METRIS™
High-Confidence AI Decision Consulting
Make better decisions about the AI already in your business: what to expand, improve, renew, restrict or stop.
We measure your AI systems, workforce practices and business workflows, so every decision rests on evidence.
Where we start
US health plans
It recommends denials, drafts notices and routes reviews. Two states now limit AI in these decisions, and two more follow on January 1, 2027.
Small and midsize businesses
It drafts documents, answers customers and screens applicants. You need to know how much to hand over, and what it should never see.
Small business services →Six small business studies →How much can you hand to AI? →
You stop renewing AI systems the evidence doesn't support, and you stop cutting the ones that are working.
You see what the AI actually changes in turnaround time, capacity and rework, measured against how the work ran before.
You find the weak output or the missing check before a customer does.
We measure what changed and what the evidence supports. We never claim a result the evidence can't show.
Before you expand, renew, improve, restrict or stop an AI use case, we measure three things.
What could go wrong, who could be affected, and how much risk remains, so you can decide whether it's acceptable.
Whether people can trace, check and correct the AI's outputs in the work that depends on them.
Whether people recognize the AI's limits, verify important outputs, and know when to pause, override or escalate.
Together, these show whether the evidence supports your next decision, and what has to change if it doesn't.
The chance, from your own data, that the line meets its target.
Example, simulated
No notice left without a person: 0 of 900 sampled.
A denial rate edging toward its ceiling.
Reviewers catching under half of AI errors.
Reported as unknown, never counted as a failure.
Tested on simulated data where the truth is known: a line that truly holds was never called Broken in 2,000 runs, and a truly broken line was called Broken in 99.9% of them.
Higher-scrutiny decisions
AI doesn't stand still. A vendor updates the model. A team changes the workflow. New people start using the tool. A pilot becomes part of everyday work.
A policy, a vendor's claim or a one-time review can't tell you whether the reasons behind your decision still hold. Trust is something you test against evidence, not something you declare.
We look at the AI in its real workflow: what it's used for, where people check or override it, and what you can show. When something important changes, we measure again, so you can decide what to expand, improve, renew, restrict or stop.
What changed in AI this week, and which line it moves, is in the Weekly AI Debrief at the top right of every page.
Suneeta Modekurty leads every Decision Review and runs the analyses herself. She is a bioinformatics data scientist and researcher who has spent ten years building AI and data science systems across several industries.
Her work follows one principle: an important AI decision should rest on evidence someone can examine, not on a vendor's claims. Her research in cancer and Alzheimer's disease includes a first-author abstract published in the Journal of Clinical Oncology (ASCO 2024 Annual Meeting), and she reviews manuscripts for PLOS ONE and Frontiers journals.
She is ISO/IEC 27701 and ISO/IEC 42001 Lead Auditor certified, has delivered live trainings on measuring AI on O'Reilly Media, and is the author of The AI-Human Synergy (2024).
AI decisions should not outlive their evidence.
A vendor updates a model. A workflow changes. Usage expands. People adapt. Each material change is a reason to recheck the evidence affected by that change.
We name the gap and the evidence needed to close it. Missing evidence never becomes a favorable finding.
We tell you before the engagement starts, and explain what specialist work or referral you may need.
Your organization's accountable leader. METRIS™ supplies the evidence record and the decision support.
Proof you can rerun
Each figure comes from code with a fixed random seed. Anyone can run it again and get the same numbers. Clients see the code behind their own reports.
# Is clinician review still catching AI errors? caught, missed = audit_window(month) # counts only p_holding = beta.sf(0.50, 1 + caught, 1 + missed) state = "Holding" if p_holding > 0.95 else \ "Broken" if p_holding < 0.05 else "Drifting" # seed 20261013 · numpy 2.5.3 · scipy 1.18.1
Have an AI decision coming up?
Start with a free check, or bring us the AI decision you need to make.
Built for leaders and teams deciding what to do with AI they already use in real work.
Your own portal with your lines, reports and decisions, and a members' community. By invitation.