Back to Guides
Enterprise Software 14 min read

Anatomy of an AI workshop for an HR team

Anatomy of an AI workshop for an HR team

A working session for a people team runs about half a day, moves through five blocks, and never opens with a slide about the future of work. The material is the team’s own job descriptions, policy drafts, interview guides, and survey comments. That last detail is why an HR room is harder to run than a sales one. The documents on the table are about named people, and the habits the room needs are not the habits a general AI class teaches.

General Assembly surveyed close to 300 HR professionals in the US and UK in July 2025 and found 82% already use AI in their day-to-day work while only 30% received training built for their job. Another 26% got no formal training at all.

Usage arrived first. The method did not follow.

What makes an HR room different

Every function brings sensitive work into a session. HR brings the record.

A sales team pastes a call note and argues about tone. A finance team pastes a variance summary and argues about rounding. When an HR generalist pastes what is in front of her, it is a performance note with a name on it, a grievance summary, or sixty survey comments that two managers could identify by writing style alone.

The first job of the session is not prompting. It is a redaction habit that survives a busy Tuesday, when nobody is watching and the policy sits in a folder nobody opens.

The second difference is structural. HR uses these tools and also owns the rules for everyone else’s use of them. SHRM’s State of AI in HR 2026, a survey of more than 1,900 HR professionals published in April 2026, found that 52% of organizations do not involve HR in their AI strategy while 28% say HR leads the upskilling effort. Left out of the decision, handed the rollout.

Our role-by-role view of what each function needs explains why that ordering problem keeps recurring.

What the team brings into the room

Five artifacts, requested a week ahead, real and unfinished. A session built on sample data teaches sample skills.

What they bring Why it is in the room What the block produces
A job description about to be posted Hiring text is HR’s highest-volume writing task; 46% name it A drafting standard the whole team reuses
A policy section being rewritten Policy is where plain language and precision fight A two-pass method: draft, then read against it
An interview guide for a live role Structured questions are easy to generate and easy to get wrong A guide checked against the scorecard
Last quarter’s engagement comments, de-identified Feedback analysis ties with job descriptions as the most common use at 46% A summary whose claims stay traceable
One message nobody wanted to write A decline, a correction, a hard conversation A tone register the team agrees on

The fifth one does the most work. It is the task people already dread, which makes it the one they will try again the following week without being asked.

Do not ask for a tool list, and do not spend the morning picking between ChatGPT, Claude, Gemini, and Microsoft Copilot. Teams of this size almost always have at least one of them already paid for. The session works on whichever account the company has approved, because the skill transfers and the license negotiation does not.

The run of show

Block Minutes What happens in it
Where this already is 25 Everyone says what they use now and for what
Redaction as a motor skill 45 Hands-on de-identification of their own text
Drafting on the five artifacts 90 The working core, with a short break inside it
The decision line 35 Which steps stay with a person, written down
Standards and two habits 25 What goes on the one-page method, what gets measured

About three hours and forty minutes with the break. Shorter than a day, longer than a lunch-and-learn, and the length is not arbitrary: the drafting block needs ninety uninterrupted minutes to get past the first impressive answer and into the second and third attempts, where the method lives.

The opening block earns its twenty-five minutes because of what it surfaces. Someone is usually using a personal account for work material. Someone else has quietly stopped using the approved tool after one bad answer and never said so.

You cannot design around either fact until it is in the room, and neither shows up on a survey.

The block only an HR room needs

Forty-five minutes on redaction, done by hand, on their own documents. No other function gets this block, and in the rooms we run it is the one that overruns.

The sequence is dull on purpose. Each person takes a real note or comment set and strips it themselves: names, team size small enough to identify, dates that pin an incident, the phrasing that gives away who wrote it.

They swap with a neighbor and try to re-identify. They usually succeed on the first pass, which is the point.

Then the account question, which is duller and matters more: which workspace is approved, what the retention setting is, who the admin is, and what happens to a conversation after it ends. SHRM found 49% of organizations have a policy regulating employee AI use and only about a quarter believe it is future-proof. That policy is usually HR’s own document, and the session is often the first time the team reads it closely.

