TL;DR
- The decision: Pick the least specialised structure that still lets the work get done without one person becoming the bottleneck for everything.
- When not acting is correct: When your team can name every employee's first child and still ship the work, the operating model isn't the problem.
- What must be true: A model only earns its complexity once the demand on the function has a shape it can't absorb without specialisation.
- How the options split: Five structures sit between fully generalist and fully external. They differ on who owns the specialist work and where the judgment sits.
- Decision rule: Size the model against the demand drivers you actually have, not against an HR-to-employee ratio.
- Expected outcome: The right fit reduces escalations to the CHRO and returns generalists to the work that actually needs a generalist.
A Head of People opens the quarterly plan at a two-hundred-person software company and finds the same line repeated under three objectives: "HR to lead". Three initiatives, one function, and a team of seven. Two are generalist HR managers carrying full business partner caseloads, one runs a recruiting pipeline that's the busiest part of the function, one owns L&D, and the remainder are shared services for payroll, benefits and the first line on every employee question. There's no centre of expertise. There's no separate talent acquisition lead. There's no people analytics role. Everyone does everything and no one owns anything end to end. The CHRO is being told to "professionalise the function", which in practice means copy the structure of a company ten times their size and hope the shape works. It won't. It will cost two years, two good people, and the trust of a leadership team that wanted help, not a transformation.
But this is the decision most teams get wrong. The instinct is to choose the most sophisticated structure you can describe. The real question is the smallest structure that survives the demand you already have, and most teams have less demand than they think. Specialists are a response to volume, complexity and risk. When the volume is missing, specialists become expensive generalists who own nothing, sit in meetings, and produce strategy decks nobody asked for. The structure on the org chart is rarely the bottleneck. The bottleneck is whether the team can absorb a pay cycle, a restructure, a hiring spike and a difficult manager conversation in the same fortnight without someone dropping a ball.
So the model you choose has to match the company you actually are. Not the company on the cover of a business magazine, and not the company the board would prefer to belong to. The one whose pay runs on the fifteenth, whose managers have already had a hard quarter, and whose HR team is six people who already know each other's coffee order.
Best tools for HR Strategy
When You Genuinely Do Not Need to Act Yet
Most HR functions consider an operating model change too often, and complete one too rarely. The signal that you don't need a new model isn't calm. It's throughput. A function whose work lands and whose escalations to the CHRO are infrequent doesn't have a structural problem. It has a workload problem, which is solved by people, not structure.
The team of four where everyone knows the names
A team of four at a ninety-person agency covers recruiting, payroll, ER, L&D, benefits and the CEO's inbox. Everyone covers everything. There's one HR system, owned by whoever set it up. The volume is high but bounded, the manager population is small enough to coach directly, and the difficult cases are few enough that the team handles them in the weekly standup. This team doesn't need a new model. It needs one more generalist and a clearer decision right between manager and function. Adopting a three-legged structure here would split the team into three groups of one and one group of two, each with a manager, and the cost of the new reporting lines would consume the bandwidth it was meant to free. The friction is real but the structure is fine, and changing it now is a vanity project.
The team of twelve where the friction is between two people
A twelve-person team at a four-hundred-person fintech has a real friction, but it sits between talent acquisition and the people operations lead. Recruiters miss the brief on senior hires, the ops lead rebuilds hiring plans every quarter, and the rest of the function is stable. The friction is structural, but the fix is a tighter handoff between two adjacent teams, not a new operating model. Rewiring the whole function to fix a misalignment between TA and ops is the change the board will fund and the change nobody will thank you for, because the original problem was an interface and you re-architected the whole house.
The team where the risk is in payroll and the structure everywhere else
A team of eighteen at a seven-hundred-person retailer has a payroll function that's two people on a system that has not been upgraded in years, run on a schedule that breaks twice a year, and surrounded by manual workarounds that nobody has written down. The rest of the function is fine. The generalists are competent, the business partners are credible, ER is steady. The model isn't the problem here. The model is the wrong place to look. The structural answer to a payroll problem is a payroll system, a runbook, and two named owners. Reorganising the function to fix payroll produces a new org chart and the same broken pay run.
