AI is changing what employees expect from HR.
People are becoming accustomed to asking questions in natural language and getting immediate answers. They expect technology to understand context, find relevant information and increasingly, take action on their behalf.
For HR teams, that creates a significant opportunity. Create or find an AI HR support solution.
AI can answer routine questions, retrieve knowledge, route requests, trigger workflows and help employees complete tasks without navigating multiple systems.
I’ll admit that I have a complex relationship with AI.
I work with video and enjoy creative writing, so I see first hand how AI is being received in some creative spaces. Some are keen to experiment with the new tools now available to us.
But there is still a huge amount of value placed on the time, effort and skill that goes into making something yourself, from scratch.
People care about where creative work comes from, whose ideas have influenced it and whether the finished piece feels genuinely considered.
AI can challenge those expectations. It raises questions about where its output comes from, what information it has used and how much human input sits behind the result. It can also make things feel remarkably similar.
In creative work, that consistency can be a problem. In HR, that kind of consistency is exactly what you want.
An employee asking about annual leave needs the right answer, based on the right policy and their own data.
Someone requesting a proof of employment certificate needs the approved template, completed with the correct information.
A case needs to follow the right process and reach the right person.
In those situations, consistency is not creative sameness, it is reliability.
And reliability depends on more than the AI itself. It depends on the information it can access, the permissions around that information, the actions it is allowed to take and the point at which a human needs to step in.
That is where AI governance becomes important.
Recent EY research found that UK respondents were particularly comfortable adopting AI when it could provide practical outcomes, with 35% citing reliability as a key benefit.
Choosing an AI vendor for HR means finding the balance between capability and responsibility.
While it is great when an AI vendor can do everything, it’s actually more important to also understand how they do it, what controls are available and whether those controls are strong enough for the people and data HR is responsible for.
There are already examples of what can happen when AI is given too much freedom without appropriate safeguards. A New Zealand supermarket’s AI recipe tool suggested combinations including chlorine gas and other unsafe substances. Chevrolet’s AI powered test chatbot was famously manipulated into agreeing to sell a car for one dollar. These examples are very different from an HR service, but they demonstrate the same underlying principle: a system can follow the logic available to it without understanding the wider consequences of an answer or action.
It means AI needs boundaries.
We should understand where it performs well, give it the right information and permissions, test what happens when things go wrong and make sure there is always an appropriate route to human judgement.
And we’ll show you how to get there, selecting and managing the right AI vendors for your service needs.
This practical guide gives HR leaders a clear roadmap to move from low-risk pilots to scalable, governed AI implementation across the employee experience. Read Now.
For years, HR self service has followed a familiar pattern. An employee has a question, searches a portal, finds a policy and raises a case if they cannot find what they need.
AI changes that interaction.
Instead of navigating through systems and menus, employees can explain what they need in natural language. AI can understand the context, find relevant information and, depending on the use case and permissions, take the next step.
For example:
“I’m moving to Spain. What happens to my benefits?”
“I need a proof of employment certificate for a mortgage application.”
“I’m getting married next month. Can you update my personal details?”
These requests could involve knowledge retrieval, employee data, case creation, document generation or a workflow. AI can potentially bring those pieces together rather than leaving the employee to work out which system or team handles each part.
But greater capability also creates greater responsibility.
That need for control is becoming more urgent as AI moves into everyday work. Recent UK research from Skillcast found that 75% of professionals use AI at least weekly, while 68% say they encounter inaccurate or misleading AI outputs at least occasionally.
There is a significant difference between AI retrieving an approved HR policy and AI accessing employee data, creating a case or initiating a workflow. Each comes with a different level of risk and requires different controls.
This is also where the question of who builds and maintains AI becomes important.
IT teams are increasingly being asked to build AI solutions internally, or are offering to build them because they already have the technical capability.
An internal team may be perfectly capable of creating something that works. But building something that works is very different from building something that can safely operate as an HR service.
HR systems contain some of the most sensitive information an organisation holds.
Personal details, pay, absence, benefits, performance information and employee relations cases all require carefully controlled access.
An internally built AI service needs to enforce those same permissions, protect information as it moves between systems, provide appropriate audit trails and prevent an employee from using a natural language interaction to access information they would not otherwise be allowed to see.
That is a significant responsibility to take on.
The challenge also grows as the organisation grows. An AI solution that works for a limited number of use cases, employees or HR systems can become considerably more complex when new countries, policies, business units, integrations and workflows are added.