Three obligations get dates on the whiteboard here, because vague warnings about bias do not change behavior:

  • EU AI Act, Article 4. The AI literacy duty on providers and deployers has applied since 2 February 2025. It binds you if you have staff in the EU, and training is how it is met.
  • New York City Local Law 144. An automated employment decision tool needs a bias audit within the prior year, a public summary of the results, and ten business days’ notice to candidates. The city has enforced it since 5 July 2023.
  • Illinois. The amended Human Rights Act took effect 1 January 2026. It bars AI that produces a discriminatory effect in hiring, promotion, discipline, or discharge, bars ZIP codes as a proxy for protected classes, and requires notice. It does not mandate a bias audit.

A session discharges the first and none of the rest. Say so plainly, because the gap is where buyers get surprised. SHRM also found that 57% of HR professionals in states with workforce AI rules are unaware of those rules.

Where the room goes quiet

One line separates safe acceleration from legal exposure, and the room always goes quiet when it lands.

An assistant may summarize a resume against a written scorecard and draft the note explaining a recruiter’s call. Once software contributes to who advances, you are operating an automated employment decision tool, and audit, notice, and record duties attach. The same line runs through promotion, discipline, and separation.

The quiet comes when somebody realizes a shortlist they produced six weeks ago crossed it. Nobody enjoys that minute. It is also the minute the session pays for itself, and it does not happen in a webinar because nobody brings a real shortlist to a webinar.

Our look at sessions for talent teams works through where that line sits step by step in a hiring process.

Three ways the session fails

The team arrives empty-handed. Without the five artifacts you run a demo. People leave entertained and change nothing on Monday. Chase the documents twice before the date, and reschedule rather than run it dry.

The person who signed the invoice skips it. When the HR director books a session for the generalists and takes the morning as calendar relief, the new standard has no sponsor. The first time a draft takes longer than the old way, the method dies quietly. Attendance by whoever approves the work is not optional.

One session, no second one. Two sessions two or three weeks apart beat a single long morning, because the gap forces people to use the method on live work and bring back what broke. The Conference Board’s July 2026 research, built on interviews at 35 companies and a survey of close to 1,300 workers, found employer training concentrated on basics, rarely reaching harder skills such as directing agents. One morning is how a program stays stuck at basics.

What the team walks out holding

Five things, all of them one page or shorter:

  1. A drafting standard for hiring and policy text, written by the team rather than handed to it.
  2. A de-identification checklist, tested on their own material.
  3. Five prompts tied to the five artifacts, in the team’s own words.
  4. The decision line, written, naming which steps stay with a person.
  5. Two measures baselined that morning, with a date to re-read them.

What they do not walk out with: a certificate, a tool-stack decision, or a finished AI policy. Those are separate purchases, and a session that promises all three is selling the morning twice.

What a session like this costs

Facilitated sessions run roughly $2,000 to $15,000 at market rates, depending on length, custom material, and whether a second session is included. Add participant time honestly: six HR people for four hours, at the May 2025 median of $36.51 an hour for HR specialists, is about $876 of salaried time before benefits.

Compare that with the per-seat route. AIHR’s Artificial Intelligence for HR lists $1,125 a seat for 33 hours across 17 modules, carrying 28 SHRM PDCs, paced at three hours a week over about eleven weeks. It is the most HR-specific syllabus on the market and worth one seat for whoever becomes the internal reference. Six seats is $6,750, three months, and six private methods at the end of it.

Our comparison of the HR options prices the rest of the field against what each one buys.

The buyers’ own preference points the same way. In the General Assembly survey, 70% of HR professionals said they want interactive workshops focused on HR use cases, and 85% of those who had received job-specific training felt very or completely confident using AI, against 63% of the self-taught.

How you know in ninety days

Baseline two numbers during the final block, not afterward. 56% of HR functions do not formally measure the success of their AI investments and only 16% use ROI as a metric, which is why so few of these programs survive a budget review.