The edge case where the team looks calm because the work is being suppressed
A team of nine at a three-hundred-person services firm reports low escalations and high satisfaction in the engagement survey, but the manager population has stopped raising issues because the last three they raised took three months to resolve and they have given up. The team looks fine because it has trained the company to stop asking. This is the case where doing nothing is wrong, and the reason it's wrong isn't visible on any current metric. The signal is the absence of escalations in a business that should be generating them.
The point of running through these four isn't to teach you what a model is. It's to show that the decision to keep the current model is a real decision, made for real reasons, and not a failure of ambition. The wrong move is to reorganise when the work is fine and to wait when the work has quietly stopped arriving.
Five Questions You Ask Yourself at 11pm
The decisions that matter are rarely the ones in the board pack. They're the ones made late, after the third hard conversation of the week, when an HR leader is staring at the ceiling and trying to work out whether the structure is the problem or they're.
Are we paying for specialists who are acting as generalists?
You have a Head of L&D. They spend half their week on ER cases. You have a talent acquisition lead. They're in the manager onboarding review every cycle. You have a people analytics partner. They're producing one dashboard a quarter. If the specialists on your team are spending their time on work that doesn't require their specialty, the model isn't wrong because you lack specialists. It's wrong because you've more specialist headcount than the work can absorb, and the specialists are doing generalist work because that's what falls on them. The fix is rarely a reorganisation. It's usually fewer specialist roles with a tighter scope and the generalist work pushed back to the generalist team.
Does one person hold the entire function together?
If the CHRO leaves and the function can't describe what it does in writing, the model is the person. This is the single most common structural failure in mid-sized HR functions and it's invisible from the org chart, because the org chart shows the structure and the structure looks fine. What is missing is the documentation of decisions, the runbooks, the escalation paths, and the small pieces of judgment that sit in one person's head. If your function can't operate for two months without its leader, the model isn't the issue. The issue is the absence of artefacts that let the model survive a transition. Get the artefacts right, and the structure becomes visible. Get the structure right without the artefacts, and the next leader starts from scratch.
Are we rebuilding work the centre should have done?
You write the same manager onboarding checklist at every business unit because the centre has not. You build your own ER playbooks because the centre has not. You negotiate your own benefits renewals because the centre has not. The work is being duplicated across business HR, and the cost isn't visible because it's split across multiple budgets. The model here may be fine. The centre is the problem. Either the centre needs to do its job or the centre needs to be dissolved and the work pushed out to the business HR teams. A centre that exists on paper but not in practice is more expensive than no centre at all.
Do the generalists have enough time to be generalists?
A generalist who spends their week chasing payroll queries isn't a generalist. They're a tier-one service desk with a job title. The signal that the model is wrong is the generalist whose calendar is full of work that anyone could do and empty of work that only they can do. If your business partners can't find time to coach managers, sit in leadership team meetings, or work on the talent review, the work that only a generalist can do isn't getting done. The model is wrong in a specific way. The shared services layer is either missing, understaffed, or designed as a routing function rather than a resolution function.
Would a new manager in this team know what to do on day three?
If a new HR manager joins and can't tell you what their first thirty days look like, what decisions they own, what they escalate, and what the escalation path looks like, the model has not been written down. The model may be right. It isn't communicable. The fix isn't a reorganisation. The fix is a one-page operating model that names the layers, names the decisions each layer owns, and names the hand-offs between them. Most operating model changes fail not because the design is wrong but because the new manager on day three can't tell you how the function works.
Three Honest Categories the Approaches Split Into
The five models you can adopt aren't five separate things. They're three categories, with the five sitting inside them. Understanding the categories first means you understand the trade you're making before you look at the menu.
