An AI inventory
A register of the models, agents, prompts and tools in use, with an owner and a purpose for each.
Services / Run
AI your business
can stand behind.
Know which AI you run, what it can get wrong and who decided it was acceptable. We help you build the inventory, assessments, testing, monitoring and human review that let you scale AI safely.
- Best when
- AI is spreading faster than anyone can oversee it, or a customer is asking how you govern it.
- You bring
- Your AI use cases and any policies you already have.
- You leave with
- An AI inventory, risk assessments, a testing and monitoring approach and a decision trail.
- Shape
- A focused assessment, then the controls you choose to build.
You might be here if
Does any of this
sound familiar?
- Teams are adopting AI tools without a shared view of what is in use.
- A customer, regulator or insurer is asking how you govern AI.
- You cannot say which models and prompts sit behind which decisions.
- You use AI to check other AI’s answers and are not sure when to trust that.
Why it matters
Governance is how
you say yes.
Without it, every AI use case becomes a debate. With it, low-risk uses move quickly and high-risk ones get the scrutiny they deserve. The point is to scale AI with confidence, not to slow it down.
Evaluation is the other half. AI output has to be tested, by machines where volume demands it and by people where judgment matters. AI checks are useful and also fallible, which is why expert review is part of the design rather than an afterthought.
What we build with you
Six controls that work together.
Start with the ones your risk and your customers call for. Add the rest as you scale.
Risk assessment
Structured questionnaires for each use case, based on a recognized framework such as the NIST AI Risk Management Framework, with a mitigation plan.
Testing before launch
Evaluation sets built from your real tasks, prompt-attack tests and safety checks, with results you can read.
Monitoring after launch
Scheduled evaluation against thresholds for quality, personal-data leakage and harmful output, with alerts.
Human evaluation
Expert review loops where an automated check is not enough, so domain specialists validate results and feed corrections back.
A decision trail
Approvals, exceptions and evidence in a form you can show a customer or an auditor.
How it works
One clear step at a time.
Inventory
Find and record what is in use, including the tools teams adopted on their own.
Assess
Classify each use case by risk, record the mitigations and agree who approves what.
Test
Evaluate before launch with realistic inputs, adversarial prompts and safety checks.
Monitor and review
Watch production against thresholds, route breaches to a person and record what was decided.
What changes
From where you are to where you want to be.
From: AI adopted tool by tool
To: A register with an owner for each use
From: Hoping the model behaves
To: Evidence of how it behaves, before and after launch
From: AI checking AI, with no human review
To: Automated checks with expert review where it matters
From: Answers to customers from memory
To: A decision trail you can show
A working demonstration
What it looks like in practice.
We built a working governance demonstration on IBM watsonx.governance and OpenPages, using a fictional staffing-agent use case. It includes an AI inventory record, risk assessments, a test suite, a prompt-attack test run, scheduled monitoring and an approval workflow. It is a demonstration, not a client deployment.
Before you start
What to expect.
Compliance is your counsel’s call.
We use frameworks such as NIST’s AI Risk Management Framework and the EU AI Act to structure assessments. Whether a use complies with a law is for your counsel and regulators to decide.
Testing shows where a system fails, not that it is safe.
Evaluation samples are finite. We report what was tested, what was not and how much to trust the result.
Questions
Common questions.
We only use ChatGPT or Copilot. Do we still need this?
An inventory and a simple policy are worth having even then. Staff may paste sensitive data into tools you do not know about, and customers increasingly ask. The depth should match your risk.
Who should own AI governance?
Usually a named business owner with support from security, legal and engineering. We help set up the roles and the review rhythm.
What is human evaluation?
Subject-matter experts reviewing AI output against a rubric, with their corrections feeding back into testing. It complements automated judges and catches what they miss.
Do you only work with IBM watsonx.governance?
It is the platform we built our lab on. The approach also works with other tooling, or with a lightweight register and process. We match the tools to your size.
How do we start?
With a focused assessment of the AI you already use. It gives you an inventory, a view of the highest risks and a practical order of work.
Scale AI with the right controls.
Tell us how AI is used in your organization today. We will suggest a proportionate starting point.