A 14-person company doesn't have an HR department. It has 100 hours a year of people work that the owner or office manager does on top of their day job — and the stakes per hour are higher than in any other role. The wins from Claude are real: better job descriptions, faster resume triage, performance reviews that actually reflect the person. The losses are louder than any other role too, because the failure mode is breaking trust with a human who works for you. This page is the careful version.
Block 60 minutes on the first Monday of every month. Open your People Playbook project. Pick one person on the team. Do the full review on them — performance trend, comp position, growth path, retention risk, the conversation you've been putting off. Twelve months, twelve people-reviews, no team member who goes a year without attention.
Small businesses don't lose people because of bad pay or bad management. They lose people because of inattention — the senior person nobody's had a real conversation with in a year, the new hire who got a great onboarding and then dissolved into the team, the high performer who silently took an outside offer because nobody asked them what they wanted next. The People Hour is the antidote. Once a month, one person, no surprises in either direction.
The first 30 minutes: walk this person's last 12 months with Claude — their trajectory, their comp position, what they've said in 1:1s, what you've been avoiding. The next 30: decide on the one thing to do this month. A comp adjustment, a growth conversation, a stretch assignment, a real review, an exit plan — or, sometimes, just a thank-you that hasn't been said specifically enough.
"Walk my roster. Who haven't I given real attention to in the last 90 days — no 1:1s skipped, no scope conversation, no recognition? Pick them. Walk me through what I should do this month."
"Who's most likely to leave in the next 6 months based on what I've shared in this project? Be specific: signals (overdue review, comp below market, scope drift, recent disengagement), and the conversation I should have now."
"What role do we structurally need in the next 6 months based on team composition, current load, and growth trajectory? Make the case. Then make the case against — what alternative to hiring should I consider?"
"From 'People things I'm avoiding' — what's the one conversation I should have this month? Walk me through how to start it on Monday. If it's an underperformance conversation, help me prepare without softening the message."
"Which team member is ready for something I haven't given them? Stretch project, scope expansion, mentorship, formal title change. Be specific about what they're ready for and what would happen if they did it well."
"What pattern am I seeing across the team that I haven't named yet? Overload signals, capability gaps repeating, comp drift, cultural drift, a meeting cadence that's broken. Don't tell me what to do — tell me what to look at."
Why this works. People decisions in small businesses get made under pressure, on Tuesday afternoons, in 7-minute slots between operational fires. The People Hour is the only hour each month that's reserved for considered people work — and the only reason most of your team will still be here in 18 months. Skip it for three months and you'll be surprised when someone resigns. Run it for twelve months and surprises mostly stop.
This is the project where the data is most sensitive and the value is most concentrated. The honesty of the "what I'm avoiding" list is what separates this from a polished HR brochure. The access controls on the data are what separates a useful project from a brewing incident. Build it carefully.
Download the template. Fill in honestly. Be deliberate about what goes into the project and what stays out of it:
Open Claude → Projects → New Project → "People Playbook." Paste the substrate. Upload the handbook and review templates. Verify your Claude account's data-residency and retention settings before adding anything personal-data-shaped.
The "people things I'm avoiding" section is the difference between a useful project and a polished one. Small-business owners avoid people conversations more reliably than they avoid anything else, because the cost of a bad people conversation feels personal in a way that a bad vendor conversation doesn't. Writing the list down gives Claude permission to surface what you'd rather not look at when you're working on adjacent things. That's the entire job. Without that section, this is a fancy job-description generator. With it, it's something closer to a co-pilot.
Used inside your People Playbook project, so Claude already has the team, the hiring profile, the comp philosophy, and the conversations you've been avoiding in context. Fill in the [bracketed] specifics and copy-paste.
