What is HR service delivery? Models, examples and best practices

You're fielding five Slack DMs about open enrollment while trying to pull a headcount report for the executive team. Your headcount is flat. Employee requests aren't. Somewhere in that gap sits a decision most People teams never made on purpose: how HR support reaches employees, who handles what, and what a system does on its own.

That decision is your HR service delivery model. Copying the IT ticketing playbook only moves the bottleneck from your inbox to a portal nobody wants to use. The model you choose decides whether your team spends the year answering the same question or shaping the company.

What is HR service delivery?

HR service delivery is the set of processes, people, channels and technology an organization uses to get HR support, information and services to its employees. It covers everything from how a pay stub question gets answered to how an employee relations case is routed and tracked.

When you design a delivery model, you're choosing how your team interacts with the workforce. Proactive delivery anticipates lifecycle needs, like starting onboarding before a new hire asks for a laptop. Reactive delivery waits for the request, then works the queue.

Most scaling companies struggle with the reactive side, and the numbers explain why. In ScottMadden's 2025 workshop on HR optimization, 65% of the work done by business-facing HR staff is transactional in a traditional model. In a transformed model that share falls to 20%, with roughly 45% of the administrative workload moving to a service center. That gap is the difference between running programs and running a help desk.

The point is making room for more personal support where it counts.

HR service delivery models compared

HR service delivery models describe how HR people and technology are organized to support employees. Most companies run a blend of two or three. Here's where each one breaks.

Centralized, decentralized and shared services

A centralized model runs all HR support through one corporate team. It keeps policy consistent and it's the default for most mid-market companies. A decentralized model embeds HR people inside departments or regions. It feels closer to the business, but it tends to produce duplicate processes and inconsistent answers.

In practice it's rarely either-or. In the National Association of State Personnel Executives' 2025 survey of 31 state HR agencies, responsibility was most often split or distributed rather than fully centralized: onboarding sat with departments in 16 states and was split in 11, while compensation was centralized in 15. Policy-heavy work centralizes. Employee-facing work stays close to the employee.

Shared services sits between the two. Routine, transactional work (address changes, verifications, benefits questions) is pooled into one hub that serves the whole company, while specialist and business-facing HR roles stay close to teams. It's efficient, and it can feel distant. After West Virginia centralized HR into a shared-services office in 2023, a 2026 audit found staff reporting that HR services felt more distant, less responsive and less tailored to their needs.

The Ulrich model

The Ulrich model, from Dave Ulrich's work in the 1990s, splits HR into three parts: a shared services function for administration, centers of expertise for specialist work like compensation or talent, and embedded HR leads who work directly with business unit heads. The first two absorb the volume so the third can focus on the business.

In practice the stool wobbles when the shared services leg can't handle daily query volume. The work lands on whoever your employees can reach in Slack.

The tiered model (Tier 0 to Tier 3)

The tiered model organizes requests by complexity into an escalation path. Tier 0 is self-service: a knowledge base, a portal or an AI assistant. Tier 1 is generalist support for first contact and routine transactions. Tier 2 is specialist support for benefits, payroll or immigration questions. Tier 3 is the center of expertise or senior HR for sensitive, complex or policy-shaping issues.

The tiers aren't the problem. The problem is what happens at Tier 0 and Tier 1, where most of the volume lives.

The employee-centric model

An employee-centric model starts from the employee's journey rather than HR's org chart. You map the moments that matter (first day, first pay run, a move, a new baby, a leave, an exit) and design support around them, in the channels employees already use. Without a delivery mechanism that can act, though, it becomes a nicer-sounding portal.

The AI-first model

An AI-first model puts an AI agent at the front of the queue. The agent answers policy questions for that specific person, completes routine transactions in the systems of record, and escalates to a human only when a case needs judgment or empathy. Humans keep Tier 2 and Tier 3. The agent takes most of Tier 0 and Tier 1.