The more the service touches, the more there is to secure, test, monitor and maintain.
And HR does not stand still.
Policies change. Employee expectations change. HR processes change. AI models evolve. New systems and integrations are introduced. The service needs to be monitored, maintained, tested and improved throughout its lifetime.
An IT team that builds an AI solution may eventually need to move on to other priorities and business as usual. That can leave HR with a service that still works, but no longer evolves at the pace of the organisation. Without dedicated ownership, specialist HR knowledge and a clear approach to ongoing governance, something that was effective at launch can become harder to secure, harder to scale and harder to improve.
As HR leaders, an internal build might feel plausible but ask yourself:
That is why the question for HR leaders is not simply, “Can we build this?”
“Can we secure this, operate it safely, scale it as we grow and keep improving it as our HR service changes?”
And, perhaps most importantly, “Should AI do this, under what conditions and with what controls?”
Choosing an AI vendor for HR starts but doesn’t end with the product demo.
A system might be impressive in a demonstration, but that does not tell you everything you need to know. For HR, it is important to know what information AI can access, what it is allowed to do, how those decisions are controlled and what happens when something goes wrong.
A useful way to assess a vendor is to look at five key areas.
Can the AI retrieve knowledge?
Can it create cases?
Can it generate documents?
Can it trigger workflows?
Can it take action in connected systems?
What information can AI use to understand a request?
Can AI work with approved HR knowledge and relevant employee data?
Can it distinguish between information that is generally available and information that is specific to an employee?
What determines what AI can see and do?
Can permissions, business rules and approval requirements be applied to different actions and users?
Can the organization decide which actions AI can complete independently and which require human involvement?
Can the vendor demonstrate how AI is tested, monitored and audited?
Can you understand what information was used, what happened and whether the outcome can be reviewed?
What happens when the AI reaches the limits of what it should handle?
Can a human intervene, take over or correct the outcome when needed?
These criteria matter because HR deals with information and decisions that can have a real impact on people.
An AI system might surface outdated information, access data a user should not see, create a case with the wrong priority level or take action that should have required human involvement.
A good AI vendor should be able to explain both what the technology can do and how it operates within the wider HR service. That includes its data sources, permissions, workflows, auditability, escalation and approach to ongoing improvement.
Throughout the whole process of selecting a vendor, ask yourself:
“Can this AI do what we need, with the controls we need, for the people and data we are responsible for?”
Ivan Harding explores the disconnect between how HR leaders perceive their organization's employee experience and how employees actually feel about it. Read Now
Different AI use cases carry different levels of risk.
The simplest use cases are questions and knowledge retrieval. An employee might ask about annual leave entitlements, parental leave or a company policy and receive an answer based on approved HR information.
Here, the key considerations are accuracy and access. Is the information up to date? Can the AI identify the right source? Does it respect the employee’s permissions?
Knowledge retrieval becomes more complex when information is spread across policies, guidance, documents and HR systems.
Connecting more sources can make AI more useful, but it also makes governance more important. Organisations need to understand what sources are being used and how access is controlled.
Then there is case creation.
An employee might describe a workplace issue in natural language and expect the AI to understand what they need, gather the relevant information and create a case for the right specialist or team.
At this point, questions around data become more important.
What information is being captured?
Who receives it?
Has the employee had an opportunity to check or correct it?
How does the AI decide how the case should be categorised or routed?
This is also an area where AI can work particularly well alongside people.
AI can handle the administrative consistency involved in routing, prioritising and categorising a case, while a case specialist focuses on understanding the situation and making the judgement required to reach an appropriate resolution.
Workflows and approvals take things another step further.
AI might initiate an employee process, generate a document, request an approval or coordinate actions across connected HR systems.
This is where the distinction between answering and acting becomes particularly important. The more consequential the action, the clearer the permissions, business rules and human oversight need to be.
For example, generating a proof of employment certificate from an approved template can be a highly structured process. There is a defined document template, defined information and a defined outcome.
A complex employee relations matter is very different. It may require context, judgement and human involvement that AI should support rather than replace.
That is why good AI governance starts with your use cases.
The terms chatbot, AI assistant and AI agent are often used interchangeably, but they describe different levels of capability. Understanding the distinction is useful when evaluating what you actually want an AI vendor to deliver.
A chatbot is primarily built to have a conversation. It can answer predefined questions, guide employees towards information and provide a simple interface for support. It may be useful for straightforward enquiries, but its ability to understand context can be limited.