I'm hiring for [ROLE]. Here's the voice memo / notes describing what I actually need: [PASTE OR ATTACH]. Draft the job description. Rules: 1. Pull from our hiring profile (project) — what we look for, what we don't optimize for. Don't drift into generic job-description language. 2. Match our voice. No "rockstars," no "ninjas," no "fast-paced environment" filler. 3. Be specific about the first 90 days — what this person will actually be doing, not abstractly "owning" anything. 4. The "what we don't optimize for" section — explicit. Anti-signals belong in the JD; they save everyone time. 5. Comp range — from our comp bands in the project. Posted publicly; we don't hide it. 6. The single question they should ask themselves before applying. After the draft, tell me: is this a hire I'm making to solve a real problem, or am I hiring to avoid a different harder problem (a difficult conversation, a process gap, a personal capacity issue)? Be willing to push back.
Attached: [N] resumes for the [ROLE] req. The JD and our hiring profile are in the project. Triage. For each resume: 1. **Match against our look-fors** (project) — specific signals present, specific signals absent 2. **Match against our anti-signals** — anything that suggests this person would be miserable here or short-tenure 3. **Confidence (1–10) on inviting to first conversation** 4. **The one question I'd want to ask them** that the resume doesn't answer 5. **Pattern from our past hires** — does this person resemble a past good hire or a past bad hire we made? Name the pattern. Output as a ranked list. Top tier (interview), middle tier (maybe — what would change your mind), bottom tier (decline politely). Critically: do not use protected-class signals (age, race, gender, national origin, religion, disability, family status, etc.) in any way. If a resume includes such information explicitly, ignore it. Score on capability and fit signals only. If you find yourself reasoning from a proxy for a protected class, flag it to me instead of using it.
Role: [ROLE]. JD attached. Our hiring profile is in the project.
Generate the interview question bank. Three sections:
1. **Capability questions** (6–8): specific to the role's actual work. Each should be a real situation we'd face, asked as "tell me about a time…" or "walk me through how you'd…". Not abstract.
2. **Culture/fit questions** (4–5): aligned to our values and our hiring profile. Avoid the generic trap-questions. The best fit questions surface judgment, not rehearsed answers.
3. **Reverse questions** (3): things we'd want THEM to ask us, that would signal real engagement. (If they don't ask any of these or anything similar, that's data.)
For each question: what a strong answer sounds like, what a concerning answer sounds like, what a disqualifying answer sounds like.
End with the one question I should ask near the end of the conversation that creates a moment of unscripted response. Different from "any questions for me?" Performance review for: [NAME, ROLE, TENURE]. Inputs I'm sharing: [1:1 notes, project outcomes, peer feedback, my own observations from the period — paste everything]. Draft the review using our framework (project). Rules: 1. Be specific. Generic reviews fail people — a sentence like "shows strong initiative" is useless without an example. Every claim needs a behavior. 2. Match what the year actually looked like. If they had a hard year, name it. If they had a quiet year, name that. The best reviews are the ones that describe a year the person recognizes. 3. The single question we always ask (project) — answer it honestly. 4. The growth conversation — what's next for them, in their interest first, ours second. 5. Things to say in the conversation but NOT in writing (project) — flag separately for me. Some things belong in dialogue, not on record. End with: the version of this review I would resent reading. Read it back and ask — would I respect the person who wrote this? If no, what's missing.
I need to have a hard performance conversation with [NAME] about [ISSUE]. Context: [How long has this been a pattern? What have I done about it already? What's the desired outcome — improvement, role change, exit?] Help me prepare: 1. The one sentence summary of the problem, said plainly. Strip out hedges. ("Your code reviews are slowing the team down by an average of 3 days" — not "I wanted to discuss your review cadence.") 2. The three specific examples I should cite — concrete, recent, observable behaviors (not interpretations). 3. What they'll likely say in response, and how I respond to each. 4. The line that defines the outcome I need — what improvement looks like, by when, measured how. 5. What I will and won't agree to. The concessions I shouldn't make. 6. The opening line. Critically: if this conversation looks like it might be heading toward termination, formal warning, or anything with legal/HR implications, tell me to call our employment lawyer before the conversation, and don't draft the language. Prep helps; legal-shaped wording needs a lawyer.
