L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 6 · Lesson 1

The Resume Stack Nobody Has Time For

AI-powered applicant screening — what it actually does, and why the shortcuts carry real costs.
When you're on both sides of the hiring table, what do you want the algorithm to see?

Destiny is 21, running a growing alterations and custom-clothing business out of a shared studio space in Atlanta. She built the client base herself, handles her own bookings, and has finally hit a wall: she needs a part-time front-desk person. She posts on Indeed. Within 48 hours she has 74 applications.

She doesn't have 74 hours. She barely has four. So she does what a lot of small-business owners her age do — she Googles "AI resume screener free" and finds a tool that promises to rank candidates by fit. She uploads the job description. She uploads the resumes. The tool gives her a ranked list. She interviews the top six.

Two weeks later, after a hire that doesn't stick, a friend asks her: "Did you notice the tool filtered out everyone who went to community college?" Destiny hadn't noticed. She hadn't been shown that. The algorithm had done its work quietly, and she'd trusted it without asking what it was actually measuring.

What AI Screening Actually Does

Resume screening tools — from basic keyword matchers to more sophisticated machine-learning rankers — do one thing at the core: they compare text in resumes against text in your job description, and score similarity. Some go further, pulling in signals from work history patterns, school names, or even writing style. The pitch is efficiency. The reality is that efficiency and fairness are not automatically the same thing.

When you feed a tool your job description, you are essentially telling it: here is what good looks like. If your description says "2 years experience," the tool will systematically down-rank recent graduates — including people who are fully capable of the role. If it uses historical hiring data from similar businesses to calibrate scores, it inherits whatever biases were baked into those decisions.

For a small business owner who is also often a young person of color, a woman, or someone without a traditional business background — this matters personally. Because you know from the job applicant side that credentials don't always tell the story of competence.

Real Friction

A 2023 Harvard Business School report found that algorithmic screeners rejected millions of "hidden workers" — people with gaps, non-linear paths, or community college credentials — who human reviewers would have seriously considered. Small business owners often explicitly want candidates with hustle, not pedigree. But generic AI tools are often calibrated for corporate hiring norms.

The Tools That Exist (And What They're Good For)

There's a wide range here, and it's worth being specific about what's available at the scale most small businesses operate at:

Keyword-based ATS filters (like built-in tools in Indeed or Workable) are the most common. They look for terms from your job post in the resume. Simple, gameable, and reasonably transparent — if you know what keywords you're requiring, you control what gets filtered.

AI ranking tools (like Manatal, Fetcher, or Findem) use ML to score candidates holistically. More powerful, but also more opaque. These are where bias risk escalates because the scoring logic isn't always visible to you.

Conversational pre-screening bots (like those in Greenhouse or HireVue's text-screening feature) conduct automated first-round Q&A via text or video, then score responses. These have the highest stakes — because the AI is now evaluating personality, not just credentials.

For a 20-person or under business, the honest answer is: keyword filtering with a clear, thoughtful job description is probably the right tier. Sophisticated AI ranking tools built for enterprise hiring introduce complexity — and liability — that you may not be ready to audit.

ATSApplicant Tracking System — software that collects, stores, and filters job applications. Even small-business tools on Indeed or Workable have basic ATS functionality built in.
Proxy VariableA feature an algorithm uses as a stand-in for something else — like using "years of experience" as a proxy for skill, or school name as a proxy for intelligence. These often encode bias without anyone choosing to encode it.
The Practical Move: Write the Job Description for the Candidate You Actually Want

The single highest-leverage action you can take before touching any AI tool is rewriting your job description to reflect the actual work, not the default credential checklist. Most JDs are copy-pasted from templates and include requirements that don't predict job performance — they just predict what kind of background someone comes from.

Ask yourself: What will this person do on day one? Day thirty? What problems will they face? Then write requirements around those answers. "Ability to handle a full booking schedule and communicate professionally with clients" is more honest than "2 years administrative experience" — and it opens your pool to people who have the actual skills without the formal title history.

If you're using AI to screen, the job description is your main dial. You control what the algorithm values by controlling what the description emphasizes. That's the leverage point most people skip over while debating which tool to use.

What Peers Get Wrong

A lot of people our age treat AI screening like it removes the bias problem — because it's "objective." It doesn't. It moves the bias upstream into the training data, the job description, and the tool's design choices. The fairness question doesn't disappear; it just gets harder to see. You still have to ask it.

Transparency and the Legal Reality

This isn't just an ethics conversation — it's also a legal one. As of 2024, New York City's Local Law 144 requires employers using automated employment decision tools to conduct bias audits and notify candidates. Illinois and California have their own AI-in-hiring disclosure rules. These regulations are young and expanding. If you're using a third-party tool that makes any consequential decision in your hiring process, you want to know: has this tool been audited? Can I explain to a rejected candidate why they were filtered out?

