AI playbook for the head of HR — recruiting, onboarding, culture

When a head of HR at a 140-person services firm told me her recruiters were "drowning in resumes but afraid to touch ChatGPT after legal said no" — I realized the AI playbook for HR isn't about tools. It's about which three plays you authorize first.

Why HR is the highest-leverage AI function in an SMB

HR touches every employee. A small change in how you screen, onboard, or communicate compounds across every team. That's why a translation-services company we studied moved screening from 3 hours to 3 minutes per role and lifted intake efficiency by 83% — not because they bought a fancy ATS, but because they put AI in front of the screening triage.

It's also why HR is the function most exposed to shadow AI. A widely-cited 2025 figure: 46% of employees have uploaded confidential data to public AI tools. In HR, that data is salaries, performance notes, and applicant resumes. The head of HR who doesn't have an authorized path is silently authorizing the unsafe one.

Definition: Shadow AI — employees using consumer AI tools (ChatGPT, Gemini, Claude personal) on work data, outside any sanctioned channel.

The 90-day HR AI playbook (six plays)

Recruiting — Play 1: structured-resume triage

The first hour you save in HR is at the top of the funnel. Build a triage prompt that takes a resume + a job spec and returns a structured rubric: must-haves matched, gaps, and 3 clarifying questions for the recruiter. No accept/reject. No score. The recruiter still decides.

You are a recruiting assistant. Compare this resume to the job spec.
Return JSON with:
- must_haves_matched: [list]
- must_haves_missing: [list]
- nice_to_haves_matched: [list]
- ambiguities: [3 questions a recruiter should ask]
DO NOT recommend hire/no-hire. DO NOT score the candidate.
RESUME: {{resume}}
JOB SPEC: {{spec}}

Why this shape: it removes the legal risk of automated decisioning while keeping the time saving. EU AI Act classifies fully-automated hiring as high-risk; advisory triage is not.

Recruiting — Play 2: scheduling + reference check drafts

The second recruiting play is unglamorous: AI drafts the scheduling emails, the reference-check questions, and the rejection notes. A senior recruiter reviews and sends. The compounding effect is roughly 5-7 hours per recruiter per week — and it's invisible to candidates because the human still owns the send.

Tool tip (Course for Business): In our 6-week program we put HR teams through the Shoulder-to-Shoulder hot seat: a recruiter sits with a coach for 90 minutes and ships their first sanctioned triage prompt for live roles. We use the Augment, don't replace rule — AI never sends a rejection or makes a hire/no-hire call alone. By week 2 most recruiters self-report 30-40% time saved on the same req volume. See course.aiadvisoryboard.me/business.

Onboarding — Play 3: the AI buddy

New hires ask 30-40 questions in their first two weeks: how do I get a laptop, where's the wiki, who owns benefits. A retrieval-grounded onboarding bot fed by your handbook and IT docs answers 70-80% of those without paging a human. The trick is grounding — never let it answer outside the indexed corpus.

Definition: RAG (retrieval-augmented generation) — the AI is forced to quote from your indexed documents and refuses if the answer isn't there.

Onboarding — Play 4: 30-day check-in synthesis

Have managers run a templated 30-day 1:1 with each new hire. Record (with consent) or take notes. AI summarizes themes across all 30-day check-ins this month: "What surprised new hires?", "Where did onboarding feel slow?", "Which teams need a better day-1 doc?". The output is a one-page memo for the head of HR and the CEO. This is where culture signal actually lives.

Culture — Play 5: pulse-survey synthesis (not score-watching)

Most culture dashboards die because nobody reads the bar charts. Switch the deliverable: AI clusters open-text answers into themes and surfaces the 5 strongest verbatims per theme. The head of HR walks the leadership team through verbatims, not scores. Adoption changes overnight.

Culture — Play 6: manager coaching prompts

Build a manager-only chatbot with three skills: (1) draft a hard feedback message, (2) prep for a difficult 1:1, (3) write a recognition note that doesn't sound generic. Most line managers in a 30-500-employee company never had formal management training. AI gives them a private rehearsal partner.

Team scan (what AI champions report after week 1)

• Recruiters self-report 30-40% time saved on resume triage with the structured-rubric prompt.
• 2-3 hiring-manager interviews per recruiter shifted from "screening" to "evaluating fit".
• New hires hit the AI buddy 8-15 times in week 1; HR ticket volume drops noticeably.
• 30-day check-in synthesis surfaces 1-2 onboarding gaps that nobody had escalated.
• Pulse verbatims get read by the exec team for the first time in 18 months.
• 1-2 managers admit they used the coaching bot before a hard conversation.
• Legal and IT confirm no PII has left the sanctioned tools (because there's a sanctioned path now).
• 1 unauthorized Slack-installed AI bot is found and decommissioned.
• 1 recruiter still bypasses the rubric — flagged for shoulder-to-shoulder coaching, not punishment.
• The head of HR has a one-page weekly digest she actually trusts.

Tool tip (Course for Business): The HR cohort in our program runs with a 1:15-20 AI Champions ratio — for a 140-person firm that means roughly 7-9 champions across HR, recruiting, and people-ops. Champions own one play each and run a 30-minute clinic per week for the rest of the team. We've found this outperforms top-down rollouts by a wide margin because line-level recruiters trust line-level peers. Five days to first automation, six weeks to durable habit. course.aiadvisoryboard.me/business.

Micro-case (what changes after 7-14 days)

A typical 180-person professional services firm runs the six plays in this order: triage prompt week 1, scheduling drafts week 1, AI buddy week 2, 30-day synthesis week 3, pulse verbatims week 4, manager coaching week 5. By day 14 the recruiting team has cleared a backlog that had been three weeks deep, the IT helpdesk gets fewer "where do I find…" tickets from new hires, and the head of HR walks into the monthly leadership meeting with verbatim quotes instead of a satisfaction score that nobody believed. The CEO's question shifts from "is HR using AI?" to "which other functions should run this playbook?".

Note on this case: This example is illustrative — based on typical patterns we observe with companies of 30-500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

FAQ

Should the head of HR own the AI policy for the whole company?

No. HR co-owns it with Legal, IT, and Security. But HR owns the rollout — because adoption is a people problem, not a tooling problem. If the AI policy lives only in a Confluence page nobody reads, it doesn't exist.

Is AI-driven resume screening legal?

Advisory triage (the rubric prompt above) is generally fine across the EU and most US states. Fully automated rejection is not — the EU AI Act and several US states (NYC Local Law 144, Illinois AIDA) require either human review, candidate disclosure, or both. Default to advisory.

How do we stop recruiters from pasting resumes into personal ChatGPT?

You give them a sanctioned tool that is faster and easier than the personal one, and you log usage. The 46%-shadow-AI figure drops sharply when there's a sanctioned path. Punishment doesn't move it; convenience does.

What about the AI buddy hallucinating policy?

Ground it (RAG) and force it to refuse on out-of-corpus questions. Add a one-line disclaimer: "This bot answers only from the handbook indexed on {date}. For exceptions, page #people-ops." We see fewer than 1 in 50 answers escalate when grounding is real.

Where does this overlap with our daily-management dashboard?

Briefly: the daily-management product surfaces what work happened across teams. The HR playbook above is what HR does inside their own workflow. They complement, but don't substitute.

Conclusion

The head of HR who runs all six plays in 90 days has done more for company-wide AI adoption than any tooling RFP would. Recruiting buys time, onboarding buys trust, culture buys signal. Start with Play 1 on Monday.

If you want every HR team member to ship their first AI automation in five days — book a 30-min call and we'll map your team's first week: course.aiadvisoryboard.me/business.