I have a 1:1 with [NAME] on [DATE]. They've been here [TENURE], and our last meaningful conversation was about [TOPIC + DATE]. Help me prepare questions specific to them — not a generic 1:1 template: 1. The growth question for someone at their tenure and trajectory (1-year tenure ≠ 4-year tenure) 2. The energy question — what they should be doing more of, what they should be doing less of, based on what I know about them 3. The "what would change my answer" question — if they were unhappy, what signal would I be looking for? Frame the question to elicit it without alarming them. 4. The recognition I owe them — something specific they did well that I haven't named. 5. The one thing I'm avoiding that's about them (project) — and how to bring it up if the conversation creates space for it. End with: should I push the 1:1 from 30 to 60 minutes for this one? Sometimes the answer is yes.
We've made a verbal offer to [NAME] for [ROLE], start date [DATE], comp [$X]. Draft three things: 1. **The offer letter** — using our standard structure. Warm, specific to them (not generic), with the comp and start date explicit. Note the standard accompanying docs they need to sign. Flag any clause that's non-standard for their state (you have the location in the project) — but do not draft legal language; that's for the lawyer / Docusign template. 2. **The first-week plan** — five days, what they should be doing each day. Day 1 should not be "fill out paperwork all day." Include who they meet, what they should accomplish by end of week 1, what success at day 30 looks like. 3. **The week-1 message to the team** — announcing the new hire. Warm but practical. Names, when they start, what they'll be working on, who's their "buddy" if we do that. End with the three things I should personally do during their first week — not delegable, only by me.
BEFORE USING THIS PROMPT: have you talked to the employment lawyer? If no, stop. Call them first. This prompt only helps you prepare for a termination the lawyer has already cleared on cause, documentation, and process. Termination: [NAME]. Effective date: [DATE]. Reason category (no-cause / performance / for-cause): [CATEGORY]. Lawyer has reviewed: [YES — name + date]. Help me prepare the non-legal parts of the conversation: 1. The opening sentence. Direct. Not soft-launched. Not "let's talk about your future here." 2. What I will and won't say. Things to acknowledge; things that belong in writing only (severance, COBRA, final paycheck, equipment — referenced in conversation, detailed in the packet). 3. The 5-minute conversation arc. Length-bounded — kindness here is brevity, not extended explanation. 4. The internal comms plan — who needs to know what, in what order, by when. Customer-facing comms if any. Team-facing comms. 5. The three things I should NOT do today, even if I feel tempted (re-litigate the decision, apologize for the decision, soften the message). The lawyer is responsible for: the termination letter language, severance terms, the release agreement, COBRA timing, final paycheck legal requirements. Do not draft any of those. If you find yourself drafting one of those, stop and tell me.
Exit conversation with [NAME], departing on [DATE]. Notes / recording / transcript: [PASTE]. Synthesize: 1. **What they said** — the stated reason for leaving, the specific things they named about culture / management / role / comp 2. **What they likely meant but didn't say plainly** — read between the lines, but flag this clearly as inference (vs. direct quote) 3. **Patterns to compare against** — does this match what other recent leavers have said? Cross-reference any prior exit syntheses in the project. 4. **The one criticism that's hardest to hear** — and most worth acting on 5. **What they appreciated** — not for ego; for retention of others. If multiple leavers have said the same thing was good about working here, that's a signal to keep doing. 6. **The action** — the one thing I should change, the one thing I should keep, and the conversation I should have with the remaining team in response. End with: should anyone on the current team get a check-in this week as a result of this exit? Specifically — does any current team member's situation rhyme with the departing person's? Name them.
