Where Does AI Help HR Most, and Where Must a Person Decide?

Use AI freely for green tasks that are administrative and low stakes, use it with review for yellow tasks that shape what people see or hear, and keep red tasks, which decide a person’s job, pay, discipline or legal rights, in human hands. Place any new task by asking who is affected, how reversible the mistake is and whether the law regulates the decision.

I am Mac Malbrough. I hold a J.D. and an MS-HRM, I am not an attorney, and this is general information, not legal advice. Owners keep asking me where AI belongs in HR, and the honest answer is that the question is too big. AI is not one thing, and HR is not one thing. The useful question is task by task, and a map makes the answer visible to everyone on your team.

The problem this solves: People on your team already use AI at work, but nobody has said which tasks are fine, which need review and which must stay with a person, so one pasted complaint or medical note can become a legal problem.

How UHR solves it: The California employee handbook template ($149) gives you the policy language to attach this map to, the California 2027 Employer Readiness Kit ($39) sorts the new laws, and the AI Use and Disclosure Policy Kit ($29) gives you a disclosure standard.

The map is also defined in the Urban Glossary, alongside AI risk tiering and impact assessment.

Why does a map work better than a policy sentence?

Most AI policies say something like “use AI responsibly.” Nobody can act on that. A map gives a person a specific answer when they are about to paste something into a tool at four in the afternoon. They can find their task, see its color and know what to do.

A map also shows leadership that you are not against AI. Most of the map is green or yellow. That makes the red zone easier to defend, because you are not saying no to everything, only to the parts where a mistake would hurt a person or the business.

How do you place a task on the map?

Ask three questions in order.

1. Who is affected, and how much? A task that only changes how your team schedules its week is different from one that changes whether a person keeps a job.
2. How reversible is a mistake? A typo in a draft is fixed in a minute. A wrong termination date, a leaked medical detail or a rejected applicant is hard to undo.
3. Does the law regulate the decision? Hiring, pay, leave, accommodation, discipline and termination all carry legal rules in California. A tool that touches them is not only an efficiency choice.

If the answer to the third question is yes, the task starts at red and has to earn its way down through human review, testing and records. If the first two answers are low impact and easy to reverse, the task is green.

What goes in the green zone?

Green tasks are administrative, internal and low in consequence. A person still reads the result, but a mistake is cheap and visible.

• Formatting, proofreading and tightening your own drafts.
• Summarizing your own meeting notes, without confidential employee facts.
• Drafting a calendar invitation or an agenda.
• Turning a long policy into a one-page outline you will check yourself.
• Organizing a list of training options by topic.
• Drafting job post language from a manager’s intake notes, before you add the required pay scale.

What goes in the yellow zone?

Yellow tasks influence what employees, candidates or leaders see or believe. They are allowed, but each one needs a named reviewer and a rule about inputs.

• Draft announcements, policy updates and manager talking points.
• Summaries of survey comments or exit feedback.
• Interview question sets and scoring guides.
• Learning outlines and training materials.
• First-line employee questions answered only from your handbook, with a person reviewing every unanswered question.
• Scenario write-ups from headcount data, with finance checking the numbers.
• Performance review language drafted from a manager’s own notes, which the manager edits and owns.
• Offboarding checklists and paperwork summaries, which a person checks against the final pay rules.

The common safeguard is a reviewer who knows the subject and has the authority to change the result. A reviewer who can only click approve is not a safeguard.

What goes in the red zone?

Red tasks decide something about a person or involve a legal duty. AI may support a person in the background, but the person decides and is accountable.

• Who to interview, hire, promote or reject.
• Pay decisions, raises and pay equity findings.
• Discipline, performance improvement plans and termination decisions.
• Complaint intake and workplace investigation findings.
• Leave eligibility, accommodation decisions and anything involving a medical detail.
• Any statement of what the law requires in a specific situation.
• Anything involving an employee’s confidential information entered into a tool you have not approved.

California has rules for the first group in particular. The civil rights regulations on automated-decision systems treat a computational process that helps make an employment decision as something an employer must be able to defend against discrimination claims, and SB 947 adds attention to automated tools used in discipline and termination. See the civil rights rules for AI hiring tools and SB 947 and automated decisions in discipline and termination for the details. The practical result is the same: a record of who reviewed the result, what they changed and why.

What about tasks that move between zones?

Tasks change color depending on how you use them. Summarizing your own notes is green. Summarizing an employee’s complaint is red, because the summary may shape how the complaint is handled. A tool that drafts interview questions is yellow. A tool that ranks the candidates is red.

The easiest way to catch these shifts is to write the data the task uses next to its color. If the input is your own words, the task is probably green. If the input is an employee’s information, it moves toward red. If the output leads to a decision about a person, it is red no matter how simple the task looked.

What does a real week look like with the map in use?

Picture an HR lead on a Tuesday. She uses a tool to tighten a reminder email, which is green, and reads the result once. She asks it to outline a manager talking-points sheet for a schedule change, which is yellow, so she edits it and sends it to the owner for approval. A manager then asks her to paste an employee’s medical note into the tool to summarize it. That is red, so she declines, summarizes it herself and files the note where medical information belongs. The map did not slow her down. It gave her a fast, defensible answer three times in one day, and it gave the manager a reason that was not personal.

Teams that work this way tend to ask better questions too. When a task does not fit the map, they add a line instead of arguing, and the map grows with real use.

How do you put the map into your handbook and your habits?

A map on a shared drive does nothing. These four steps give it force.

1. Print the map on one page with the three colors and your task list.
2. Add a two-sentence rule to your handbook: AI may be used for green and yellow tasks as shown on the map, and red tasks stay with a person.
3. Name an owner who updates the map every quarter and when a new tool is adopted.
4. Ask every manager to sign that they have read it, and keep the record.

Review it at least quarterly. New tools change what is possible, and California law on AI in employment is moving quickly.

How does the UHR handbook template put the map into policy?

The problem for most small employers is that the map exists in someone’s head and the handbook says nothing. When an employee pastes a complaint into a free tool, there is no written rule to point to. The California employee handbook template ($149) gives you the annotated, editable policy base for confidentiality, technology use and employee relations, so your AI map attaches to language that already fits California. You edit it to match how your business really works, and every manager receives the same rule.

For the 2027 changes, the California 2027 Employer Readiness Kit: Law Sorter, Contract Rewrites, Supervisor Briefing and Calendar is $39, and the AI Use and Disclosure Policy Kit for Solo Consultants is $29 for a short disclosure standard you can adapt.

What should you do this week?

1. List the 15 HR tasks your team does most and place each on the map using the three questions.
2. Circle every task where an employee’s information goes into a tool and check that the tool is approved.
3. Write the two-sentence handbook rule and share it with your managers.
4. Put a quarterly review on the calendar with a named owner.

If you would like an outside review of your own map, the HR compliance check-up covers it, and fractional HR support can maintain it with you.

For a related angle, see the manager skill of AI judgment.

Sources

• California Civil Rights Department, automated decision system regulations
• California SB 947 (2026), read on the California Legislative Information site

McKinley holds a J.D. but is not a licensed attorney. Articles here are general information, not legal advice. For your specific situation, talk with an employment attorney.

Data current as of November 2026. Sources are linked where each figure appears.