Four measures that hold up:

  • Artifact reuse. How many of the five documents produced a template the team used again. Two is a working session. Zero is a demo.
  • Cycle time on one named task. Pick the job description, from intake to posted. It is frequent, it is visible, and it is nobody’s vanity metric.
  • Unapproved accounts. Count them in the opening block and again at ninety days. The target is zero, and this is the measure a general counsel cares about most.
  • Quality, checked by a person. SHRM’s 2026 worker survey of more than 5,000 people found 44% of AI users describe their own output as “AI slop”. The fix is a review step in the standard, not a cleverer prompt.

The reassurance worth giving a nervous team is a real one. The BLS projects employment of HR specialists to grow 6% from 2025 to 2035, faster than the average for all occupations.

The function is not shrinking. What changes is the method inside it, which is the whole argument in our case that buying licenses is not the same as building capability.

Frequently asked questions

How long should an AI workshop for an HR team be?

About three and a half to four hours of working time, split across five blocks, with a second shorter session two or three weeks later. A full day loses the room after lunch. Ninety minutes cannot fit both the redaction work and enough drafting attempts to form a habit.

What should AI training for HR actually cover?

De-identification of employee and candidate material, drafting on the team’s own hiring and policy text, a written line between summarizing and deciding, and one measurement habit. Tool selection and vendor features are the least useful part of the morning.

Can HR put employee data into ChatGPT or Microsoft Copilot?

Only through an approved workspace, with the retention setting understood and names and identifying detail removed first. The rule that matters is not which product you use, it is whether the people doing the pasting have practiced the redaction step on their own documents.

Does a workshop meet our AI compliance obligations?

One of them. The EU AI Act’s Article 4 literacy duty is met by training staff, and it applies only if you employ people in the EU. Bias audits, candidate notice, and record keeping under New York City and Illinois rules are separate duties that no session discharges.

Should we buy a certificate or run a session?

Both, in that order of size. One seat on an HR-specific certificate for the person who becomes your internal reference, and one facilitated session for the team so that everyone drafts to the same standard. Six certificate seats produce six methods and no shared method.

Who should be in the room?

The whole people function, including the person who approves the budget, plus recruiting if it sits inside HR. Keep it under about ten so that everyone works on their own document rather than watching someone else work on theirs.

What does the team need to prepare?

Five real artifacts, sent a week ahead: a job description, a policy section, an interview guide, a de-identified comment set, and one message they have been avoiding. Confirmation of which AI account is approved, and who administers it.

How do we measure whether AI training for employees worked?

Baseline two numbers in the session itself and re-read them at ninety days: cycle time on one named recurring task, and how many of the session’s artifacts became reused templates. Add a count of unapproved accounts, which should fall to zero.

Is role-based AI training better than one company-wide session?

For HR, clearly. A company-wide session cannot teach de-identification on grievance notes or the automated-decision line in hiring, and those are the two things an HR team most needs. Shared ground across every function is worth one short all-hands block, not a morning.

Key takeaways

  • An HR session is set apart by its subject matter, not its slides. The documents are about named people, so redaction comes before prompting, and it gets forty-five hands-on minutes.
  • Five real artifacts, requested a week ahead, decide whether the morning becomes a method or a demo. The message nobody wanted to write earns its place the fastest.
  • The decision line between summarizing and deciding is what the course pages leave out. Once software contributes to who advances, audit, notice, and record duties attach.
  • Only the EU AI Act literacy duty is met by training. The New York City audit and notice rules, live since July 2023, and the Illinois notice duty from January 2026 are separate obligations.
  • Price the whole morning, including participant time. One certificate seat plus one session for the team costs less than six seats and produces a shared standard instead of six private ones.
  • Baseline two measures before the room empties. More than half of HR functions measure nothing, which is why so few of these programs survive the next budget review.

If you are deciding between a certificate for one person and a session for your people team, book a free AI readiness call. We will walk through your five artifacts and tell you which of the two you are missing.

Last Updated: Nov 15, 2026

AW

Arthur Wandzel

SFAI Labs helps companies build AI-powered products that work. We focus on practical solutions, not hype.

Make your team fluent in AI — then automate what proves out

  • Hands-on training applied to your own documents and workflows
  • Often reimbursable through state small-business training programs
  • Built for 5–50 person firms with no IT department

Related articles