If the AI system with the most freedom in your area made a costly mistake tonight, whose phone would ring?

Pick the answer that best reflects operational reality today.

Which seat are you in?

Required for question routing and matched-pair comparison across your organization.

Thank You for Your Interest

This research study specifically benchmarks organizations where builders, executive leaders, or risk officers manage AI systems in production.

You can share this assessment with colleagues in engineering or leadership to see where your organization places on the Readiness Matrix.

Is AI live in your area right now?

Required for survey routing and cohort categorization.

Is your AI mostly built by your team, or does it come inside software you buy?

E.g., Microsoft Copilot, Salesforce Agentforce, ServiceNow, or custom in-house LLM pipelines.

Focus on your most autonomous system

Now think of one AI system: the one in your area that's allowed to do the most on its own.

Keep this specific system in mind for the next questions. You will answer about what this system can actually do and who it impacts.

What can it do without a person approving first?

Select the highest level of freedom it possesses today.

If it got something wrong, who would feel it?

Select all that apply.

Before it launched, was it tested on real cases from your own data, not the vendor's demo?

Real enterprise cases, production edge cases, and golden test sets.

Does it run with its own limited permissions, not someone's login or an admin account?

Principle of least privilege, scoped service accounts, and token boundaries.

If it did something wrong at 2 a.m., would you find out before a customer told you?

Proactive observability, runtime anomaly detection, or threshold alerts.

Has anyone actually tried switching it off or rolling it back?

Tested circuit breakers, kill-switches, or state rollback drill.

Does a person or an automated check look at its output before it takes effect?

Pre-flight evaluation, policy guardrail, or sampled inspection.

Can you name the number it's supposed to improve?

A defined business KPI, operational efficiency target, or revenue metric.

Do you know what it really costs, including the people who check or fix its work?

Total cost of ownership including tokens, hosting, and human remediation time.

Is one named person, not a team, accountable when it gets something wrong?

A designated individual owner who answers for operational failures.

Has anyone written down how many mistakes are acceptable?

Documented error budgets, SLA thresholds, or formal risk tolerance criteria.

Is there a date when someone will decide whether to keep it, scale it back or switch it off?

A scheduled review milestone where deprecation or decommissioning is an explicit option.

In the past year, has something you said wasn't ready launched anyway?

Engineering authority in deployment decisions.

In the past year, has an AI target pushed a launch your engineers said wasn't ready?

Strategic time-to-market pressure vs engineering caution.

When does your team usually first hear about a new AI system?

Governance and compliance involvement window.

Has AI in your area done any of these?

Select all that have occurred in production.

Has any AI system been rolled back, scaled down or switched off after launch?

Operational retractions, pilot cancellations, or rollbacks.

In a normal week, how many hours does your team spend checking or fixing AI output?

Human-in-the-loop oversight and rework cost.

What's the biggest thing in the way of putting AI live in production?

Pick the primary blocker inside your company.

Think about that one system again. Would you put your name to it today?

If asked to sign a formal production sign-off sheet right now.

Gut feel, 0–10: how ready is your area for AI that acts without a person approving each step?

0 = Completely unprepared · 10 = Fully battle-tested & ready

Company size

Approximate total workforce headcount.

Industry

Your organization's primary domain.

Region

Where your primary team or operations are situated.

One thing that makes AI safe to sign for where you work, or the one thing that's missing.

Optional. Selected anonymous observations will be analyzed on DayZero2DayOne.