The single layer: everyone does everything
A team of any size where the work isn't divided by specialty. Everyone carries a caseload, everyone covers payroll, ER, recruiting and the inbox. The team is sized for the volume and the volume is the constraint. This category works when the volume is bounded, the manager population is small enough to know personally, and the risk profile is low enough that the generalist can hold it. It fails when the volume grows past what generalists can absorb without dropping specialist work. The failure is silent. Recruiting slips, ER escalations rise, pay runs get manual, and the team compensates by working longer hours rather than by changing shape. The single layer is the cheapest model to run and the most expensive to outgrow, because the team you've when you finally notice is exhausted.
The layered structure: specialisation by function
A team that's divided into business-facing generalists and a centre that owns specialist work. The centre might have an L&D lead, a talent acquisition lead, a people analytics partner, and a shared services desk. The generalists sit with the business and carry caseloads. The centre builds, maintains, and consults. This is the structure most mid-sized functions reach for, and it's the structure most teams adopt too early. It works when the centre has real, repeatable work that the generalists would otherwise duplicate. It fails when the centre is a title inflation exercise, when the work the centre owns could be done by one generalist with the right tooling, or when the hand-off between centre and generalist is so expensive that the business prefers to ask the generalist directly. The cost of the layered structure isn't the headcount. It's the hand-offs. Every hand-off is a delay, a miscommunication, and a context rebuild.
The distributed structure: the function is bought, not built
A model where some or all of the HR function is delivered by external providers, fractional leaders, or a combination of internal coordination and external execution. The internal team becomes small, focused on judgment, culture, and the manager interface, and the operational work is bought. This category works when the operational work is genuinely transactional, when the internal team is too small to justify specialist hires, and when the provider market is mature enough to deliver reliably. It fails when the operational work isn't actually transactional, when the in-house team lacks the judgment to manage the provider, or when the cost of the provider relationship exceeds the cost of the work being delivered. The failure mode here's the function that outsources its operations and discovers that it has also outsourced its institutional memory, and the next time the business asks for something new, nobody inside the building knows how to do it.
Five Diagnostic Questions to Self-Assess
The five questions below aren't for the board pack. They're for you, on a Tuesday afternoon, when you want to know whether the model is the problem.
What proportion of your team's week is on work only a generalist can do?
The way to answer this isn't a survey. It's two weeks of every team member logging their time in fifteen-minute blocks against four buckets: work only they can do, work a specialist should own, work a shared service should resolve, and work the manager should handle. If the largest bucket is the second or third, your model has the wrong shape. The team is doing work the structure should be routing elsewhere. If the largest bucket is the fourth, the team is doing work the manager should own and the manager is outsourcing their job to you. If the largest bucket is the first, the model is probably fine and the constraint is volume, which is solved by people.
What work happens more than once across the team?
The second diagnostic is duplication. If the same manager onboarding plan exists in three versions in three business units, the centre has not done its job or the centre doesn't exist. If the same ER investigation is built from scratch by three different generalists, the playbook is missing. If the same offer template is being rewritten by every recruiter, the tooling is missing. The model isn't the issue. The artefacts are. Build the artefact once at the centre, push it out to the business, and the duplication disappears.
What escalates to you that the model should have absorbed?
You're the CHRO. The work that lands on your desk is the work the model failed to absorb. Track it for a month. Categorise the escalations: specialist decisions the centre should have made, manager decisions the business should have made, and genuinely CHRO-level decisions only you should handle. If the largest category is the first two, the model is the wrong shape. If the largest category is the third, the model is fine and you're over-extended, which is a different problem and a different solution.
What would break if the strongest person on the team left tomorrow?
The strongest person on the team is usually the bottleneck, and the bottleneck is usually the proof that the model has not been written down. Name what would break. If the answer is "everything", the model is the person. If the answer is "one or two specific things", the model has the right shape and one specific role is under-documented. Fix the documentation, not the model.
What work has the team stopped doing because the model has not been updated?
The team used to do something the model doesn't currently absorb. Maybe it's manager development. Maybe it's workforce planning. Maybe it's the analytics behind the talent review. The work has fallen off because nobody owns it, and the model was designed when the work was not there. The fix is either to add the role to the model or to accept the work isn't getting done. Pretending it's happening is the worst of the three options.