Compensation conversation with: [NAME]. Current comp: [$X]. Tenure: [X]. Last comp change: [DATE + AMOUNT]. Context: [Why now? Their ask, your own initiative, market shift, role change?] Walk me through it: 1. **Where they sit in our band** (project) — middle, top, bottom. Where they should sit based on tenure + performance, your honest read. 2. **The market check** — what does a comparable role pay in our market in our location, based on what we know. Flag if data is thin. 3. **The case for the increase I'm considering** — the version I'd be comfortable defending to the whole team if it were posted. 4. **The case for a different number** — higher or lower than what I'm considering. Be willing to push. 5. **The non-comp elements** — title, scope, flexibility, equity, growth path. Sometimes the comp conversation is really one of these in disguise. 6. **The opening line of the conversation** — assuming I'm offering an increase. Not "we wanted to give you a raise" — something that names the why. Critically: if this is a counter to an outside offer, that's a different conversation. Tell me to walk through our policy on outside offers (project) before drafting anything.
It's the end of Q[N]. What I'm sharing: [1:1 notes from the quarter, any team-wide signals (Slack tone, meeting energy, request patterns), any departures or near-departures, anything from the team I've noticed]. Read the team sentiment honestly: 1. **The temperature now** vs. last quarter — better, worse, flat. Cite specifics. 2. **The energy patterns** — who's gaining energy, who's losing it. Be specific. 3. **The cohesion question** — is the team operating as a unit, or are there sub-groups forming / strain between people? Read the signals. 4. **The capacity question** — based on the volume of "I'm a bit slammed" comments vs. last quarter, is the team operating in a sustainable mode? 5. **The one structural issue** that's likely behind multiple individual signals — sometimes 4 individual complaints are 1 structural problem in disguise. 6. **The retention risk list** — who's most likely to leave in the next 2 quarters, ranked, with reasoning. End with: the one team-wide intervention worth considering this quarter (or the explicit recommendation NOT to intervene — sometimes the answer is to let things settle).
Situation: [DESCRIBE — what you're seeing, who's involved, what's been happening, what you're tempted to do about it]
Diagnose:
1. **Is this an HR issue or a management issue?** Most things owners and managers label "HR" are actually unmade management decisions waiting on a process to handle them. Be direct.
2. **If management:** what's the conversation I should have, with whom, when? Name it.
3. **If HR (genuinely policy-shaped or legal-shaped):** what's the right process — and at what point does this need a lawyer rather than a manager?
4. **The thing I'm tempted to do that would be wrong:** the workaround that feels easier today and harder in 6 months.
5. **The one question** I should ask the person in question before I do anything else, that would clarify whether this is what I think it is.
Push back if I'm trying to use process to avoid a hard conversation. That's the most common pattern here. Process is not a substitute for management; it's what management produces when it's done well. Prompt 12 is the one to use most. The single most common error in small-business "HR" is reaching for a policy or a workflow when the actual fix is a 15-minute conversation. Run this diagnostic before drafting any people-process intervention. The frequency with which the answer is "this is management, not HR" will surprise you — and the resulting conversations will be more honest than the processes you would have built instead.
The integration with the most measurable payback for small-business people work: when a candidate gives a verbal yes, one prompt produces the offer letter, the first-week plan, the team announcement, the Drive folder, the calendar invites, and the day-30 check-in schedule. What used to be 3–4 hours of clicking and copy-pasting becomes 20 minutes of reviewing.
In your People Playbook project → Connect apps → Google Workspace (Drive + Gmail + Calendar), Docusign. Read/write access on Drive (for creating the employee folder), Calendar (for scheduling onboarding meetings), and Docusign (for sending the offer letter and accompanying documents from your pre-built templates).
The offer letter template stays in Docusign — Claude doesn't draft legal language; it populates a template the lawyer has already reviewed. That separation matters and is the same one we drew in Prompt 08.
Inside the project, save this as "New hire — offer to day 30." Run it after the verbal yes.