An AI assistant provides broader support. It can understand natural language, retrieve relevant information, use context and, where it has the right integrations and permissions, help employees complete tasks.
An AI agent goes further. It interprets an objective, works out the steps required to complete it and uses connected tools or workflows to get it done.
For example, an employee might ask for a proof of employment certificate. An agent could identify the appropriate approved template, retrieve only the relevant employee information it is authorised to access, generate the document and return it to the employee.
An agent could also handle a more involved HR request by deciding which steps are needed, carrying them out in sequence and involving a person where the process requires human judgement.
That does not mean an AI agent should have unrestricted autonomy.
In HR, we need to ensure AI agents operate ethically and securely. There should be situations where an agent is allowed to act independently and others where it needs rules, approvals or human intervention.
This distinction is important when comparing vendors.
A product may describe itself as an AI agent, but the label tells you very little about how it actually operates. Look at what sits behind the interaction.
What systems can it access?
What actions can it take?
Which permissions apply?
Can workflows and approval rules be enforced?
Can its actions be reviewed afterwards?
Can a person intervene?
Ensuring AI takes the right action within clearly defined boundaries is key.
Learn how to transform HR into a people-first function that builds trust, designs better experiences, and drives real business results in this interactive, 10-minute guide. Read Now.
For HR, AI governance ultimately comes back to people.
An employee might use AI to ask about their benefits, absence, pay, performance or a difficult situation at work. These situations can be deeply personal and employees need to know that using an AI powered HR service will not expose their sensitive information to others, or others’ information to them.
That makes privacy a central part of a well managed employee experience.
A well governed AI service should collect and use only the information needed to fulfil the request. It should respect existing access controls and keep sensitive information within the boundaries already established by the organisation.
It should also be clear about what happens to employee information. HR teams should understand where data is processed, whether customer data is used to train models, what is retained and who can access it.
But privacy is only one part of the picture.
Guardrails need to cover what AI can see, what it can decide and what it can do.
For example, an AI system might be allowed to retrieve an approved policy without human intervention. It might be allowed to generate a standard document using an approved template. It might be allowed to create and route a case.
A different process might require approval before any action is completed.
And some situations should remain firmly within the domain of human judgement.
Employees also need a way out.
AI should not become a barrier between an employee and HR. If someone is dealing with a sensitive workplace issue, the right outcome might be a conversation with a person rather than another automated response.
If the AI is uncertain, cannot answer appropriately or encounters a situation requiring judgement, it should make that transition easy.
The same principle applies to HR teams. People need visibility into what AI is doing and the ability to intervene when necessary.
This is particularly important as AI capabilities evolve.
Models change, knowledge changes, policies change and organisations introduce new workflows. Governance therefore cannot be a one time approval completed when the technology is first introduced.
It needs to be an ongoing part of how the service is managed.
The strongest approach is to put clear boundaries around AI while keeping the human experience at the centre.
AI should make it easier for employees to get the right help, not make it harder to reach a person when they need one.
When evaluating AI vendors for HR, some capabilities are useful differentiators. Others should be considered non negotiable.
These are the foundations that should be in place before an organisation starts giving AI responsibility for employee service.
AI needs access to the systems and information that make HR service useful, but those connections need to be controlled.
A vendor should be able to show:
Which HR systems can be connected.
What information AI can retrieve from each system.
What actions AI can take.
How connections are secured.
Whether access can be limited to specific use cases.
An integration should give AI the information or capability it needs without creating unnecessary access.
AI should not create a new route to information an employee could not otherwise access.
Check that the vendor can enforce:
Employee and manager permissions
Access to sensitive HR information
Role based access
Restrictions on specific actions
Approval requirements where appropriate
If the underlying HR system says an employee cannot see something, asking an AI should not change that.
If AI takes part in an HR interaction or process, HR should be able to understand what happened.
That means having visibility into:
The request.
The information used.
The action taken.
The outcome.
Any human intervention.
This is particularly important when AI moves from answering questions to taking action.
There must always be a clear path from AI to human support.
That means employees can reach a person when:
Their situation is sensitive.
Their request needs judgement.
AI cannot provide a reliable answer.
They simply want to speak to someone.
The handover should preserve context wherever possible. Employees should not have to start their story again.
Once AI is live, HR needs to know whether it is actually helping.
The vendor should provide insight into things such as:
What employees are asking.
Where AI is resolving requests.