For most small businesses, the legal exposure is currently lower than for large employers — but the standard of "could you explain this decision to the person affected" is a reasonable personal bar to hold yourself to regardless of legal requirements. It's also just how you'd want to be treated when you're the one sending in the application.

Lesson 1 Quiz

AI screening, bias, and the job description as your real control panel.
1. At its core, most AI resume screening tools compare text in resumes against text in your job description and score what?
Right. The core mechanic is text similarity matching. Everything else — ranking, scoring — is built on top of that.
Not quite. The core function is comparing text similarity between the resume and job description. Additional features layer on top of that.
2. Destiny's screening tool filtered out community college applicants without her realizing it. This is an example of what concept?
Exactly. School type was being used as a stand-in (proxy) for some measure of "quality" — without Destiny choosing that or even knowing it was happening.
This is a proxy variable problem — the algorithm used school type as a stand-in for candidate quality without Destiny ever choosing that criterion. That's different from intentional discrimination.
3. You're a 22-year-old opening a small photography studio and need a part-time booking coordinator. Which job description requirement is most likely to unfairly narrow your applicant pool?
Right. A bachelor's degree requirement for a booking coordinator role is a credential proxy that doesn't predict performance — and it immediately narrows your pool based on background, not ability.
Look at which requirement is credential-based rather than skill-based. The degree requirement has no demonstrated connection to booking-coordinator performance and filters by background, not ability.
4. Which AI hiring tool carries the highest bias risk for a small business owner, and why?
Correct. Conversational AI screeners evaluate communication style, tone, and demeanor — signals that have significant cultural and socioeconomic variation. And you usually can't see the scoring rubric.
Conversational AI screeners are the highest-risk tier because they evaluate subjective personality signals — things with real cultural bias risk — using criteria you can't inspect or audit.
5. As of 2024, which city requires employers to conduct bias audits on automated employment decision tools and notify candidates?
Right. NYC Local Law 144 was among the first laws of its kind, requiring bias audits and candidate notification for automated hiring tools.
New York City's Local Law 144 is the relevant regulation here — one of the first laws requiring bias audits and candidate disclosure for AI hiring tools.

Lab 1: Audit the Job Description

You're the consultant. Your client is about to run a biased screen and doesn't know it.

Your Role

A friend who runs a small bakery café is about to post a job for a "Café Coordinator" — someone to handle scheduling, supplier calls, and social media. She copied a job description from a template. It requires a "Bachelor's degree in business or hospitality" and "3+ years of administrative experience." She's about to paste it into an AI screener.

You're playing the role of a consultant who knows how AI screening works. The AI in this chat is playing a knowledgeable peer who will push back on weak reasoning — not just agree with you.

Start by telling the AI what specific changes you'd make to the job description before it goes into any screening tool — and why. Be specific. Then we'll work through how the AI screener would behave differently with your revised version.
HR Audit Lab
AI Peer · Lesson 1
Alright — I've seen the job description. "Bachelor's degree in business or hospitality, 3+ years administrative experience." Classic template paste. Before you tell me what to change: what's your actual theory of what this role requires? If you were doing this job tomorrow, what would day one look like? Start there and then tell me what you'd cut or rewrite.
Module 6 · Lesson 2

Writing Job Posts That AI Won't Wreck

How to use AI to draft job descriptions that attract real candidates — without accidentally encoding the wrong signals.
What does "good fit" actually mean, and who gets to define it?

Marcus, 23, is three years into running a mobile auto-detailing business in Charlotte. He needs a second detailer. He's heard that tools like ChatGPT can write job descriptions fast, so he opens a chat window and types: "Write me a job description for an auto detailer."

What comes back is slick — professional-sounding, well-formatted. It asks for "strong communication skills," "a proven track record," "attention to detail in a fast-paced environment." Marcus posts it immediately. The applications he gets are almost entirely people who've worked at dealerships or corporate detail shops. Meanwhile, the person he eventually hires — someone from his neighborhood who'd been detailing cars for fun and side cash since high school — almost didn't apply because the listing felt like it wasn't talking to him.

The AI gave Marcus the average of what job descriptions look like, not what his actual business needed. That's the trap.

AI as a Writing Tool, Not a Thinking Tool

AI writing assistants — ChatGPT, Claude, Gemini — are legitimately useful for drafting job descriptions. They can help you structure the document, suggest language, and catch things you forgot to include. But they default to the center of the distribution: the average job posting for that category, weighted toward corporate norms and formal credential language.

The fix is to do your thinking before you prompt. If you hand AI a blank-canvas request, it fills in its assumptions. If you hand it a detailed brief, it helps you build what you actually need. The difference in output quality is significant.

Effective prompting for job descriptions means front-loading the context: what's the business, what's the actual day-to-day work, what's the culture like, what matters more than credentials, and who specifically are you trying not to exclude. That last one is worth saying explicitly in the prompt — AI tools respond to it.

Prompt Comparison

Weak prompt: "Write a job description for a part-time cashier at my food truck."