Five Operating Models, Reviewed
A solo generalist
One HR professional, sometimes with a part-time assistant or a contracted payroll provider, carrying the full function for a company of somewhere between fifty and two hundred and fifty employees. The generalist handles recruiting, onboarding, payroll, benefits, ER, the manager interface and the executive interface, often with a provider handling payroll runs and benefits administration. There's no specialisation because the volume doesn't justify it, and the work that does require specialist judgment is escalated to external advisers on a case-by-case basis. This model earns its place because it's the smallest structure that lets the work happen. It costs less than a team, requires less coordination, and keeps the function close enough to the business that judgment doesn't need to travel through layers. The model falls short when the volume crosses a threshold the solo generalist can't absorb, and the threshold varies by industry, by manager quality, by the pace of hiring, and by the regulatory environment. The failure isn't dramatic. It's a quiet one: the pay run slips by a day, the recruiting pipeline dries up, an ER case takes three months to resolve, and the generalist is working evenings. The solo generalist is right when the company is genuinely small, the manager population is experienced enough to handle most people decisions without HR, and the operational work can be split between the generalist and a small number of reliable providers.
A pooled generalist team with no specialisation
A team of generalists, usually between three and eight people, covering the function between them without internal specialisation. Each generalist carries a caseload, the team handles operational work collectively, and there's no centre, no separate talent acquisition, no separate L&D, no separate ER lead. The team might be structured by business unit or by workstream, but the work each person does is broad rather than deep. This model earns its place when the volume justifies more than one generalist but the demand drivers don't justify any specialist hires. The pooling gives the team resilience, cover for absences, and the ability to scale slightly without adding headcount at the specialist level. It's the model most teams should reach for when they outgrow a solo generalist and is the one most teams skip on the way to something more complex. The model falls short when the team reaches a size where coordination starts to consume more time than the work requires, and when the workstream with the highest volume (usually recruiting or ER) starts to pull attention away from the rest of the function. The fix is rarely a reorganisation. It's usually one specialist hire in the workstream with the highest volume, paired with the existing generalist team.
A three-legged structure with business partners and centres of expertise and shared services
A structure that divides the function into three layers. Business partners sit with the business and carry the manager interface, the talent review, the leadership team relationship and the judgement-heavy work. Centres of expertise own the specialist workstreams: talent acquisition, L&D, total rewards, people analytics. Shared services owns the operational work: payroll, benefits administration, the HRIS, the case routing, the first line on every employee question. The structure is the most common ambition for mid-sized HR functions and the most commonly adopted in aspiration and abandoned in practice. This model earns its place when the function is large enough that each leg has at least three people, when the demand on each leg is high enough to justify a full team rather than a single specialist, and when the hand-offs between legs are clear enough to make the structure faster than the alternative. The model fails when any leg is staffed by a single person, because a leg of one is a bottleneck dressed as a structure. It also fails when the hand-offs are unclear, because every unclear hand-off becomes a meeting, a delay, and a context rebuild. The weakness most teams don't name is the cost of the centre. A centre of expertise that's not large enough to do its job becomes a title inflation exercise. Specialists with no team to lead, no clear deliverables, and no internal customers using their work end up in meetings. The three-legged structure is right when the function has the volume to fill each leg, the manager population to keep business partners busy, and the leadership discipline to keep the centre honest.
Embedded business partners with a deliberately thin centre
A structure where the function is mostly business partners, sitting deep inside the business, with a thin centre that owns only the work that genuinely can't be distributed: the HRIS, payroll, compliance, and a small number of shared policies. The specialists sit inside the business rather than in a central team. A recruiter embedded in the engineering business unit is hired by the engineering leadership and reports through the HR function for HR matters but is part of the engineering team for everything else. A people analytics partner sits inside the strategy team. The centre is small and focused on the operational backbone. This model earns its place when the business units are different enough that the centre can't serve them all, when the volume in each unit justifies an embedded role, and when the manager population is sophisticated enough to use a business partner rather than ask the centre. The model is common in companies that grew by acquisition and can't easily standardise across business units. It falls short when the business units have more in common than they admit, because the embedded model repeats work across units that could be done once. It also fails when the centre is so thin that it can't maintain the operational backbone, and the embedded business partners end up doing their own payroll queries because the shared services function is under-resourced.
