Ask most people what their finance, HR or IT department is, and they picture one team doing everything. Ask anyone who has redesigned one, and they will tell you the opposite: a well-run support function is three different jobs wearing one badge. The art is deciding which work goes where, and which work shouldn't be done in-house at all.

The quick version

  • Shared services pool the routine, high-volume work from across the organisation (payroll, expenses, IT tickets) into one internal unit, so it is done once, cheaply, and consistently instead of duplicated in every team.
  • A centre of excellence (CoE) concentrates scarce, deep expertise (reward design, cyber-security, data science) in one place so specialists set standards and solve hard problems, rather than being thinly spread.
  • Functional outsourcing hands a slice of work to an external provider, a make-or-buy decision driven by cost, scale and whether the work is core to who you are.
  • The common trap: copying the structure without the operating model. The boxes are easy; the handoffs, the self-service and the "who owns this?" rules are where it lives or dies.

The idea in depth: one function, three jobs

The clearest articulation of this split comes from HR, and it has quietly shaped how almost every large support function is now built. In Human Resource Champions (1997), David Ulrich proposed what became known as the "three-legged stool": a function divided into shared services (the transactional engine), centres of excellence (the deep specialists), and business partners (senior advisers embedded with the line, translating strategy into people decisions). The model spread far beyond HR, finance, IT, legal and procurement all borrowed the shape, because it answers a real problem: different kinds of work reward completely different management. You can read a clear summary of the three legs in AIHR's guide to the model.

So the move is to stop asking "how do we make finance better?" and start asking "which of these three things are we actually trying to fix?" Slow expense reimbursements are a shared-services problem, fix them with standardisation and self-service, not by hiring a strategist. A botched acquisition integration is a centre-of-excellence problem, you needed scarce M&A expertise, and it wasn't concentrated anywhere. Vague, reactive support to the executive team is a business-partner problem. Naming the leg tells you the remedy.

flowchart TD
  L(["The business / employees"]) --> S(["Shared services
routine, high-volume work
done once, cheaply"])
  L --> P(["Business partners
embedded advisers
strategy → decisions"])
  P --> C(["Centres of excellence
scarce deep expertise
standards + hard problems"])
  S -.escalate hard cases.-> C
  C -.set standards + tools.-> S
Ulrich's "three-legged stool", generalised to any support function, routine work, embedded advice, and concentrated expertise, with traffic flowing between them. Leaders Loop

Shared services is the leg with the hardest economic logic, and the one most often confused with simple cost-cutting. Bryan Bergeron, in Essentials of Shared Services (Wiley, 2002), defines it as concentrating existing functions into a semi-autonomous business unit that serves the rest of the organisation as if it were a supplier competing in the open market, a form of "internal outsourcing." That last phrase is the point. A shared-services centre that simply centralises the same messy processes saves little; one that treats internal teams as customers, publishes service levels, and is measured on cost and satisfaction behaves like a business and improves like one.

So the move is to give your shared-services unit a real service contract, not just a reporting line. Define what "good" looks like (a payslip query answered in 24 hours; a new starter set up before day one), measure it, and let the internal customers see the numbers. Without that, centralisation just moves the bottleneck somewhere less visible.

When to build it, and when to buy it instead

Functional outsourcing is the same question pushed one step further: if shared services is doing work once, internally, why not have someone outside do it instead? The cleanest way to reason about this is older than any HR model. Economist Ronald Coase asked in 1937 why firms exist at all rather than buying everything on the open market; Oliver Williamson later built that into transaction cost economics, the idea that you keep work inside the firm when the cost and risk of contracting for it outside (writing the contract, policing it, being held hostage by a supplier you depend on) exceeds the cost of just doing it yourself. Williamson's own Nobel lecture lays out the logic of markets versus hierarchies in plain terms.

The practical translation: outsource work that is standardised, well-defined and easy to specify in a contract, payroll runs, data-centre hosting, first-line IT support. Keep work that is bound up with your competitive edge, hard to specify, or dangerous to depend on a single supplier for. So the move, before any outsourcing decision, is to ask three blunt questions: Is this core to why customers choose us? Can we write a contract that actually captures what good looks like? And if the provider failed or held us to ransom, how exposed would we be? "No, yes, not very" is a candidate to outsource. Any other answer, think hard.

flowchart TD
  A(["A slice of functional work"]) --> B{"Core to our
competitive edge?"}
  B -->|"Yes"| K(["Keep in-house
(often a centre of excellence)"])
  B -->|"No"| C{"Easy to specify
in a contract?"}
  C -->|"No"| K
  C -->|"Yes"| D{"Acceptable supplier
dependency / switching risk?"}
  D -->|"No"| S(["Shared services
(do it once, internally)"])
  D -->|"Yes"| O(["Outsource it"])
A make-or-buy screen built on transaction-cost logic: core, specifiable, and dependency risk decide build vs. share vs. buy. Leaders Loop

Shared services is "internal outsourcing." Outsourcing is just the same decision, one supplier further out.

