comparing units before and after managers are given time, training, and authority. Article 15's discussion of causal evidence explains what such a test requires.

Select an organization with at least 100 employees that you know well (you may disguise it) or one whose HR arrangements are publicly documented. Prepare a report of 2,500 to 3,000 words that:

1. maps its HR architecture across decision rights, delivery roles (including line managers' HR responsibilities), policies, processes, and accountability;
2. assesses whether the HR structure fits the business structure and the organization's size and stage, drawing on Ulrich, Younger and Brockbank (2008) and Baron and Hannan (2002);
3. selects two policies and traces the intended, actual, and perceived practice of each, using evidence from at least two levels, such as HR staff and employees;
4. identifies who is accountable for each gap you find and whether they have the time, skill, and authority to close it; and
5. recommends changes, with their costs, the risks of redesign, and the evidence that would show whether they worked.

Use at least five academic sources, including at least three from this article's reference list. Distinguish clearly between what you observed, what you were told, and what you infer. Where you could not obtain evidence about practice, say so and explain how that limits your conclusions.

Assessment rubric (100 marks)

1. Baron, J. N., & Hannan, M. T. (2002). Organizational blueprints for success in high-tech start-ups: Lessons from the Stanford Project on Emerging Companies. California Management Review, 44(3), 8–36. https://doi.org/10.2307/41166130
2. Boglind, L., Hällstén, F., & Thilander, P. (2011). HR transformation and shared services: Adoption and adaptation in Swedish organisations. Personnel Review, 40(5), 570–588. https://doi.org/10.1108/00483481111154441
3. Bowen, D. E., & Ostroff, C. (2004). Understanding HRM–firm performance linkages: The role of the "strength" of the HRM system. Academy of Management Review, 29(2), 203–221. https://doi.org/10.5465/amr.2004.12736076
4. Cooke, F. L. (2006). Modeling an HR shared services center: Experience of an MNC in the United Kingdom. Human Resource Management, 45(2), 211–227. https://doi.org/10.1002/hrm.20105
5. Hewett, R., Shantz, A., Mundy, J., & Alfes, K. (2018). Attribution theories in human resource management research: A review and research agenda. International Journal of Human Resource Management, 29(1), 87–126. https://doi.org/10.1080/09585192.2017.1380062
6. Khilji, S. E., & Wang, X. (2006). 'Intended' and 'implemented' HRM: The missing linchpin in strategic human resource management research. International Journal of Human Resource Management, 17(7), 1171–1189. https://doi.org/10.1080/09585190600756384
7. Nishii, L. H., Lepak, D. P., & Schneider, B. (2008). Employee attributions of the "why" of HR practices: Their effects on employee attitudes and behaviors, and customer satisfaction. Personnel Psychology, 61(3), 503–545. https://doi.org/10.1111/j.1744-6570.2008.00121.x
8. Purcell, J., & Hutchinson, S. (2007). Front-line managers as agents in the HRM-performance causal chain: Theory, analysis and evidence. Human Resource Management Journal, 17(1), 3–20. https://doi.org/10.1111/j.1748-8583.2007.00022.x
9. Ulrich, D., Younger, J., & Brockbank, W. (2008). The twenty-first-century HR organization. Human Resource Management, 47(4), 829–850. https://doi.org/10.1002/hrm.20247
10. Wright, P. M., & Nishii, L. H. (2013). Strategic HRM and organizational behaviour: Integrating multiple levels of analysis. In J. Paauwe, D. E. Guest, & P. M. Wright (Eds.), HRM and performance: Achievements and challenges (pp. 97–110). Wiley. Quotations are from the working-paper version: CAHRS Working Paper 07-03, Cornell University (2007). https://ecommons.cornell.edu/server/api/core/bitstreams/8318d2bf-0b5a-4685-9e2d-bc3cf9703964/content

Data, legal, and press sources

• Afros, A. (2022). Workers' voice within and beyond participation committees: A case study of two Bangladeshi garment factories (Better Work Discussion Paper 46). ILO and IFC. https://www.betterwork.org/wp-content/uploads/Discussion-Paper-46-Workers-Voice-within-and-beyond-Participation-Committees.pdf
• Bangladesh Labour Act, 2006 (Act No. XLII of 2006), consolidated unofficial English text up to 2018, published by the Metropolitan Chamber of Commerce and Industry, Dhaka. https://mccibd.org/wp-content/uploads/2021/09/Bangladesh-Labour-Act-2006_English-Upto-2018.pdf
• "Parliament passes labour bill; layoff compensation, trade union rights among key reforms." The Business Standard, 9 April 2026. https://www.tbsnews.net/bangladesh/parliament-passes-labour-bill-layoff-compensation-trade-union-rights-among-key-reforms
• US Department of Labor, Bureau of International Labor Affairs. (2025). 2024 findings on the worst forms of child labor: Bangladesh. https://www.dol.gov/sites/dolgov/files/ILAB/child_labor_reports/tda2024/Bangladesh.pdf
• van Wanrooy, B., Bewley, H., Bryson, A., Forth, J., Freeth, S., Stokes, L., & Wood, S. (2014). The 2011 Workplace Employment Relations Study: First findings (4th ed.). Department for Business, Innovation and Skills. https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/336651/bis-14-1008-WERS-first-findings-report-fourth-edition-july-2014.pdf
• Jeffery, R. (2018, June 22). HR must be more outward looking, says Dave Ulrich. People Management. https://www.peoplemanagement.co.uk/article/1746990/hr-must-more-outward-looking-dave-ulrich-love-hr-conference

The CIPD survey (Reilly, Tamkin and Broughton, 2007, The changing HR function) is cited as reported in Ulrich, Younger and Brockbank (2008); it was not consulted directly. Boglind et al. (2011), Cooke (2006), Khilji and Wang (2006), and Purcell and Hutchinson (2007) were consulted at abstract level, and claims drawn from them are limited to what their abstracts state; Khilji and Wang's Pakistani banking setting is taken from secondary accounts. Bowen and Ostroff's nine features were checked against the summary in Hewett et al. (2018). Nishii et al. (2008) was read in its CAHRS working-paper version. The five-element framework, comparison tables, stage table, Megh Courier case, decision options and commentary, and diagnostics are original teaching material for this series.

Next in The Human Capital MBA series: Article 14, "The HR Manager as a Business Partner," develops the business partner role introduced here.