Strong prompt: "Write a job description for a part-time cashier at my Caribbean food truck in Houston. The real job is: take orders during lunch and dinner rush, handle a Square POS, keep the line moving, stay friendly under pressure. I want to attract people who are used to working fast and communicating with customers, but I don't need previous restaurant experience — someone who's worked retail, done door-to-door, or just has natural people skills would be great. Please don't include college degree requirements or corporate buzzwords like 'proven track record.' Keep it real and direct."

The Language of Inclusion (and Exclusion)

Research on job description language shows consistent patterns: masculine-coded words ("competitive," "dominant," "aggressive") correlate with fewer women applying. Credential inflation (requiring degrees for jobs that don't need them) disproportionately screens out Black and Latino candidates. Long lists of requirements cause women and some minorities to self-select out if they don't meet every single one, while other demographics apply anyway.

Tools like Textio and Gender Decoder can scan job descriptions for coded language and flag patterns before you post. They're not perfect, but they give you a second set of eyes when your own perspective has obvious blind spots. Even running your draft through a general AI with the prompt "flag any language in this job description that might inadvertently discourage qualified candidates from non-traditional backgrounds" can surface useful friction.

The practical move: draft with AI, then audit the draft. Don't treat the first AI output as final. That extra step is where bias gets caught before it costs you a hire.

Credential InflationAdding degree or years-of-experience requirements that aren't actually predictive of job performance — usually copied from templates, not based on the role's real demands.
Masculine CodingLanguage in job postings associated with stereotypically masculine traits — "competitive," "assertive," "dominant" — shown in research to reduce applications from women without improving hire quality.
Salary Transparency: Why AI Should Help You Name the Number

One of the highest-impact changes you can make in a job posting — and one that AI can help you research — is including the actual pay range. Studies consistently show that salary transparency reduces negotiation gaps that disproportionately affect women and people of color. It also reduces wasted time for everyone: candidates don't apply for jobs that can't pay them, and you don't have to manage uncomfortable salary conversations at the offer stage.

As of 2024, Colorado, California, New York, Washington, and several other states legally require salary ranges in job postings. Even if your state doesn't, publishing the range signals that you run a business with clear communication — which is exactly the culture good candidates are looking for.

Ask AI to help you research typical pay ranges for your role and location. Use something like: "What's the typical hourly rate for a part-time café coordinator in Nashville, Tennessee in 2024?" Cross-reference with Indeed Salary Tool or the Bureau of Labor Statistics. Then put the number in the posting.

Peer Reality Check

A lot of small business owners our age are posting AI-generated job descriptions verbatim — copying whatever ChatGPT drafts without editing the credential language or reading it from the applicant's perspective. The result is job postings that look professional but feel impersonal and gatekeeping. Your competitive advantage as a young business owner is being the employer who sounds like a real human who actually thought about who they're looking for.

Putting It Together: A Job Description Workflow

Here's a repeatable sequence that takes about 30 minutes and produces a better result than any template:

Step 1: Write a bullet list of what the person will actually do — not job titles or generic responsibilities. Real tasks. Specific situations.

Step 2: Write a separate bullet list of what skills, traits, or experiences predict success in those tasks. Be honest about which ones are must-haves vs. nice-to-haves.

Step 3: Feed both lists to an AI with explicit instructions about tone, credential language to avoid, and who you're trying to reach.

Step 4: Read the AI output from the perspective of someone from a different background than yours. Would your ideal candidate recognize themselves in it?

Step 5: Audit for coded language using Textio, Gender Decoder, or a second AI prompt asking it to flag potential exclusion signals.

Step 6: Add the pay range. Always.

Lesson 2 Quiz

Job description writing, AI prompting, and the language of inclusion.
1. Marcus got generic job description output from ChatGPT because he gave it a generic prompt. What does AI default to when it receives a blank-canvas request?
Exactly. AI reflects the distribution of its training data — which skews toward formal, corporate, credential-heavy language. You have to explicitly push against that.
AI defaults to the center of the distribution in its training data — which means formal, corporate, credential-heavy language. That's not a flaw, it's just how these models work. You have to prompt against it deliberately.
2. Which addition to a job posting prompt is most likely to produce a more inclusive and accurate result?
Right. Specificity about real tasks and explicit openness signals are the two highest-leverage things you can add to a job description prompt.
The most impactful additions are specificity about real work and explicit statements about the backgrounds you want to remain open to. "Professional tone" and "standard templates" push the output toward the exact corporate defaults you're trying to avoid.
3. Research shows that words like "competitive," "dominant," and "aggressive" in job postings correlate with what outcome?
Correct. This is well-documented in hiring research — masculine-coded language measurably reduces the applicant pool by discouraging women from applying, without improving the quality of hires.
Masculine-coded language ("competitive," "dominant," "aggressive") correlates with fewer women applying. This is one of the most consistently replicated findings in hiring research.
4. You run a small landscaping business in Colorado. You want to post a job without including a salary range to keep negotiating flexibility. What's the problem with this in 2024?
Right. Colorado was one of the first states to mandate salary transparency in postings. Keeping the range off the listing isn't a legal option there.
Colorado is one of several states (along with California, New York, Washington) that legally requires salary ranges in job postings. Omitting it isn't a legal strategy — it's a compliance violation.
5. After drafting a job description with AI, what is the most important additional step before posting it?
Exactly. The perspective shift — reading as the candidate you want, not as the employer — is where most bias gets caught before it does damage.
The critical step is reading it from your ideal candidate's perspective and auditing for language that might discourage them. Adding more requirements or grammar-checking doesn't address the inclusion problem.

