There is a version of AI in HR that lives entirely in conference keynotes and vendor slide decks. Then there is the version I work in, which involves late nights in Wisconsin building Claude API integrations that actually run inside real HR workflows, serving real employees, with real consequences if they break.

I am Drew Soule, an HR executive based in Southeast Wisconsin with 15 years across hypergrowth tech, aerospace, financial services, and healthcare. I have navigated labor relations across a 1,200-person multi-site manufacturing and supply chain organization with active CBAs. I have served as Sr. HRBP to an 800-person global Product and Engineering org running 10 to 40 concurrent employee relations cases every single week. I built an entire People Programs function at a healthcare organization in 90 days. And I have been building custom HR workflow automations using the Claude API in production environments.

I am also a person born with a unique physical disability. I lead every room I enter. I say that not for pity, and not for applause. I say it because the lens you develop when the systems around you were not built for you is the same lens that makes you a better AI in HR practitioner. You learn to interrogate default assumptions. You learn that what gets built reflects who was in the room when it was designed.

That matters enormously right now.

What Most AI in HR Conversations Miss

Generative AI HR use cases get discussed constantly. What rarely gets discussed is the gap between a ChatGPT demo someone ran in a browser and a Claude API HR workflow you have actually stress-tested inside a people operations context.

The difference is significant. Running a language model in a chat interface is not the same as integrating it into a structured workflow where the inputs are messy, the stakes involve real employees, and the output needs to be consistent enough to hold up under scrutiny. HR analytics AI integration, for instance, is not just feeding a spreadsheet into a model. It is making decisions about which signals are worth amplifying, which are misleading, and which carry risk of embedding historical bias at scale.

When I diagnosed that 53% of exits at a healthcare organization were happening within the first 90 days, the insight did not come from a model. It came from structured pattern recognition across qualitative exit data, tenure data, and manager-level cohort analysis. What AI did was accelerate the synthesis and surface correlations I would have taken weeks to triangulate manually. That still produced a 22% attrition reduction and over $400K in cost avoidance. But none of that happens if you do not understand the underlying people problem well enough to ask the right question of the system.

That is the practitioner edge. You cannot prompt your way to it without the HR expertise underneath.

Building Claude API HR Workflows in Practice

Here is what Claude API HR workflows actually look like in my work.

In high-volume employee relations contexts, one of the most time-consuming tasks is documentation. During periods when I was managing 10 to 40 concurrent ER cases weekly, the cognitive load of context-switching between cases while maintaining accurate, appropriately toned written records was significant. I built automations using the Claude API that take structured case inputs and generate first-draft documentation aligned to our internal standards, flagging tone inconsistencies and identifying places where additional factual specificity was needed before the document could be considered complete.

This is not AI replacing HR judgment. It is AI handling the mechanical drafting load so the HR practitioner can focus on the judgment that actually requires human expertise. The model does not decide whether a termination is warranted. The model helps me document a progressive discipline conversation accurately and quickly so I can move to the next case without losing fidelity.

In labor relations contexts, I have used similar Claude API integrations to synthesize CBA language across a multi-site manufacturing environments and surface relevant precedent quickly during grievance handling. A 1,200-person organization with active collective bargaining agreements generates a significant volume of contractual interpretation questions. Reducing the time to locate relevant language from 20 minutes to under two minutes per query adds up fast when you multiply it across a full year of labor relations activity.

These are not hypothetical generative AI HR use cases. These are HR automation practitioner decisions made in production environments with real constraints.

AI Ethics in People Operations Is Not Optional

“AI ethics people operations” is a phrase that shows up on a lot of conference agendas. What it means in practice is something most organizations have not fully worked through.

When I build an automation that touches employee data, there are a set of questions I treat as non-negotiable before anything goes into production. Who has access to the outputs? What data is the model actually seeing, and is there a retention implication? If the model surfaces a pattern in the data, is that pattern actionable, or is it a proxy for something that would be discriminatory to act on directly? Is the data secure?

That last question is the one most practitioners underestimate. HR analytics AI integration creates real exposure when you are working with demographic data, performance data, and tenure data in the same environment. A model can surface a correlation between a manager cohort and attrition that looks like a performance signal but is actually a reflection of who that manager tends to hire and how that group has historically been evaluated. Acting on that correlation without interrogating its source is how bias gets industrialized.

I spent years supporting a large Product and Engineering organization where representation gaps were real and measurable. The last thing that environment needed was an AI system that laundered those gaps through a veneer of algorithmic objectivity. The AI ethics work is not separate from the HR work. It is the same work.

The Disability Lens Is a Design Asset

I want to be direct about this because I do not believe in being vague about things that matter.

I was born with a physical disability. I use a wheelchair. I have navigated every element of a professional career in a world that, on balance, still designs its physical and organizational infrastructure around the assumption of able-bodied participation. That experience is not incidental to my work as an HR leader. It is foundational to it.

Disability inclusion and belonging are not soft values. They are diagnostic capabilities. A leader who has personally experienced what it feels like to be excluded by default design is calibrated differently when evaluating whether a new system is likely to exclude someone. That calibration applies directly to AI in HR.

When I build automated workflows that touche any part of the employee lifecycle, I think explicitly about who the system might fail. Employees for whom English is a second language. Employees with cognitive disabilities who may interact with an AI-generated communication differently than a neurotypical employee. Employees in roles with limited digital access who may never see the AI-generated output at all, only a downstream effect of it.

Disability belonging workplace is a real practice. It requires that the people building the systems have interrogated their own assumptions. I have done that work. It shows up in everything I build.

What Wisconsin HR Practitioners Need to Think About Now

I am building this work out of Southeast Wisconsin, and I want to anchor this for other HR tech practitioner Wisconsin professionals because the context here is specific.

The Midwest HR environment often involves multi-site organizations, manufacturing, supply chain, and labor relations complexity that is different from the coastal tech-company context where most AI in HR content is generated, where I was directly exposed to it. The organizations I have worked in here have unionized workforces, plants across multiple states, and HR teams that are lean relative to headcount. The ROI calculus for HR automation practitioner decisions looks different here than it does in a 350-person SaaS company in San Francisco.

That is not a limitation. It is an opportunity. If you can build AI in HR workflows that hold up inside a 1,200-person manufacturing org with active CBAs and a lean HR team, you have built something genuinely robust.

The AI in HR practitioner conversation needs more voices from that environment. People who have done the labor relations work, the multi-site coordination, the high-volume ER case management, and are now integrating AI into those workflows from a position of operational credibility.

That is the work I am doing. And I believe the organizations that take it seriously now, with real practitioners building real systems and asking the hard ethics questions before deployment, are the ones that will have a durable advantage in people operations over the next decade.

Not because AI is magic. Because the humans building it understood the work first.