The headline that buried the lede
OpenAI just published a case study about HSP GRUPPE, a German network of tax advisory, auditing, and law firms. The top-line numbers are eye-popping: 84% weekly active usage across 755 weekly active users, 98.6% of surveyed employees reporting higher productivity, and 500,000+ messages exchanged with ChatGPT over six months.
But honestly? The metrics are the least interesting part.
What's actually worth paying attention to is how they got there—because HSP didn't just deploy ChatGPT Enterprise. They treated AI adoption as an organizational transformation, not a software rollout. And the difference between those two approaches is the gap between 84% adoption and the dismal ~10-20% engagement most enterprises see after the pilot-project honeymoon ends.
Process nerds inherit the AI earth
Here's the part that doesn't make it into most AI case studies: HSP GRUPPE has been investing in process standardization and digitization for two decades. Long before ChatGPT existed, they had embedded quality management, workflow discipline, and a culture of continuous improvement across the firm network.
That foundation matters more than any prompt engineering workshop.
CEO Carsten Schulz frames the challenge perfectly: "Technology is moving faster than organizations can transform. Our challenge wasn't introducing AI—it was helping the organization absorb it."
This is the insight most companies miss. AI adoption isn't a technical problem—it's an organizational absorption problem. You can't AI-native your way out of broken processes. You can't prompt-engineer around unclear decision rights. If your firm doesn't already have the muscle memory for systematic process improvement, ChatGPT Enterprise becomes expensive autocomplete instead of an operating model upgrade.
The architecture of adoption
HSP's rollout strategy had three load-bearing pillars:
Monthly AI forums where employees shared practical use cases with each other. Not top-down training—peer learning at scale.
Custom Agents as organizational memory. HSP didn't just let everyone freestyle with ChatGPT. They identified successful patterns and codified them into shared Agents. One example: the AI Client Communication agent, which supports first drafts with consistent tone and structure while keeping professional review and final responsibility with the human expert.
Another: the Booking Assistant SKR03 & SKR04, which helps with preparation and classification of accounting questions. Again, the professional makes the final call.
This is the critical move: successful experiments become repeatable workflows that everyone can leverage—not just the AI-fluent early adopters.
Strong governance boundaries. ChatGPT Enterprise's security features provide the technical foundation, but HSP layers on its own data protection, confidentiality, and governance requirements. The use of client-related information is controlled. Professional review isn't optional.
Notice what's not in that list: mandatory training. Feature rollouts. Executive mandates.
Instead, they built infrastructure for organic diffusion of successful patterns.
What 98.6% productivity improvement actually looks like
The survey results sound almost too good to be true: 98.6% reporting higher productivity, 84.6% reporting improved work quality, 95.9% reporting weekly time savings.
But the real evidence is in the work patterns:
Partner Magdalene Posnak cut real estate investment analysis time from nine hours to around two. The time saved went to client advisory—not headcount reduction.
Managing Partner Frank Heibel uses ChatGPT as a "technical sparring partner" to work through complex tax questions and refine client communications. His quote: "It completely changes the level at which you can work."
Managing Director Marco Sell uses AI-powered dashboards to explore financial scenarios with clients during meetings. Lawyer Jan-Henrik Leifelt uses ChatGPT to strengthen—not replace—professional judgment.
The pattern: less time preparing information, more time applying expertise.
HSP estimates approximately 40,000 hours of additional annual capacity. Around 28,000 hours could support productive, generally billable specialist work; roughly 12,000 hours could support administration and client service. Based on conservative hourly rates, they estimate a theoretical annual revenue potential of approximately €3.8 million.
Crucially, this is a capacity scenario—not realized or guaranteed revenue. The firms are already operating at capacity with substantial backlogs, so freed-up time gets redirected to additional client work, shorter turnaround times, and stronger service.
The agentic future: redesigning workflows, not automating tasks
Here's where it gets really interesting. HSP is already piloting ChatGPT Work with a small group of developers and administrators, exploring how agentic AI can safely automate complex workflows.
Schulz's vision goes beyond task automation. He points to year-end accounting as an example: today, accountants often discover missing information only when they begin preparing annual accounts—months after bookkeeping has been completed.
With ChatGPT Work, HSP is exploring workflows where AI continuously reviews bookkeeping throughout the year, identifies missing information, and proactively requests documents from clients. So much of the preparation has already happened before an accountant even opens the file.
Schulz: "We identified 88 opportunities for automation. But I think that's already thinking too small. The real question isn't how we automate today's processes. It's how we redesign them. That's what opens the door to a completely new world."
This is the shift from AI-as-copilot to AI-as-orchestrator. Not helping humans complete existing tasks faster, but restructuring the work itself.
What this means for the rest of us
HSP GRUPPE is not a typical enterprise. They're a network of independent tax advisory, auditing, and law firms with 20 years of process discipline and quality systems already baked in. The numbers in this case study reflect a shared workspace covering 81 organizational groups—HSP plus Kanzleipakt.
So you can't just copy-paste their playbook. But you can extract the architectural principles:
- Build on operational foundations first. Strong processes accelerate AI adoption. Weak processes get amplified by AI.
- Create peer learning infrastructure. Regular forums where successful ideas spread quickly matter more than top-down training.
- Codify successful experiments into shared capabilities. Turn one person's hack into everyone's repeatable workflow.
- Design for professional judgment, not automation. The greatest value comes when AI enhances expertise rather than replacing it.
- Think in workflows, not tasks. Redesigning end-to-end processes creates more leverage than automating isolated activities.
The most striking thing about the HSP case study is how unsexy the success factors are. No exotic prompting techniques. No cutting-edge models. No AI research team.
Just disciplined organizational design, strong governance, peer learning at scale, and a CEO who understands that absorbing AI is harder than deploying it.
Schulz again: "Everyone talks about AI replacing tax advisors. I think that's the wrong conversation. The real opportunity is helping qualified professionals become dramatically more effective."
That's the bet. And if you're measuring success by weekly active usage instead of headcount reduction, you're probably making the right one.