Lab 2: Rewrite the Posting

Draft a real job description using the workflow — then defend your choices.

Your Role

You're opening a small vintage clothing resale shop in a college town. You need a part-time floor associate — someone who can help customers, process inventory, and manage the shop's Instagram. You're drafting the job posting from scratch using the lesson's 6-step workflow.

Share your job description draft with the AI. It will give you direct, honest feedback — flagging credential inflation, exclusionary language, and missing salary transparency. You'll need to defend your choices or revise them on the spot.

Draft a job description for the vintage shop floor associate role and paste it here. Use what you learned in Lesson 2 — real tasks, honest requirements, no credential inflation, salary included. Then explain the specific choices you made.
Job Description Workshop
AI Peer · Lesson 2
Go ahead and share your draft. I'm looking for three things specifically: (1) Are the requirements actually connected to the real work? (2) Is there any language that would quietly discourage someone from a non-traditional background? (3) Is there a salary range? Don't explain what you intended — let the text speak first, then we'll talk through it.
Module 6 · Lesson 3

Onboarding, Scheduling, and the HR Stack

Once someone says yes, AI can help you not drop the ball — without becoming a surveillance machine.
How do you build a system that supports your employees without treating them like data points?

Priya is 22 and runs a small catering operation she started out of her apartment two years ago. She now has four part-time employees and is drowning in the logistics of managing them — shift scheduling, tracking hours for payroll, making sure everyone has the right food-handler certifications, and remembering who's available when. She's doing all of this in a combination of group texts and a Google Sheet that she's afraid to look at.

A friend recommends she try a platform called Homebase. Within an afternoon, Priya has everyone's availability in one place, can send a shift schedule with one tap, and the system automatically flags when someone's working more hours than they're contracted for. She's surprised by how much of her mental load just… moved out of her head.

But a week later, one of her employees — Devon, 19 — quietly asks her: "Does that app track where I am on my phone?" Priya doesn't actually know. She'd signed up without reading the data permissions. It's a fair question, and it's one she should be able to answer.

The HR Stack for Small Business: What's Actually Useful

There's a cluster of AI-assisted tools that genuinely solve real problems for small business owners managing a handful of employees. The honest framing is: they're useful, they create efficiency, and they also collect data about your employees that you have a responsibility to understand.

Scheduling tools (Homebase, Deputy, When I Work) use AI to optimize shift schedules based on availability, labor cost targets, and demand patterns. For a business with variable hours — a restaurant, a retail shop, a mobile service — these are legitimate time savers. They can automatically account for overtime rules, required rest periods between shifts, and minimum staffing requirements.

Onboarding tools (Rippling, Gusto, BambooHR at the small end) automate the paperwork side of bringing someone on — I-9 verification, W-4 collection, direct deposit setup, and benefits enrollment. They also track training completion and certifications, which matters in industries like food service, childcare, or construction where regulatory requirements are real.

Payroll and time-tracking integrations are where AI adds the most obvious value: eliminating manual entry, catching discrepancies, and automating tax filings. Getting payroll wrong is expensive both financially and in terms of employee trust. An automated system with good data is more reliable than a spreadsheet maintained by an owner who's also doing ten other things.

The Honest Trade-off

These platforms are genuinely useful. They're also subscription costs ($20–$100+/month depending on team size and features) that add up. And they collect significant data about your employees' work patterns, hours, and behavior. You should know what data is collected, how it's stored, and whether it's sold to third parties — not just for legal compliance, but because your employees are people who trusted you with a job.

Employee Monitoring: Where Useful Becomes Surveillance

Here's where we need to draw a real line. There's a significant difference between tools that help coordinate a team and tools that surveil workers. The line isn't always obvious in marketing copy — "productivity tracking" can mean anything from a time-clock app to software that takes screenshots every five minutes and scores keystrokes.

GPS tracking is the most common friction point for small business owners with field workers — delivery drivers, cleaners, landscapers, mobile service providers. There are legitimate use cases: fleet routing, verifying job completions, safety. But GPS tracking that runs 24/7, including off-shift hours, is a legal problem in several states and a trust problem everywhere. The rule of thumb that holds up legally and ethically: track the work, not the person.