A fractional or outsourced model where the function is bought rather than built
A structure where most of the function is delivered by external providers, fractional leaders, or a combination. The internal team is small, often a Head of People and one or two generalists, and the operational work is bought from a payroll provider, an benefits administrator, a recruiter on contract, an ER adviser on retainer, and an L&D partner for specific programmes. The internal team owns the judgment, the culture, the manager interface and the executive interface. This model earns its place when the internal team is too small to justify a full specialist stack, when the operational work is genuinely transactional, and when the provider market is mature enough to deliver reliably. It's the right model for a company that has decided HR isn't a core competency and is willing to pay for the work to be done well. It falls short when the work isn't actually transactional, when the in-house team lacks the judgment to manage the provider relationship, or when the cost of the providers exceeds the cost of the work. The failure mode is the function that outsources operations and loses the institutional memory that comes with doing the work. The next time the business needs something new, nobody inside the building knows how to do it.
The Decision Table
| Situation | Scale | Setup | Primary Pain | Recommended Starting Point |
|---|---|---|---|---|
| Solo founder or owner-led company, low hiring volume | 30 to 80 | A single HR generalist with payroll and benefits outsourced | The generalist is stretched across operational and judgement work | A solo generalist, with payroll on a provider and benefits administration externalised |
| Growth company, hiring is the bottleneck | 100 to 250 | A small generalist team with recruiting dominant | Hiring volume consumes the team's bandwidth | A pooled generalist team, with one recruiting specialist paired to the team |
| Multi-business-unit company, varied demand | 300 to 800 | A generalist team that has hit a coordination ceiling | Different business units need different support, the centre cannot serve them all | Embedded business partners with a thin centre, owning only the operational backbone |
| Single-business company, consistent demand across units | 500 to 1500 | A generalist team plus informal specialists whose scope is unclear | Specialists are doing generalist work, generalists are doing specialist work | A three-legged structure, only if each leg has at least three people; otherwise a pooled team with named workstream leads |
| Stable company, low growth, experienced managers | 200 to 500 | A generalist team with a strong operational backbone | The function is running fine but the leadership wants a "strategic" HR presence | Do not reorganise. Add a single senior HR leader to free the CHRO for executive work, keep the model as is |
| Acquired company, integration underway | 500 to 2000 | Multiple HR functions being merged | Different systems, different policies, different cultures, no shared operating model | A pooled generalist team as the integration target, with embedded leads in the acquired units for the first eighteen months |
| Regulated industry, high ER and compliance exposure | 300 to 1000 | A generalist team with high ER volume | ER cases consume the team's attention and slow everything else | A pooled generalist team, with one ER specialist and a clear escalation path to external counsel |
| Holding company or portfolio of small businesses | 50 to 400 per unit | Each unit has a small HR presence or none | Coordination across units is impossible, duplication is high | A fractional model at the centre, with a small in-house team owning coordination and the units buying HR as needed |
The Ratio Nobody Agrees On
The HR-to-employee ratio is the most quoted and least useful number in the operating model conversation. Functions with low ratios claim they're running lean. Functions with high ratios claim they're running strategic. Both are usually rationalising headcount they already have. The ratio is a lagging indicator of what a function was three years ago, not a leading indicator of what it should be tomorrow.
The reason the ratio is a poor guide to structure is that it treats all employees as equivalent. They're not. A company of two hundred software engineers isn't the same operating challenge as a company of two hundred retail associates, a hotel of two hundred housekeeping staff, or a hospital of two hundred nurses. The demand on HR varies by industry, by manager quality, by pace of change, by regulatory environment, by the number of jurisdictions, by the seniority of the workforce, and by the operational complexity of the underlying business. A single ratio can't capture any of that.