It's early. In NASPE's 2025 survey, only three of 31 state HR agencies had AI tools implemented and in use, eleven were piloting, and early use was concentrated in front-line applications such as chatbots or virtual HR assistants. AI starts at the front door.

This is the model we build at Kinfolk. The honest caveat is that it depends on integrations, permissions and audit trails more than on the AI model behind it. An agent that can read the handbook but can't update the HRIS is a chatbot with better manners.

HR service delivery examples: five requests and how they travel

The clearest way to understand a delivery model is to follow a request through it. Here are five from every People team's queue.

A PTO balance check. Tier 0 in theory. In practice the employee can't find it in the HRIS, so they DM someone in People Ops. In an AI-first model the agent reads the balance from the HRIS and answers in Slack.

An employment verification letter. A Tier 1 transaction: confirm employment, pull the dates and title, generate the letter, send it. An agent with HRIS access can do it in the thread where it was asked.

An address change. Tier 0 if your HRIS offers self-service for it. In ScottMadden and APQC's 2024 benchmarking, employee direct access is offered for personal data changes at 60% of organizations, payroll requests or changes at 54% and benefits transactions at 53%. That leaves plenty of routine updates landing on a person.

New-hire onboarding. Not a ticket at all. It's a lifecycle program: accounts, equipment, channels, first-week schedule, benefits enrollment. Reactive models turn each piece into a separate request. Proactive models let the HRIS record trigger the whole sequence. At sports analytics company Hudl, onboarding coordination was cut from 60 to 20 minutes per hire after moving that coordination to AI agents in Slack.

An employee relations case. Tier 3, and it should stay there. A complaint about a manager needs a human, a secure case record and confidentiality. A good model keeps this work away from automation and gives your team the time to do it properly.

Why ticket queues don't fix HR service delivery

A ticket queue organizes work. It doesn't do the work. That's the flaw in copying the IT playbook for HR, and it's why so many portals end up ignored. Sensitive, complex cases still need human routing and a secure case record. You don't want to automate those.

The problem is the rest of your volume. Forcing employees to log into a portal for a quick policy check or a pay stub question creates what John Seddon calls "failure demand": demand caused by a failure to do something right, or make it easy, the first time. Employees give up on the portal and DM whoever they know in HR. The queue looks quiet. The work didn't go anywhere.

Pointing someone at a Confluence page isn't service delivery either. It's a reading assignment. Real delivery means the request gets resolved: the balance checked, the letter sent, the record updated. That's autonomous orchestration, and it's why we think service desks built for IT tickets are only half the answer for HR. Choose the model first, then the tools.

How to design an HR service delivery model

Designing a delivery model comes down to three moves: audit what employees ask for, define what you deliver, and decide what does the work.

Step 1: Audit your request volume and failure demand

Start with where employees get stuck. Count the shadow requests in DMs, emails and hallway conversations, not just the ticketing tool. Then look for repeats. If the same question about PTO rollover or sick days arrives every week, that's your first candidate for an agent.

Step 2: Define your service catalog

Write down what your team delivers: onboarding, offboarding, benefits changes, payroll corrections, verification letters, leave, policy questions. Note the steps to resolve each and which tier owns it. If a process needs six approvals and three forms, fix it before you write it down.

Step 3: Decide what executes the work

Your HRIS is your system of record. A ticketing tool tracks work. Neither completes a request on its own. Whatever sits in front of employees needs to act on the systems of record with the right permissions. That's the orchestration layer, and most delivery models leave it out. Tools like Kinfolk connect to your HRIS and handle requests in Slack, so the agent executes the workflow autonomously instead of returning a link.

Change management and rollout

Tell employees what changed and where to go. Make the human escalation path visible from day one: if the agent can't resolve something, a person picks it up in the same thread. That's what makes your rollout stick.

Permissions, privacy and governance

HR service delivery governance is the set of access controls, audit trails and privacy rules that keep employee data visible only to the people (and agents) authorized to see it.