For office or shop workers, productivity monitoring tools that log activity, take screenshots, or score "engagement" are increasingly common post-pandemic. Research consistently shows they damage trust without reliably improving performance. For a small business that depends on a small team doing good work, the trade-off almost never makes sense. You'd know immediately if someone isn't showing up or doing their job — you don't need an algorithm to tell you that.

GPS GeofencingSetting a virtual boundary around a location. When an employee's phone enters or exits the boundary, the system logs it. Useful for verifying job site arrivals; problematic when running continuously outside work hours.
Passive MonitoringData collection that happens in the background without active employee input — screenshots, keystroke logging, email scanning. Often disclosed in employment agreements but rarely discussed directly.
AI-Assisted Onboarding: Making the First Week Not Terrible

Beyond paperwork, AI tools can help you build a new employee's first week in a way that actually sets them up for success. Most small businesses have a chaotic onboarding experience — not because the owner doesn't care, but because they've never had time to document how things actually work.

AI can help you create onboarding documents and checklists from a rough description of what a new person needs to know. You can prompt something like: "Write a first-week onboarding checklist for a new barista at a small specialty coffee shop. Include training milestones, key policies they need to know, and who to contact for different types of questions." Then customize the output for your actual business.

Tools like Notion AI or Confluence let you build a simple internal wiki — a home base where new employees can find the answers to the questions they'd otherwise interrupt you to ask. For a tiny team, this feels like overkill until the third time you're explaining the same thing to a second new hire. At that point, documentation pays for itself immediately.

The bigger point is that good onboarding is one of the highest-ROI investments a small business can make. Research from the Society for Human Resource Management puts the average cost of employee turnover at 50–200% of annual salary. If you lose someone in the first 90 days because they felt lost and unsupported, you're paying that cost. An AI-assisted onboarding system that costs you two hours to build can prevent that.

What Peers Are Navigating

If you've ever started a job where you spent the first week not knowing where anything was, who to ask, or what the actual expectations were — that's what bad onboarding feels like from the inside. As a young business owner, you have a specific opportunity here: you remember how that felt recently enough to build something better. That's a real competitive edge in hiring and retention.

Lesson 3 Quiz

HR tools, onboarding, monitoring, and where the line is.
1. Devon asked Priya whether the scheduling app tracked his location on his phone. What was Priya's fundamental mistake before this conversation?
Right. As an employer using a platform that collects data about your employees, you have a responsibility to understand what's collected — before you ask employees to use it, not after they ask you about it.
The core failure was not reading the data permissions before deploying the tool. As the employer, you're responsible for understanding what data is collected about your employees — not discovering it when they ask you.
2. What is the most ethically defensible rule of thumb for GPS tracking of field employees?
Correct. "Track the work, not the person" is the standard that holds up legally and ethically. GPS that runs off-shift is a legal issue in many states and a serious trust violation regardless.
"Track the work, not the person" is the defensible standard. That means GPS can be active during job hours for legitimate work coordination — fleet routing, job site verification — but should not run during off-shift hours.
3. Research consistently shows that employee productivity monitoring tools (screenshots, keystroke logging) have what effect on small teams?
That's the consistent finding. Trust damage is real and compounding; performance improvement is inconsistent and often illusory. For a small team, this trade-off almost never makes sense.
Research consistently shows surveillance monitoring damages trust without reliably improving performance. For small teams where relationships are everything, this is a particularly bad trade-off.
4. You run a small cleaning service with four employees. Which tool would give you the most legitimate operational value with the lowest surveillance risk?
Right. Geofenced clock-in tracks the work — verifying job site arrivals and departures — without continuous surveillance. That's the right tier for a cleaning service.
Geofenced clock-in at job sites tracks the actual work without continuous surveillance. It solves the legitimate problem (verifying job completions and hours) without the trust cost of 24/7 tracking or invasive monitoring.
5. The Society for Human Resource Management estimates employee turnover costs how much relative to annual salary?
Correct. The 50–200% range is the widely cited SHRM figure — which is why investing in good onboarding and retention is not a soft metric. It's a real cost-avoidance calculation.
SHRM estimates 50–200% of annual salary — a wide range, but even the low end makes a strong case for investing in onboarding and retention rather than cycling through new hires.

Lab 3: Build the Onboarding Checklist

Turn a chaotic first week into a system — using AI as your draft partner.

Your Role

You run a small food truck that does lunch service five days a week in a downtown business district. You've just hired your second employee — a 20-year-old who's never worked in food service. Their first day is Monday. You have no formal onboarding document — just a lot of knowledge in your head.

Your job is to work with the AI to build a first-week onboarding checklist. You'll describe your business and what the new hire needs to know. The AI will help you structure it — but will also challenge you when you're missing things a new employee would actually need or when you're including things that belong in a policy manual instead.