What you size against instead is the demand driver. The table below isn't a benchmark. It's a way of thinking about which demand drivers raise the cost of the function and which structural choices absorb them.
| Demand driver | What raises it | What lowers it | Which model absorbs it best |
|---|---|---|---|
| Hiring volume | High growth, low brand, niche roles, senior hiring | Stable headcount, strong employer brand, entry-level hiring | A pooled team with a dedicated recruiting specialist, or a three-legged structure with talent acquisition in the centre |
| ER and manager sophistication | Inexperienced managers, regulated industry, dispersed workforce | Strong manager training, low regulatory exposure, experienced managers | A pooled generalist team with one ER specialist, or external counsel on retainer |
| Operational complexity | Multiple jurisdictions, complex benefits, hourly workforce, frequent pay cycles | Single jurisdiction, simple benefits, salaried workforce, monthly pay | Shared services inside a three-legged structure, or a payroll provider with a thin internal team |
| Strategic work | Growth, change, M&A, leadership transitions | Stability, low change, mature leadership | Business partners embedded in the leadership team, supported by a thin centre |
| Manager development | Large manager population, rapid hiring of new managers, distributed teams | Small manager population, experienced managers, low turnover | A centre of expertise with one L&D specialist, or external L&D partners on a project basis |
| Workforce planning | Growth, contraction, acquisitions, restructures | Stable headcount, mature business | A three-legged structure with people analytics in the centre, or a fractional people analytics partner |
The model absorbs the demand driver that the function has the most of. A function dominated by hiring needs a recruiting specialist. A function dominated by operational complexity needs a shared services layer. A function dominated by change needs embedded business partners. A function with low volume across the board needs none of these.
Moving Between Models Without a Reorg
Most operating model changes are sold as transformations and delivered as reorganisations. The board approves a new structure, the CHRO redraws the org chart, the new roles are advertised, and eighteen months later the function looks the same as before because the work has not changed. The reorganisation was the visible part of the change. The underlying behaviour, who owns what, who escalates to whom, who builds the artefacts, didn't change. The new structure sits on top of the old habits and the old habits win.
The reason the centre of expertise is usually the wrong place to start is that the centre is the most expensive piece of the new structure. The specialists in the centre are the hardest to hire, the slowest to pay back, and the easiest to lose. If you build the centre first and the demand isn't there, the centre becomes a cost centre with no internal customers and no outputs. The specialists leave. The generalists stop asking the centre for help. The centre gets dissolved.
The change to make first is in the shared services layer, not the centre. Shared services is the cheapest piece of the new structure, the easiest to outsource if it doesn't work, and the layer that frees the generalists to do the work only they can do. When shared services is operating, the generalists have time to absorb specialist work. When the generalists have that work in their hands, the demand for the centre becomes visible. You build the centre once you can see the work, not before.
The incremental change works like this. Start by mapping what the generalists are spending their week on, in the four buckets from earlier. Identify the work that shared services should resolve. Build the smallest shared services layer that absorbs that work, whether that's one new hire, an outsourced provider, or a self-serve HRIS. Watch the generalists' week change. Identify the work that the generalists are now spending too much time on. That work is the candidate for the centre. Build the smallest specialist role that absorbs that work, whether that's a recruiting specialist, an L&D lead, or a people analytics partner. Watch the demand on that role. If the demand is real and repeatable, the role earns its place. If the demand fades, the role folds back into the generalist team.
The move to embedded business partners comes last, and only if the business units have become different enough that the centre can't serve them. Most companies never reach this stage. Most companies that adopt this stage too early end up rebuilding the centre they dissolved, because the duplication across embedded business partners exceeds the cost of the centre they gave up.
What to Put in Writing
The artefacts below are what turn a good decision into a defensible one. They're also what most teams skip, because the artefacts feel like documentation rather than delivery. They're delivery. The team that has these artefacts can survive a leadership change. The team that doesn't can't.