We've made a verbal offer to [NAME] for [ROLE]. Start date [DATE]. Comp [$X]. Hiring manager: [me / NAME]. Their location: [STATE]. Produce the full new-hire package: **1. Offer letter (Docusign send)** Populate our standard offer letter template with: name, role, comp, start date, signing deadline (5 business days), reporting line. Flag any state-specific clauses I should verify with the lawyer for their location. **2. Drive folder + accompanying docs** Create /People/[NAME]/ with subfolders: 01_Offer, 02_Onboarding, 03_Reviews, 04_Comp_History. Drop offer letter copy into 01_Offer. **3. First-week plan (Day 1 — Day 5)** Using our standard onboarding playbook (project), customize for this role: - Day 1: setup + welcome + first specific task (NOT just paperwork all day) - Days 2–4: meet the team in this order (most important relationships first), three specific deliverables they should attempt by end-of-week-1 - Day 5: check-in with me + one social moment with the team **4. Calendar invites** Schedule: their Day 1 morning kickoff with me, their first 1:1 (Day 3), their day-30 check-in. Send invites to me, them, and their hiring manager. **5. Team announcement message** Draft for me to send to the team — warm but practical. Their name, start date, role, what they'll be working on, who their buddy is (recommend from team based on the project). Include one specific thing about them that signals they were chosen for fit, not just credentials. **6. Day-30 check-in agenda** Draft the questions I'll use 30 days in — using Prompt 06's structure (growth, energy, what would change my answer, recognition). Critically: do not send anything outbound yet. Reply with everything for my review. I'll approve before the offer goes out and before anything posts to the team.
Before live use, run this against your most recent successful hire. Three checks:
"Quick note: Maya starts with us on Monday, May 27th. She's joining as our Director of Revenue Operations — owning the reporting setup across HubSpot + NetSuite that's been on our list for the better part of a year. Maya came out of a 200-person logistics company where she built the same kind of reporting stack from scratch; we hired her because in the second interview she walked me through her actual NetSuite-to-HubSpot reconciliation model on a whiteboard, and I knew within ten minutes she'd seen the problem we have. Her buddy for the first month is Sarah. Day-1 lunch is on me — anyone free Monday at 12:30, come say hi."
When a verbal yes comes in: run the prompt, review the six outputs, approve, and watch the offer letter go out, the folder appear, the calendar fill in, the team announcement queue up. What used to be 3–4 hours of operational work over a few days becomes 20 minutes of review. More importantly, your new hire's first impression of your operation is one where the offer landed within hours of the verbal yes, the day-1 schedule was specific, and the team welcome was personal. That's how strong people stay through their probation period.
Keep yourself in the loop on the announcement send — the team announcement is your voice; it should pass through you before it goes out. The folder, the calendar invites, the offer letter from the template — those can fire on approval.
What this isn't. It isn't an autonomous hiring system. The legal documents come from templates the lawyer has reviewed; the team announcement comes from you, not the AI; the comp number was set by you. Claude assembles, drafts, and schedules. You decide. That's the right division of labor for the most sensitive role in this resource — and the only one where the autonomous version would damage trust faster than it would save time.
L4 in HR is where the trust risk is highest and the savings are smallest. Three sketches, with explicit limits on each — and one whole category of HR automation that this resource recommends against, full stop.
Every application that hits the form gets auto-scored against the JD using Prompt 02's logic. Posts a ranked daily list to you. Top-tier candidates get a same-day "we'd like to talk" reply (drafted, approved by you, sent through your normal hiring email). Middle-tier get a "thanks, we'll review carefully and reply within 5 business days" auto-reply. Bottom-tier get a polite decline draft for your approval.
The single most important control: bias auditing. Resume-screening AI has well-documented bias risks — proxies for protected classes can sneak into "fit" signals. Mandatory quarterly review: pull a sample of declined candidates, blind the AI's reasoning, and have a human read the resumes fresh. If the patterns diverge systematically by anything that correlates with a protected class, the hiring profile in the project is encoding something it shouldn't be, and the agent goes offline until it's fixed.
The Offer-to-Onboarding integration above, extended through day 90. The agent owns the schedule and the agenda for the 30-day, 60-day, and 90-day check-ins. Drafts the agenda based on what's happened in the new hire's first month (1:1 notes, manager observations, peer feedback if collected). After each check-in, drafts a brief synthesis for you: how's this person tracking against day-90 success, what surprised you, what to course-correct.