An honest limitation. These models describe a tendency, not a guarantee, and the three-legged stool in particular has aged. Ulrich himself, speaking at the 2018 Love HR conference, said the model was "a good idea" in its day but "in 2018, it's not a good idea because the world has changed" (reported by People Management). The common failure is the "Bermuda Triangle": work falls between shared services, the CoE and the business partners, each assuming someone else owns it, and the employee gets bounced around. The structure only works when the handoffs and decision rights between the three are deliberately designed, which is exactly the part organisations skip when they copy the org chart and call it done.

A worked example

Take a mid-sized company, call it Harlow Group, around 4,000 staff across several countries. (Illustrative figures throughout; this is a teaching example, not a real company.) Its HR is the classic "one team doing everything" model: a generalist in each business unit handling everything from a payroll query to a redundancy programme to advising the divisional MD. Everyone is busy; nothing is fast; the same employment-law question gets answered five different ways in five regions.

Reframed through the three legs, the fix becomes obvious. The high-volume, repeatable work, contracts, onboarding, leave, the payslip queries, goes into a single shared-services centre with a self-service portal and a target of resolving 80% of queries at first contact (illustrative). The scarce, high-stakes expertise, reward design, employment law, organisational development, is pulled into small centres of excellence that set one consistent standard instead of five. And a handful of senior business partners stop doing admin entirely and sit with the divisional leaders on workforce planning.

Then the make-or-buy screen runs over each leg. Payroll itself, standardised, contractible, not a source of competitive advantage, is outsourced to a specialist provider; the shared-services centre manages the relationship rather than running the calculations. Reward design stays firmly in-house: it is bound up with how Harlow competes for talent, and no contract could capture it. What Harlow ends up with isn't really "cheaper HR", though it is cheaper. It's a function where routine things are fast and cheap, hard things are done well and once, and senior people spend their time on decisions only they can make.

Note the order of operations: Harlow designed the work first, then the structure, then the sourcing. Reverse that, buy the shared-services software, then figure out the processes, and you get the most common failure mode, an expensive centre that automated the old mess.

Frequently asked questions

What's the difference between a shared-services centre and a centre of excellence?

Volume and depth. A shared-services centre handles high-volume, routine work and is optimised for cost, consistency and speed, think a single payroll team serving the whole company. A centre of excellence handles low-volume, high-expertise work and is optimised for quality and standards, think a small reward-design or cyber-security team whose job is to be the best in the organisation at one hard thing. One is an engine; the other is a brain trust.

Is shared services the same as centralisation?

Related, but not identical. Centralisation is simply pulling decisions or work into the centre. Shared services is a specific, more disciplined version: it pools the work and runs it like an internal supplier with service levels, internal customers, and a cost-and-satisfaction scorecard. Plenty of centralisation happens with none of that discipline, and tends to recreate the bottleneck it was meant to remove. (More in centralisation vs decentralisation.)

When should we outsource a function rather than build shared services?

Outsource when the work is standardised, easy to specify in a contract, and not core to your competitive advantage, and when the risk of depending on an external provider is acceptable. Keep it in-house (often as a centre of excellence) when it is bound up with what makes you distinctive, hard to write a watertight contract for, or dangerous to be held hostage on. This is the transaction-cost logic: build when contracting out would cost or risk more than doing it yourself.

Why do these models so often disappoint in practice?

Usually because the organisation copied the boxes and skipped the operating model. The three legs only work if the handoffs between them, who escalates what, who owns the employee's problem end to end, what self-service covers, are deliberately designed. Skip that and work falls into the gaps, employees get bounced between teams, and people conclude "the new structure made things worse." It wasn't the structure; it was the missing wiring between the parts.

Is the three-legged stool still the right model?

As a way of thinking about a function's distinct jobs, the split between routine work, deep expertise and embedded advice is durable. As a rigid blueprint, even its author has moved on, Ulrich now argues the model needs rethinking for a more digital, networked world. Treat it as a lens for asking "which kind of work is this, and where should it live?" rather than a fixed shape every function must take.

Related in the Toolkit

This whole topic is one chapter of a bigger design conversation: how you split a function maps onto how you structure the whole organisation, and the part that most often breaks, work falling between the legs, is really a failure of decision rights.

Where to go next