AI in HR: Where Should a Mid-Sized HR Team Start?

AI in HR creates value when it is applied to a specific process, rather than introduced as a growing collection of unused features. This is particularly important for mid-sized companies. After all, a small HR team rarely has the time or the need to launch a large-scale AI project, but could gain a great deal by choosing a single, highly targeted AI adoption scenario within HR.

So where to start? Let’s explore how you can identify the process best suited to AI, what data and agreements need to be prepared, where human decision-making should remain, and how to ensure that AI becomes a genuine help to the team rather than just another feature switched on.

Quick Take

• When introducing AI in HR, it is better to focus on a clear, frequently recurring problem rather than chasing the most impressive technological possibilities.
• Four criteria will help identify suitable AI use cases in HR: the frequency of the problem, the cost of errors, data availability and the ability to measure results on a quarterly basis.
• One of the biggest challenges of AI in HR is inconsistent employee data and processes carried out differently across teams.
• Data privacy and ethics in AI for HR starts with clearly defined human responsibility: the system may perform an action, make a recommendation, or merely prepare material to support a decision.
• AI adoption in HR should only be scaled up once the first modest result is measurable and the team is actually utilising the chosen solution.

AI adoption in HR should start with a problem

The possibilities of AI are indeed vast. Given this, it is hardly surprising that, once discovered, you may be tempted to apply the technology across as many processes as possible. However, when assessing potential AI use cases in HR, the first project should not be the most ambitious. Instead, it should be the one that is easiest to implement and test.

A good first step in AI in HR starts with a process that involves a lot of repetitive work, clear rules and sufficient data. What does the HR team do frequently and spend a lot of time on, yet does not require constant human attention? Perhaps it is research, routine document preparation or gathering material for interviews? In other words, which task currently prevents the team from focusing on higher-value actions?

Once such a process has been identified, it is worth asking: does it really need a new tool? An organisation already using SAP SuccessFactors can begin its first use of AI in HR by utilising the existing platform and the data stored within it. That said, having an existing platform does not automatically mean that you can implement a new solution without preparation. The most common challenges of AI in HR often arise outside of the technology. You may need to align the data, process, configuration and change communication. However, in such cases, all preparation centres on a single specific problem, rather than an abstract AI programme detached from day-to-day HR work.

Which HR process should you choose first?

Four fairly simple criteria will help you find the answer to this question – use them to evaluate HR processes currently taking place within your organisation:

• Frequency of repetition. The more frequently a task is performed, the sooner even a small improvement can show its value.
• Cost of error. Consider not only the time wasted, but also what happens when information is missed, misinterpreted or communicated too late. The consequence could be the loss of a strong candidate, a decision based on incorrect data, or additional costs incurred in rectifying a mistake that has already been made.
• Data availability. AI cannot provide reliable assistance if the required information is fragmented, incomplete or stored without a common structure.
• Measurable results. Choose a process where progress can be assessed within a quarter, for example, in terms of time saved, shorter task duration or frequency of use.

What does a practical first AI use case look like?

Useful examples of AI in HR often begin with a simple question: where is the HR team’s time being spent today? In one organisation, this might be preparing for performance review meetings, where information has to be gathered from several sources. In another, it might be the initial stages of candidate screening, where the same actions are repeated. Elsewhere, a great deal of time is spent answering routine employee queries, even though the necessary information already exists in the organisation’s records.

These situations vary, but they have one thing in common: the work is repetitive, the data required for it already exists, and the time saved can be observed and measured. SAP also points out that Joule can help with day-to-day tasks, find information and provide answers based on HR policy documents. However, whether a specific feature is suitable for solving a particular problem must be assessed on a case-by-case basis – just as with the currently available SAP SuccessFactors AI features.

Challenges of AI in HR

Although we often hear that AI can solve almost any business problem, this is not always the case. If the same process is currently carried out in five different ways, AI is unlikely to produce reliable results. This technology requires consistency and a repetitive structure; therefore, if your team has not had these in place until now, you will need to prepare before implementing AI in HR. This is where many of the biggest challenges of AI in HR lie: first, you need to create a common system, get used to following it, and only then entrust part of the work to artificial intelligence. For example, when job titles, objectives, skills or employee data are entered inconsistently, the system does not receive a reliable context. Even a technically functioning feature may then produce a result that the team cannot rely on, or which will need to be carefully checked each time.

Reliable AI starts with reliable data and processes

AI data readiness is not merely a technical matter of data transfer. The HR team must agree on who enters the information, in which fields it is stored, what minimum quality requirements apply, and which source is considered the primary one. Only then can AI in human resource management draw on a sufficiently consistent organisational context.

The process itself is just as important. If operational objectives are formulated in one way in one team and in another way in another, AI will merely accelerate the existing inconsistency. In such situations, it is first necessary to agree on a common process logic. The same principle applies to HR workflow automation: the team first needs to choose one process with a clear problem, owner and outcome.

AI adoption in HR involves more than enabling a feature. The HR team must ensure that the process, data and responsibilities are sufficiently clear for its practical use.

Who has the final say? Data privacy and ethics in AI for HR

Simply agreeing that a result generated by artificial intelligence will be reviewed by a person from the team is not enough. Data privacy and ethics in AI for HR begin with a specific decision on what role the AI system will play in the chosen process. Practically speaking, you should choose one of three levels:

• The system independently performs a clearly defined, low-risk action.
• The system provides a recommendation, but a human evaluates it and makes the final decision.
• A human manages the entire process, whilst the system merely collects, summarises or prepares the material required for the decision.