Start by telling the AI what your food truck serves, where you operate, and what the new employee will be doing in their first three days. Be specific — the AI needs real details to give you useful output.
Onboarding Builder
AI Peer · Lesson 3
Before we build the checklist — I want to ask you something most owners skip: what's the single most common reason a new employee fails in their first 30 days on a food truck? If you can answer that, I can help you build an onboarding system that actually addresses the real risk, not just the paperwork. Tell me about your operation and what that failure mode usually looks like.
Module 6 · Lesson 4

Performance Reviews, Feedback, and Letting People Go

The most human parts of managing people — and where AI can help without replacing your judgment.
Can you automate compassion? And should you try?

Jordan, 24, manages a small team of four at his screen-printing business in Columbus. One of his employees — Amara, 20 — has been consistently late for three weeks. Jordan likes her. She does good work when she's there. He's been avoiding the conversation because he doesn't know how to have it without it feeling like an attack.

A business mentor suggests he try drafting the feedback conversation using AI — not to script the whole thing, but to help him organize what he actually wants to say. He opens ChatGPT and types out the situation honestly: the pattern, his relationship with the employee, his goal (fix the lateness, keep the person). The AI helps him structure a framework for the conversation: lead with the impact, not the judgment. Ask before assuming. Be specific about what needs to change and by when.

Jordan reads the draft and realizes something: he'd been avoiding the conversation not because he didn't know what to say, but because he'd been framing it in his head as "confronting" Amara rather than "supporting" her. The AI didn't have that insight — but the process of writing it out surfaced it. He has the conversation the next morning. It goes well. Amara had been dealing with a transit issue and just needed to shift her start time by 20 minutes.

What AI Actually Does Well in Performance Management

Jordan's experience points at the real use case: AI as a thinking partner and draft generator, not as the decision-maker. The situations where AI adds genuine value in performance management are:

Structuring feedback conversations. Most managers — especially first-time ones — struggle not with knowing what the problem is, but with figuring out how to say it clearly without it feeling like an ambush. AI can help you organize: what's the specific behavior, what's the impact, what's the ask. The "SBI" model (Situation, Behavior, Impact) and similar frameworks are easy to prompt an AI to apply to your specific scenario.

Writing performance review documentation. Consistent, written performance documentation protects both the employee and the employer. AI can help you draft reviews that are specific and behavioral rather than vague and judgmental — "missed three opening shifts in October" rather than "has an attitude problem." That specificity matters legally if a situation escalates, and it matters for the employee's ability to understand and respond to the feedback.

Generating talking points for hard conversations. Terminations, final warnings, PIPs (performance improvement plans) — these have standard structures for good reason. AI can help you draft the structure. You bring the human context and judgment about whether you've reached the right decision.

What AI Cannot Do Here

AI cannot tell you whether someone deserves to keep their job. It cannot weigh the full context of a person's situation, the history of your relationship, what their performance actually means for your business, or what a fair outcome looks like. It can help you prepare for a human conversation. The conversation itself is yours to have.

Performance Improvement Plans: Structure Without Bureaucracy

A PIP sounds like something for a big corporation, but the underlying logic applies at any scale: if an employee is underperforming, they deserve a clear documented statement of what needs to change, a specific timeline, what support you'll provide, and what happens if things don't improve. That's fair. It also protects you legally if the situation eventually leads to termination.

AI can generate a solid PIP template from a description of the role, the performance issue, and your expectations. The key is making sure the improvement goals are specific and measurable — not "improve your attitude" but "respond to all customer messages within 24 hours" or "complete opening procedures without reminders three times per week." Vague PIPs are useless for both parties.

One thing worth being honest about: for a small business with one or two employees, a formal PIP might feel unnecessarily bureaucratic. A clear conversation with a written follow-up email summarizing what was discussed often serves the same function. The documentation principle is what matters — the format can scale to fit your size.

SBI FrameworkSituation, Behavior, Impact — a structure for delivering feedback. Describe the specific situation, the observable behavior, and the impact it had. Keeps feedback concrete and avoids character judgments.
PIPPerformance Improvement Plan — a documented agreement specifying what needs to change, by when, with what support, and with what consequences. Standard in large companies; scalable for small businesses.
Termination: The Conversation AI Can Help You Prepare For

Letting someone go is the hardest HR action most managers face. It's the point where having your thinking organized matters most — because you won't be at your clearest when you're in the room. AI can help you prepare in several concrete ways:

Drafting a termination script. Not a word-for-word performance, but a structure: what you'll say first, how you'll explain the decision, what the logistics are (last day, pay, return of equipment), and how you'll close. Having this in writing helps you stay on message when the conversation gets emotional.

Reviewing documentation for completeness. If the termination follows a performance issue, AI can help you review whether you have adequate documentation — written warnings, PIP completion status, specific incident records — before you have the conversation. This matters for your legal protection, especially in states where wrongful termination claims are more common.