Where people are being escalated.
Where employees are repeatedly asking for help.
Which knowledge areas may need attention.
How AI is affecting HR service demand.
This turns governance into an ongoing process rather than a one time vendor assessment.
It also highlights an important difference between buying AI and operating an AI powered HR service.
The technology needs an owner.
Someone needs to understand what employees are asking, identify where the service is falling short, maintain the knowledge behind it, review how workflows are performing and make improvements as HR processes change.
That ongoing responsibility is difficult to maintain when AI is treated as a one off IT project.
Good AI governance does not end when the vendor is selected. How AI is introduced matters just as much.
The temptation can be to launch broadly because the technology can handle a wide range of requests. A more controlled approach is to start with a small number of clearly defined use cases and learn from them.
Choose processes that are:
Frequent enough to generate meaningful usage
Straightforward enough to define clear outcomes
Well supported by reliable HR information
Low risk if something needs to be corrected
This could mean starting with common policy questions, knowledge retrieval, simple HR requests or document generation before moving into more complex processes.
Decide what you want to improve before AI goes live.
Depending on the use case, that might include:
How quickly employees get an answer
Whether they find the information useful
How often they still need HR support
How long a process takes to complete
How much manual work is removed
Whether employees complete the process successfully
The important thing is to measure the outcome, not simply the amount of AI being used.
Early usage should reveal where the service works well and where it needs attention.
You might discover that:
Employees are asking questions the knowledge base does not cover
A process needs clearer rules
Some requests consistently need human support
Employees are unsure what information to provide
A particular workflow needs refinement
Those findings are valuable.
Fix the underlying issue before giving AI responsibility for a wider range of tasks.
Once an AI service is performing reliably, its role can grow.
That might mean moving from answering questions to helping complete requests, then into more structured workflows and actions.
The goal is not to automate as much as possible.
The goal is to give AI the right level of responsibility for each HR process, based on what it can reliably handle and what employees actually need.
Applaud takes an HR first approach to AI.
The focus is not simply on adding AI to an existing HR system. It is using AI to make the experience of getting HR support easier for employees and more manageable for HR teams.
AI can sit across the different parts of the HR service, bringing together knowledge, employee information, case management and workflows.
Employees can interact with Applaud through the channels they already use, including the Employee Portal, Teams, Slack and Outlook. They can ask questions in natural language, find relevant information and, where appropriate, move from asking for help to getting something done.
That might mean:
Finding an answer from approved HR knowledge
Creating or updating an HR case
Generating a document
Starting an HR workflow
Supporting an approval
Connecting information from different HR systems
The important part is what sits around the AI.
In Case Management, for example, AI can handle the administrative consistency involved in assigning, prioritising, categorising and planning cases, while case specialists focus on the judgement and human interaction needed to find a resolution.
That plays to the strengths of both.
AI brings consistency, speed and administrative capability. People bring context, empathy and judgement.
The same principle applies to employee requests.
Applaud’s AI Assistant can do more than answer a question. Where the appropriate permissions and workflows are in place, it can take action.
An employee can generate a proof of employment certificate, submit a time off request or update personal details through the interaction rather than being sent to another system to complete the task themselves.
This is where the difference between an AI that can answer and an AI that can operate as part of an HR service becomes important.
Applaud combines AI with the knowledge, permissions, workflows, case management and analytics needed to deliver HR services responsibly. This gives HR teams visibility into what is happening while giving employees a simpler way to get things done.
It also means AI does not have to make every decision itself.
Where a process needs human judgement, the service can involve HR at the appropriate point. Where a process is structured and repeatable, AI can handle the steps consistently.
For HR leaders assessing AI vendors, that distinction matters.
The question is not simply whether an AI can answer questions or perform tasks.
It is whether the technology can operate as part of a complete HR service, with the context, controls, ownership and human support needed to make it useful in the real world.
That is the role AI can play in HR: helping employees get the right support, helping HR teams spend their time where human judgement matters most and making the overall service easier to navigate.
AI does not need to replace the human side of HR to be valuable.
In many cases, its greatest value comes from handling the things it is particularly good at consistently, reliably and within clear boundaries, so people can focus on the things that require people.
Scott Oakes is a Senior Technical Communications Specialist with a background in advertising, communication, creative writing, and video production. He plays a key role in shaping Applaud's YouTube channel, combining creativity and clarity to make technical how-to videos more engaging, ensuring that complex concepts are easy to understand.