The greater the potential impact of a decision on an employee or candidate, the more important human oversight of the AI’s logic becomes. This is particularly clear in recruitment. SAP SuccessFactors Recruiting still offers applicant screening that identifies skills in candidate data and shows how they match the requirements of a role. At the same time, SAP’s current talent acquisition direction centres on SmartRecruiters for SAP SuccessFactors, with Winston providing AI-supported candidate matching, screening and shortlisting. In either case, AI can organise relevant information and support the evaluation, but the recruiter or hiring manager should remain accountable for the final decision.

This is not merely a question of using a specific function. In SAP’s principles of responsible AI, human oversight is mentioned alongside transparency, explainability, responsibility and accountability. Consequently, responsible AI in HR must be assessed not by whether a person is formally involved in the process, but by whether they understand the information provided by the system and actually retain the right to make the decision.

However, clear boundaries for decision-making are not enough. Another question must also be addressed: who can view and use the information? This can create tension between HR, which sees an opportunity to reduce operational workload, and IT, which is responsible for risk and control. Agreeing the boundaries together helps prevent that tension from delaying an otherwise manageable use case. AI governance in HR must define the limits of access – what data is required for a given process, who can access it, and what information should not be used as input. These agreements must be adopted jointly by HR, IT and, where necessary, the legal function.

How can AI in HR be integrated into everyday practice?

Simply activating an AI feature does not guarantee success. The benefits of AI in HR become measurable when the team returns to the feature because it improves a specific part of their work. Therefore, the first stage usually requires one person in charge, a clearly defined user group, and one or two small, tangible results that can be observed without a complex reporting system.

Instead of measuring whether the feature is enabled, it makes more sense to measure its usage and effect on the process. Do managers actually rely on the prepared meeting points? Do staff find answers more quickly? Does the HR team spend less time on repetitive tasks? This is precisely what ‘measuring adoption rather than deployment’ means – assessing not the technical roll-out, but actual usage and the change it brings about.

The team also needs to know when they can rely on an automated action, when a recommendation needs to be reviewed, and when a decision must remain entirely in human hands. An example of such an assistant is SAP Joule, but success is determined not just by the product’s name, but by its clear role within a specific process.

Once the first use case becomes a stable part of day-to-day work, you can decide if it is worth applying the same principle to another process. Such a staged AI rollout within a mid-sized HR team allows you to learn from the team’s real-world experience without the pressure of having to overhaul the entire HR function straight away.

FAQ

Where should a mid-sized HR team start with AI?

For a mid-sized HR team, it is best to start with a single, clear day-to-day problem rather than a broad AI project. The first AI in HR trial should take place within a familiar process, have a specific owner, a defined user group and a result that can be verified within a single quarter.

Which HR process is the right first AI use case?

The right first process is one that is frequent, sufficiently standardised and has a clearly measurable outcome. When evaluating AI use cases in HR, it is worth comparing the frequency of occurrence, the cost of error, data availability and the ability to quickly verify the benefits. For example, preparing for performance review meetings may be a more suitable choice than a complex function that the team will use infrequently and whose impact will be difficult to measure.

What needs to be in place before introducing AI in HR?

Before you start, you need consistent data, a standardised process, a clear single source of truth and clearly defined responsibilities. AI readiness in HR means that the team has agreed who fills in specific fields, what quality requirements apply to them, who will use the output and who will make the final decision. Without this foundation, AI in practice will merely replicate existing inconsistencies.

How do we introduce AI in HR responsibly?

Responsible AI implementation begins with defining the system’s specific role: it may independently perform a low-risk action, provide a recommendation to a human, or simply prepare material to support a decision. Data privacy and ethics in AI for HR also require defining what data the system may use and what the limits of its decision-making are. The greater the impact on an employee, the more important human oversight becomes.

How should HR balance AI opportunity with governance and control?

Controls should be applied in proportion to the risk of the specific use case. AI governance in HR must not impose the same restrictions on automated data collection as on recommendations that affect an employee’s career. HR, IT and, where necessary, the legal function should agree on access, oversight and accountability, whilst allowing the team the opportunity to safely and practically test approved, low-risk applications in real-world work.

Who remains accountable when AI supports an HR decision?

Ultimate responsibility remains with the person or function that managed the decision prior to the introduction of AI. Human oversight in the context of AI is not merely a formal endorsement of the result: the decision-maker must understand the information provided, assess the broader context and, where necessary, reject it. For example, a candidate’s skills match may assist in the selection process, but should not independently determine its outcome without additional human assessment.

How do we get an HR team to actually use AI?

A team begins to use AI regularly when the feature solves a recognisable day-to-day problem and fits naturally into the existing process. Effective change management for AI adoption in HR involves a single point of contact, a clearly defined initial user group, brief training and regular feedback.

How do we measure whether AI adoption in HR is working?

AI adoption in HR is effective when people regularly use the chosen feature and this leads to an improvement in a specific process. Before implementation, it is worth recording the baseline situation and then comparing usage frequency, time saved, task duration, errors or rework. Success is indicated by the resolution of the initial problem and the continuation of the usage habit, rather than by the number of AI features enabled.

Once readiness is clear, technical questions remain

Once the process, data and success metrics are clear, you can assess which SAP SuccessFactors AI capabilities are best suited to your organisation. Jigsaw Cloud can help you select the right use cases and plan their implementation.