Anticipating questions. You can ask an AI to generate the questions an employee is likely to ask during a termination conversation — about unemployment eligibility, references, final pay timing, benefits continuation — so you have answers ready. Being unprepared for those questions makes the conversation harder for everyone.

What AI cannot tell you: whether the termination is the right call. That judgment is yours. Is this a performance issue that could be fixed with different support? Is this about fit, not ability? Is there something in this person's situation you haven't fully considered? Those are human questions about a person's livelihood. Take them seriously before the conversation, not after.

The Real Peer-Level Reality

A lot of young managers avoid feedback and termination conversations because they feel the emotional weight of them — and that's actually appropriate. The discomfort means you're taking the stakes seriously. What AI helps with is the preparation side: organizing your thinking, structuring what you want to say, making sure you're not missing something important. That preparation makes the conversation shorter and clearer, which is better for both of you. The discomfort doesn't go away, and it probably shouldn't.

Building a Feedback Culture Before You Need It

The best time to build feedback norms on a small team is before something goes wrong. Regular, low-stakes check-ins — even informal ones — make the harder conversations easier because feedback stops feeling like an event and starts feeling like a normal part of working together.

AI tools like 15Five or Lattice (designed for larger teams but scalable) can automate weekly or bi-weekly check-in prompts to employees — asking how work is going, what's blocking them, what they need. For a very small team, you can do this manually with a simple template. The practice matters more than the platform.

For a business owned by someone your age, building this culture early is a genuine differentiator. Your employees — also likely young — will have other options. The businesses that retain people create environments where feedback flows both ways: where you're as willing to hear from your team as you are to deliver your own assessments. AI can help you systematize that habit. The intention to build it is yours.

Lesson 4 Quiz

Performance management, feedback, and the limits of AI judgment.
1. What was the most important insight Jordan had when he used AI to draft his feedback conversation with Amara?
Exactly. The AI didn't have the insight — the process of articulating the situation honestly did. That's the real use case: AI as a forcing function for your own thinking.
The key insight was Jordan's own realization — surfaced by the writing process — that he'd been framing the conversation as confrontation instead of support. The AI didn't generate that insight. The act of writing things out clearly did.
2. In a Performance Improvement Plan, what makes an improvement goal actually useful?
Right. "Respond to customer messages within 24 hours" is a useful PIP goal. "Improve your attitude" is not — it's unmeasurable and gives the employee no actionable direction.
Specific and measurable is the standard that matters. Vague goals ("improve attitude") are useless for both parties — the employee can't act on them and they don't hold up legally either.
3. You're preparing to terminate an employee after documented performance issues. Which use of AI in this process is most appropriate?
Right. Preparation is the appropriate use case. AI helps you organize, anticipate, and document — but the decision and the conversation are yours.
AI is useful for preparation: structuring the conversation, anticipating questions, reviewing documentation. It cannot evaluate whether the decision is right, and having AI conduct the conversation would be both inappropriate and damaging to trust.
4. What does the SBI framework stand for, and why is it useful for feedback conversations?
Correct. Situation, Behavior, Impact. By anchoring feedback to a specific situation and observable behavior, you avoid the character-judgment territory ("you're lazy") that makes feedback feel like an attack.
SBI stands for Situation, Behavior, Impact. It's useful because it anchors feedback to observable specifics rather than character assessments — which makes it easier to hear and easier to act on.
5. You own a small graphic design studio and haven't given your one employee any feedback in six months because nothing has gone seriously wrong. What risk does this create?
Exactly. When feedback only happens when something is wrong, it carries alarm-signal weight that makes it harder for both parties. Regular low-stakes check-ins normalize the practice.
The risk is cultural and relational: when feedback only happens in crisis, it becomes loaded with alarm-signal weight. Regular, low-stakes feedback normalizes the conversation so it doesn't feel like a threat when something actually needs to change.

Lab 4: The Feedback Conversation

Prepare for a hard conversation — and defend the judgment calls you make along the way.

Your Role

You run a small social media management business with two employees. Your second hire, a 21-year-old named Kenji, is talented but has missed two client deadlines in the past month and hasn't been responsive to your messages during business hours. You haven't said anything yet. You need to have a conversation with him this week.

You're going to use the AI as a thinking partner to prepare for the conversation. The AI will ask you hard questions about what you actually know vs. what you're assuming, help you apply the SBI framework, and push back if your proposed approach is too vague or too harsh.

Start by telling the AI everything you know about the situation with Kenji — what happened, when, what the impact was on clients, and what you've noticed. Be honest about what you don't know. Then tell the AI how you're thinking about approaching the conversation.
Feedback Prep Lab
AI Peer · Lesson 4
Before you tell me what you're going to say to Kenji — I want to know what you actually know versus what you're assuming. "Not responsive to messages" could mean a lot of things. "Missed deadlines" needs context. Walk me through the specific incidents: what was due, when, what happened, and what you've actually observed. Then we'll figure out what kind of conversation this actually needs to be.

