Short answer
AI can help HR draft content, retrieve approved information, analyse a defined workforce question and prepare human conversations. Choose a use case by the decision it supports, the people affected, the source material and the reviewer who owns the result. A tool that produces plausible language is not automatically reliable enough for an employment decision. Start with a bounded workflow and test its output in context.
Start with the decision, not the model
The SHRM State of AI in HR 2026 overview frames AI adoption as a people-leadership issue as well as a technology issue. The accessible overview describes research among HR professionals but does not establish that a particular tool or use case will work in every organisation. The practical first question is smaller: what work decision is slow, inconsistent or poorly evidenced today?
For example, “use AI for HR” is not a decision. “Help managers prepare consistent questions for a new-starter conversation using approved onboarding material” is. It has a purpose, source set, reviewer and a way to check whether the output is usable.
Map the use case before buying or building
| Use case | Useful output | Source and review check | Do not assume |
|---|---|---|---|
| Drafting HR content | A first draft of a job description, policy explainer or learning aid | Approved source material, named editor and factual check | The draft is accurate, current or approved because it reads well |
| Knowledge retrieval | A cited answer from an approved policy library | Source link, version date, access rules and owner | A general model knows the current local policy |
| Workforce analysis | A reproducible summary of a defined pattern | Population, measure, time period, calculation and analyst review | Correlation establishes cause or predicts an individual outcome |
| Conversation preparation | Questions and a short context brief | Clear purpose, permitted context and human follow-up | A prompt replaces a manager or makes a sensitive decision |
| Skills and work evidence | A structured record of an example or requirement | Source, date, context and reviewer | A model can infer a complete personal capability profile |
The UK CIPD's AI skills-planning guidance provides a practical people-profession lens on AI-driven skill change. Use it to frame work and capability questions; do not turn a generic framework into an automatic individual assessment.
Choose the level of human review
Not every output needs the same approval route. A wording suggestion for an internal draft may need an editor. A summary that could affect a manager conversation needs the manager to check source context. A recommendation connected to hiring, pay, performance, promotion, work allocation or discipline requires an accountable human process and appropriate policy review.
The NIST AI guidance takes a risk-based approach to managing AI-related risks. In HR, that means looking at consequence as well as technical quality. A concise output can still be unsafe if its source is stale, its scope is unclear or its audience treats it as a decision.
Copy this AI use-case card
| Field | Question to answer |
|---|---|
| Work problem | Which decision or task needs improvement? |
| People affected | Who will receive, use or be represented in the output? |
| Permitted sources | Which documents, records or inputs may be used? |
| Output | What form of draft, answer, analysis or brief is expected? |
| Reviewer | Who checks accuracy, context and suitability before use? |
| Limits | What must the output not decide, infer or disclose? |
| Test | What small, real workflow will show whether it helps? |
| Review date | Who will decide to continue, change or stop? |
A fictional pilot
This example is fictional. A people team wants help preparing new-starter conversations in three locations. It does not ask a model to decide whether any new starter is performing well. It supplies the approved induction plan and asks for a short question set about access, training and first-shift support.
An HR partner reviews the questions against local operating conditions. Managers use the approved version for one cohort, record only agreed follow-ups, and identify which questions employees found unclear. The review examines whether managers received the material, whether conversations happened and whether the resulting actions were usable. It does not claim that a better prompt caused retention.
Where Lontra fits
Lontra can help a team explore one bounded employee-work question through structured conversations and manager briefs. It is useful when the decision needs employee context alongside operational evidence. Managers receive briefs rather than raw conversations; people remain accountable for interpretation and action. A trial is one campaign with up to 30 invitations over 60 days, without a card. Explore the product.
For creating and reviewing a specific generative output, use the generative AI HR workflow guide. Once you have chosen a workflow, use the AI HR implementation guide to organise the pilot. If you are deciding whether another tool is needed, begin with the HR tech stack audit.
Sources
- SHRM State of AI in HR 2026 overview
- CIPD AI skills planning guidance
- NIST artificial intelligence and risk-management resources
Frequently asked questions
- What are useful AI use cases in HR?
- Useful use cases start with a defined work problem, such as drafting a job description, retrieving an approved policy, analysing a workforce pattern or preparing a conversation. The required source quality, reviewer and consequence should determine the design.
- Can AI make HR decisions?
- An AI output can organise information or draft material, but it should not be treated as the accountable decision-maker for employment, performance, pay, promotion or other consequential outcomes.
- How should an HR team choose an AI pilot?
- Choose one bounded decision, identify permitted inputs and a reviewer, define a practical quality check, and decide in advance what evidence will determine whether to continue, change or stop.