| Artefact | Who owns it | When it is written | What it prevents |
|---|---|---|---|
| One-page operating model | The CHRO or Head of People | At every model change and refreshed annually | The next leader rebuilding the function from scratch |
| Decision rights matrix | The CHRO, signed off by the leadership team | At every model change and refreshed annually | The escalation that ends with the CHRO making a decision the manager should own |
| Hand-off map between business HR and the centre | The centre lead, in partnership with the business partner lead | When the three-legged structure is adopted | The duplication and the dropped ball at the boundary |
| Manager onboarding curriculum | The L&D lead or generalist team | Refreshed every six months | The manager who does not know what they own and who to ask |
| ER playbook | The ER lead or generalist team | Refreshed annually and after every significant case | The investigation that takes twice as long as it should and produces a worse outcome |
| Recruiting scorecard by role | The talent acquisition lead | Refreshed at every hiring plan cycle | The hire that fails because nobody agreed what success looked like |
| Pay cycle runbook | The payroll lead or shared services lead | Refreshed before every pay cycle | The pay run that breaks because the runbook is in someone's head |
| People analytics dashboard | The people analytics lead | Refreshed monthly | The talent review that runs on anecdote |
| Annual people strategy | The CHRO or Head of People | Refreshed annually | The year of reactive decisions because there was no plan |
| Stakeholder map of HR's internal customers | The business partner lead | Refreshed at every model change | The generalist who is surprised by a stakeholder relationship they did not know existed |
The artefacts are the difference between a function that runs on the people in it and a function that runs on the design of it. The first is fragile. The second survives the people.
Questions to Ask Before You Commit
These questions are written to be copied into an email and sent to a provider, an adviser, or your own team. A bad answer to any one of them is a signal to slow down.
Demand proof. What specific work will this specialist role own, and what work will it take away from the generalist team? A bad answer is vague: "strategic projects", "leadership development", "culture initiatives". These aren't jobs. They're categories. A good answer names the workstream, the volume, the deliverable, and the internal customer.
Hand-off clarity. Where does this role's work begin and end, and who is on either side of the hand-off? A bad answer is "we'll figure it out". A good answer names the hand-off points, the format of the hand-off, and the escalation path when the hand-off breaks.
Demand volume. What is the volume of work this role will absorb, and what happens to that work if the demand is lower than expected? A bad answer is "we'll find other things for them to do". A good answer names the volume threshold below which the role folds back into the generalist team.
Centre discipline. If we build a centre, what is the centre's product, who are its internal customers, and what evidence will we've that they're using it? A bad answer is "the centre exists to support the business". A good answer names the deliverable, the internal customer, and the usage evidence.
Specialist cost. What is the fully-loaded cost of this specialist, and what is the work they replace or absorb that has its own cost today? A bad answer is "the cost is justified by the strategic value". A good answer names the cost of the specialist, the cost of the work they replace, and the net change.
Outsourcing option. What would it cost to buy this from a provider instead of building it in-house, and what would we lose? A bad answer is "we want to keep it in-house". A good answer names the cost of the provider option and the specific work that would be lost.
Failure signal. What would tell us, six months in, that this change has not worked? A bad answer is "we'll review it". A good answer names the specific metric, the threshold, and the action you would take if the threshold is missed.
Documentation. What artefacts will this role own, and where will they live? A bad answer is "we'll document as we go". A good answer names the artefacts, the owner, and the location.
The Cost of Getting This Wrong
The cost of a bad operating model decision is rarely the new headcount. The new headcount shows up on the org chart and the budget. The cost that doesn't show up is the opportunity cost of the work the team stopped doing while it was reorganising. The talent review that didn't happen. The manager development programme that fell off. The ER cases that took twice as long. The pay run that broke because the runbook was in transition. The hiring that slowed down because the recruiting specialist was being hired rather than recruiting. None of these are visible as a cost. All of them are real.
The second cost is the people who left during the transition. Every reorganisation creates uncertainty, and uncertainty creates attrition. The generalists who were not sure their role survived, the specialists who were hired into a structure that was being rebuilt around them, the business partners who could see the model was not right but were not consulted. The people who leave during a reorganisation are often the people you can least afford to lose, because they're the ones with the institutional memory the new structure has not documented.
So the decision isn't which model is best. The decision is which model is the smallest one that lets the work happen, and which one you can build without losing the team that does the work. The team is the function. The model is the container. Get the container right and the team can absorb the work. Get the container wrong and the team spends two years rebuilding what they already had.