This is the L4 with the most genuine upside in HR. Onboarding done well is the single highest-leverage retention investment, and most small businesses do it poorly because the operational lift is high. The Orchestrator makes the lift small. Build it second.
The most aggressive HR L4 and the one most worth thinking hard about. Continuous reading of: anonymous monthly pulse survey responses, 1:1 notes (with explicit team consent), exit interview synthesis, public signals (Slack tone, meeting energy if you collect it). Outputs: monthly trend report on team sentiment, retention risk flagging, early-warning signals.
Two things make this build different from a marketing or sales pipeline:
Done well, this is a useful early-warning system for a small business that genuinely cares about its team and wants to act on signal earlier than annual reviews. Done poorly, it's a mechanism for breaking trust at scale.
Three categories where this entire resource recommends against autonomous AI use, even at L4:
These aren't risk-averse limits. They're correctly-placed limits. The cost of getting them wrong is not measured in hours saved.
If you're imagining any of the three L4 builds — pause. Spend two quarters at L2–L3 with the People Hour first. By the end of Q2 you'll know exactly which one fits your business, you'll have built genuine team trust in your AI use (because nothing employee-facing went out unreviewed), and you'll know which legal/HR specialists to involve before building. When you're at that point, this is the L4 category I'd most strongly recommend bringing in outside help on — second only to finance.
HR has the smallest dollar payback of any role in this resource and the largest trust-cost when it goes wrong. Each failure mode below has cost a real small business a real employee. Catch them in yourself.
Sending an AI-drafted message to an employee — review, comp change, recognition, hard conversation — without telling them and without your judgment-edit on the language. The moment they realize, trust evaporates and is very hard to rebuild.
The rule: every word that goes to an employee passes through you, in your voice, with your judgment. AI helps you prepare; it does not communicate with your team on your behalf.
The hiring profile encodes a pattern from your past hires — and your past hires share a demographic that correlates with a protected class. Now the AI ranks similar candidates higher and dissimilar ones lower, "objectively." You've built a discrimination machine without intending to.
Mandatory quarterly bias audit. Spot-check declined candidates. If divergence patterns correlate with anything protected, the hiring profile is wrong and the agent stops until you fix it. This is non-negotiable.
Termination wording. Accommodation requests. Harassment complaints. Severance terms. Non-compete enforcement. State-specific employment law. AI gives you a confident, plausible answer, you act on it, and 18 months later you're settling a claim that would have been a $500 lawyer conversation up front.
The prompts in this page enforce the line: tax-shaped questions get framed for the CPA (Finance), legal-shaped questions get framed for the lawyer (here). The framing helps. The lawyer still happens.
The AI-drafted review is structurally fine — sections complete, scores tidy, language professional. And it could be about anyone. The employee reads it and knows: nobody actually thought hard about my year. Trust in the process drops; trust in you drops a little too.
The standard from Prompt 04: every claim has a behavior. Every section is recognizable to the specific human reading it. If you wouldn't want to receive this review, don't deliver it.
You don't want to have a hard conversation, so you build a process. Now there's a 90-day improvement plan with biweekly check-ins and a documentation rubric. You've turned a 15-minute conversation into 90 days of administrative theater, and the underlying issue is still unresolved.
Prompt 12 exists for exactly this. The diagnostic is "is this an HR issue or a management issue?" Most of the time the honest answer is "management" — meaning a conversation, not a workflow.
The pulse survey is anonymous, except the AI can identify the author from word choice within three weeks. You read the patterns, you act on them, and your team eventually figures it out. The cost is not a single employee's trust — it's the whole team's, simultaneously, and there is no path back from this one.
Either the anonymity is technically and procedurally real, or you don't promise anonymity. There is no middle ground that survives contact with a moderately observant team.
Download the template. Fill in the things you've been avoiding. Save the prompts. Wire offer-to-onboarding. Run the People Hour. In a year, your team will have had more deliberate attention than most small-business teams ever get — and you'll have built that attention without ever letting AI replace the part of the job that has to come from a human.