Module 6 Test

Hiring, HR, and AI — 15 questions. Score 80% or above to pass.
1. What is the core mechanism of most AI resume screening tools?
Correct. Text similarity matching is the foundation of resume screening tools — everything else layers on top.
The core mechanism is text similarity scoring between the resume and job description. Additional features build on that foundation.
2. An algorithm filters out resumes that list community college attendance without the employer choosing that criterion. This is an example of:
Right. School type functions as a proxy for something else — without anyone explicitly choosing it as a filter criterion.
This is a proxy variable — the algorithm uses school type as a stand-in for some measure of quality, encoding bias without explicit intent.
3. New York City's Local Law 144 requires employers using automated employment tools to do what?
Correct. Local Law 144 requires bias audits of automated hiring tools and candidate notification — one of the first laws of its kind in the US.
NYC Local Law 144 requires bias audits of automated employment decision tools and disclosure to candidates that such tools are being used.
4. When generating a job description with AI, what is the single most important thing to include in your prompt?
Exactly. Specificity about real work and explicit openness signals are what separate a useful AI-generated JD from a generic template output.
The highest-leverage prompt additions are real task specifics and explicit guidance about inclusive backgrounds — not formality, credentials, or competitive benchmarking.
5. Which of these job description requirements is most likely to function as credential inflation?
Right. A degree requirement for a barista role has no demonstrated connection to performance — it's a credential that filters by background, not ability.
The bachelor's degree requirement for a barista is the textbook example of credential inflation — an arbitrary requirement that screens by background rather than predicting actual job performance.
6. Which states require salary ranges to be included in job postings as of 2024? (Select the best answer.)
Correct. Colorado was an early mover, followed by California, New York, Washington, and others. The trend is expanding.
Colorado, California, New York, and Washington all have salary transparency requirements in job postings as of 2024 — with more states moving in that direction.
7. Devon asked his employer whether a scheduling app tracked his phone location. What responsibility does this illustrate for small business owners?
Right. Deploying a platform without understanding its data practices is a failure of due diligence that lands on the employer, not the platform vendor.
The employer is responsible for understanding what data any platform collects about their employees — before asking employees to use it, not after they ask.
8. What does the rule of thumb "track the work, not the person" mean for employee GPS monitoring?
Correct. The principle distinguishes legitimate work coordination from continuous surveillance — the latter being both a legal issue and a trust violation.
"Track the work, not the person" means GPS is appropriate during work hours for coordination — not running continuously including off-shift hours.
9. The SHRM estimate that employee turnover costs 50–200% of annual salary primarily justifies what business investment?
Exactly. If losing someone early costs that much, investing in the onboarding experience that keeps them is a straightforward cost-avoidance calculation.
The turnover cost data makes structured onboarding and retention investment an easy calculation — the cost of building a good first week is tiny compared to the cost of losing a hire in the first 90 days.
10. Jordan's most valuable realization while using AI to prepare for Amara's feedback conversation was:
Right. The AI didn't generate that insight. Writing the situation out honestly forced Jordan to examine his own framing.
The key insight was Jordan's own — surfaced by the process of articulating the situation, not generated by the AI. That's the real use case: AI as a forcing function for clearer thinking.
11. Which statement best describes the SBI framework for feedback?
Correct. Situation, Behavior, Impact. By staying specific and observable, you avoid character-judgment territory that makes feedback harder to hear.
SBI: Situation, Behavior, Impact. It works because specific and observable beats abstract and judgmental every time in a feedback conversation.
12. What is the most important characteristic of a useful Performance Improvement Plan goal?
Right. Vague goals are useless for everyone. Specific, measurable, time-bound goals are actionable for the employee and defensible for the employer.
Specific and measurable is the standard. "Respond to clients within 24 hours, three times per week for 30 days" is a useful PIP goal. "Improve your communication" is not.
13. Which type of AI hiring tool carries the highest bias risk and why?
Correct. Scoring communication style, tone, and demeanor introduces cultural and socioeconomic bias — and employers typically can't inspect the scoring rubric.
Conversational AI screeners evaluate subjective personality signals — communication style, demeanor, tone — that carry significant cultural variation and bias risk. And you usually can't see how they're scored.
14. You're preparing to terminate an employee after three documented performance warnings. Which use of AI is appropriate?
Right. Preparation — structure, anticipating questions, reviewing documentation — is the appropriate lane for AI in termination. The decision and the conversation are yours.
AI is appropriate for preparation: drafting the conversation structure, anticipating likely questions, reviewing your documentation. It cannot make or validate the termination decision.
15. Why do regular low-stakes feedback check-ins matter more than annual reviews for small teams?
Exactly. When feedback is routine, it stops being an event. That makes every future conversation — including hard ones — easier for both sides.
Regular check-ins normalize feedback as a communication habit rather than an event. That removes the alarm-signal weight from future hard conversations — which makes them shorter, clearer, and less stressful for everyone.