When You Are Ready to Go Further
If the questions in this article have surfaced a real decision in your function, the next step is to see how your peers have made it. HROpsLab maintains an independent library of how real HR functions have structured their operating models, what they chose, what they changed, and what they would do differently. We don't sell software, payroll, or advice. We publish what works and what doesn't, drawn from the teams who lived through it.
The library is free to browse. If you want to compare your structure against how other functions of a similar size and shape have organised themselves, the place to start is the operating model case studies.
Independent reviews for HR leaders. HROpsLab is a review publication. We sell nothing.
[Two-CTA block: Case Studies | Talk to Expert]
[Related blogs: 6-9 handpicked related posts]
Frequently Asked Questions
What is an HR operating model?
An HR operating model is the structure the HR function uses to divide and deliver its work: who owns what, who sits where, who handles which kind of decision. It isn't the org chart. The org chart is one output of the model. The model itself is the decision rights, the hand-offs, the work split between generalist and specialist, and the artefacts that document all of it. Two functions with the same org chart can have very different operating models, and two functions with very different org charts can have the same operating model.
When is a team too small to specialise?
A team is too small to specialise when the specialist role would be staffed by one person and the work wouldn't justify a full-time specialist. The threshold isn't a headcount. It's a volume question. If the workstream the role would own (recruiting, L&D, ER, analytics) generates enough repeat demand to keep a specialist fully occupied, the role earns its place. If the demand is part-time, the work is better absorbed by a generalist with the right tooling or by an external provider on a project basis.
What does an HR business partner actually do?
An HR business partner sits with the leadership team of a business unit and owns the people decisions that require judgement: restructures, senior hiring, talent reviews, leadership development, complex ER. They don't own payroll, benefits administration, or the HRIS. They do own the interface between the business and the centre of expertise. The work that makes the role valuable is the judgement-heavy work that the manager can't do alone and that the centre can't do without context.
Are centres of expertise worth it below a thousand employees?
Centres of expertise can be worth it well below a thousand employees, but only if the centre has at least three specialists and clear internal customers. A centre of one isn't a centre. It's a specialist with no team to lead, no peers to consult, and no clear deliverables. The centre earns its place when the demand on the workstream is repeatable, when the generalist team needs the centre's output, and when the cost of the centre is justified by the work it absorbs from the generalists. Below a thousand employees, this is rarely true for more than one or two workstreams.
How do you know the current model has stopped working?
The model has stopped working when the signals that the structure is failing become routine: escalations to the CHRO that the structure should have absorbed, work the team has stopped doing, duplication across the team that the structure should have routed, hand-offs that break, or specialist work being done by generalists because the specialist role doesn't exist. One signal is noise. A pattern of signals is a model that has stopped fitting.
Should you outsource part of the HR function?
Outsourcing is right when the work is genuinely transactional, when the in-house team is too small to justify specialist hires, and when the provider market is mature enough to deliver reliably. It's wrong when the work requires deep institutional knowledge, when the in-house team lacks the judgement to manage the provider, or when the cost of the provider exceeds the cost of doing the work in-house. The most common failure is outsourcing operational work and discovering the in-house team has also lost the institutional memory that came with doing the work.
How long does a model change take?
A small model change (one new specialist hire, a new shared services layer) takes six to nine months from decision to operation, because the hire takes three months and the integration takes another three to six. A larger model change (a three-legged structure replacing a pooled team) takes eighteen to thirty-six months, because the centre takes a year to build, the hand-offs take another year to stabilise, and the business takes a further year to use the centre properly. The mistake is treating it as a twelve-month transformation. It isn't. It's a multi-year change in how the function works.
What do you do when the model is right but the people are in the wrong seats?
The fix is role redesign, not model redesign. The model is the container. The seats are who sits in it. If the work is in the right place and the people aren't, the answer is to move the people, redefine the roles, or in some cases replace the people. The model change is the wrong tool for a people problem. Diagnose the people problem separately and act on it without redrawing the structure.
Independent reviews for HR leaders.
