The AI policy problem most companies are now facing
AI adoption is easy. Managing it is the hard part.
Most companies didn't decide to adopt AI. Their employees did it for them.
Long before leadership teams sat down to talk about policy, governance, or risk, employees were already pasting client data into ChatGPT, drafting memos in Claude, and running analyses through Gemini.
By the time the conversation about “an AI strategy” reached the executive table, AI was already woven into daily operations — just without any of the structure around it.
It's not whether to use AI. It's how to manage what's already happening.
The problems nobody saw coming
At first, letting employees experiment with AI feels harmless. Maybe even smart. But within a few months, the cracks start showing.
- 01Sensitive company and client data ends up inside public AI tools.
- 02Nobody centrally knows which tools are being used, or by whom.
- 03Departments expense overlapping subscriptions to three or four different AI products.
- 04Compliance, legal, and IT have no real guardrails in place.
- 05AI-generated work goes out the door without proper review.
- 06Different teams standardize on different platforms, fracturing how the company communicates and operates.
It's the same shadow IT and SaaS sprawl problem companies wrestled with for a decade — only now the data leaving the building isn't just a file. It's the company's intellectual property feeding someone else's model.
Why “just pick one” doesn't work
The natural reaction is to consolidate. Pick an enterprise AI platform, roll it out company-wide, and call it solved. For mid-sized companies, two problems surface fast.
The economics rarely match the usage
Enterprise AI licensing is expensive — often disproportionately so. You pay a flat per-seat cost across the whole organization, even though only a fraction of employees actually use AI daily.
People prefer different models for different work
Forcing everyone into one tool ignores how people actually do their jobs — and adoption suffers because of it.
- WritersClaude
- EngineersChatGPT
- Workspace teamsGemini
The black box of third-party platforms
A wave of third-party AI management platforms promise centralized access, governance, and security in one place. The pitch is compelling. The reality is murkier.
Most executives can’t answer
- 01What do the underlying AI models actually cost?
- 02How is token usage being calculated?
- 03How much markup is the middleman adding?
- 04Are we paying for capacity we'll never use?
Who actually owns AI internally?
IT? Operations? Compliance? Innovation? Without a clear owner, governance stays inconsistent and spend keeps creeping. AI spending is quietly becoming the next black box on the corporate P&L.
What we built at Roth & Co
So we built our own internal AI platform.
We ran into all of this ourselves. We wanted our team to have access to the best AI models without giving up security, visibility, or cost control — and without locking everyone into a single vendor. So we built our own internal AI platform.
Employees get secure access to multiple leading AI models through one controlled environment. Management gets centralized oversight, reporting, and budget control.
Full usage reporting
Analytics across every team, tool, and token.
Centralized budget control
Token allocation and spend caps, set per team.
Permission-based access
The right people, the right models, the right data.
Security & governance
Guardrails and policies enforced by default, not by trust.
Adoption visibility
See utilization across teams — what's working, what isn't.
Multi-model flexibility
Claude, ChatGPT, Gemini and more — no vendor lock-in.
Scalable cost management
Pay for real usage, not blanket enterprise licensing.
The goal wasn’t to build another chatbot. The goal was to build something a business could actually manage, govern, and scale.
The real question
The conversation is shifting fast.
“Which AI tool should our employees use?”
“How does AI responsibly become part of how our company operates?”
The companies that figure that out — not just the technology, but the governance, the economics, and the ownership — will have a meaningful operational edge over the next few years. The ones that don't will keep paying for tools they can't see, governed by no one in particular, with data flowing to places they can't track.