Role-based access

Flat permissions are the fastest way to turn a helpful HR agent into a data incident. Your delivery layer must inherit permissions from the HRIS: managers see their direct reports, employees see their own records, and an agent retrieves only what the person asking is allowed to see. Enforce it at the data layer, not in the prompt.

Auditability

Tools change. AI models improve. Your need for a record of who approved a compensation change, who requested a leave and when doesn't change. Build on an orchestration layer that logs every action, so the audit trail outlives whichever tool you're using this year.

HR service delivery best practices

These come from where delivery models actually fail: measurement, process and channels.

Measure experience, not just throughput. In LACE Partners' 2024 survey of shared services leaders in HR, 74% measured service effectiveness with operational measures versus 53% with employee experience measures, and 24% didn't measure service effectiveness at all. If you only track tickets closed, you'll never see the failure demand.

Fix the process before you automate it. An AI wrapper on a broken six-step approval only makes the bad process run faster.

Go where employees already are. If you launch a portal and people keep DMing your team in Slack, the portal has failed, not the employees. Bring the delivery to the channel. It's why we build in Slack.

Set service levels, even if nobody asked for them. Without a target, nothing is late. When the University of Northern Colorado audited its HR function in 2024 with no service level agreements in place, 59 of 89 open requisitions, about 66%, had been open for more than 90 days. Without SLAs, the auditors noted, there were no clear metrics for judging HR performance at all.

Keep humans on Tier 3, on purpose. Automation earns trust when it's obvious what it won't touch. Publish the boundary.

Track resolution, not deflection. Deflection counts the tickets that didn't get created. Resolution counts the requests that got done. Only one tells you whether employees got what they needed.

Measuring HR service delivery

HR service delivery ROI comes down to how fast requests get resolved, how many get resolved without a human, and how many hours come back to your team.

Resolution rate and time to resolve

Resolution rate is the share of requests fully handled without escalation to a person. Time to resolve is how long the employee waited. Both need a defined window. Hudl, a Kinfolk customer, published 70% of requests handled by AI without escalation in six months and 80% of requests meeting first-response SLA targets. That's the shape of a useful benchmark: a rate, a window and a target. For lifecycle events, add time to productivity: how quickly your new hires get accounts, equipment and answers.

Cost per interaction

Take the loaded hourly cost of the person answering a routine request, multiply by the minutes each one takes, then by monthly volume. Use your own numbers; the point is to make the admin overhead visible. Then compare it with an agent doing the same work. Kinfolk's figures are up to 80% of Tier 0–1 employee requests handled without a human and roughly 45 days a year reclaimed for HR teams.

Frequently asked questions

What is the difference between shared services and HR service delivery?

HR service delivery is the whole framework for getting HR support to employees. Shared services is one model inside it, where routine work is pooled into a single hub.

What are the tiers of HR service delivery?

Four tiers, by complexity. Tier 0 is self-service through a knowledge base, portal or AI assistant. Tier 1 is generalist support for routine requests. Tier 2 is specialist support such as benefits or payroll. Tier 3 is the center of expertise or senior HR for sensitive or policy-level issues.

How does an HR service delivery model affect employee experience?

It decides how long employees wait and whether they trust HR to get things done. Resolving requests where people already work removes friction. A portal usually adds it.

What is the best HR service delivery software?

It depends on your model. A tiered model with a human Tier 1 needs case management. An AI-first model needs an orchestration layer that connects to your HRIS, respects its permissions and completes requests. Kinfolk is built for the second, in Slack.

How do you measure HR service delivery success?

Track resolution rate, time to resolve, self-service completion, employee satisfaction with HR and the hours returned to your team. Measure experience alongside throughput, or you'll miss the requests that never reached the queue.

Where HR service delivery goes next

Traditional models gave HR a way to organize volume, not a way to remove it. The next step is a model where routine requests get resolved where your employees already work, sensitive cases stay with people, and every action leaves a trail.

If you want to stop being the human router for routine requests, see how it works. Book a demo