I’ve had the privilege of working with some brilliant colleagues throughout my career. But the strongest professional relationships, especially in the C-suite, aren’t built overnight. They’re built over time on a foundation of trust and a clear understanding of roles, responsibilities, and decision-making authority.
The same principle applies to AI. People won’t rely on AI systems unless they understand and trust the information and decisions behind them. And for AI to deliver meaningful results, it needs a strong foundation.
That foundation has two essential components. The first is a single, trusted source of truth. That means one authoritative record that determines which system prevails when data conflicts arise, identifies who owns each asset, and shows whether it is compliant at any given moment. The second is governance: clear decisions made in advance about where human oversight is required before an automated action can proceed.
In short, the foundation requires data you can trust and a live human in the loop.
You might have gotten away with delaying those decisions in the past. Not anymore, because the money is already spent. Our recent survey shows that 56% of organizations are implementing AI broadly across different workflows or at the scale where the business depends on it. Just 2% of those surveyed reported not using AI at all. Similar to relationships in your career, automation must be built on clear understanding, which builds trust.
AI CAN RECOMMEND, BUT NOT ANSWER FOR THE RESULT
AI does a good job processing information and recommending actions when it can pull information from a trusted system of record. It cannot establish or verify what exists in your environment, who owns it, what it connects to, what changed last month, or whether any of it would survive an audit. Those are matters of record, and someone has to keep the record and be accountable for it.
Fortunately, creating that record is a more practical process than many organizations realize. Start with discovery—an authoritative inventory of what exists across every environment. Attach ownership and lifecycle to each asset to identify them. Keep the change history, because that is what an auditor asks for. Then decide which system wins when the two systems disagree, and document that answer.
That process tends to crystallize too late for many teams, like when an auditor asks how a change was approved. Nobody wants the answer to be that the system recommended it.
FIRST, SORT ACTIONS INTO TWO PILES
Some teams are already scrambling to cover for this. In our survey, which included 1,500 IT professionals, 68% said they had seen AI create a hallucination with potential operational impact. Just over half said a team member caught it before it caused problems; 16% said that hallucinations caused issues.
For a lot of companies, that catch rate is the entire control framework. Catching stuff is great, but this framework depends on the right person being alert at the right moment. That’s not how to run a business.
Instead, sort actions into two piles before anything goes live. The first pile is routine and reversible work that can run on its own. This includes restarting a service or applying a standard patch. The second pile is anything that changes a system-wide setting or touches a live incident. That should undergo human review. After sorting them, write each group into the workflow rather than leaving it to individual judgment.
Don’t mistake this being handled on paper for being handled in real life. Our survey found 85% of IT professionals saying that every AI agent and workflow has a named owner in their organization. But only 42% say that ownership is clear in practice. Where policies exist, just 24% of employees say they are followed very consistently.
GUARDRAILS HELP ACCELERATE GOOD DECISIONS
You might have read the last few paragraphs and thought, “This sounds like more tedious governance.” To be fair, governance was the most common answer IT professionals gave in our survey when asked what slows their AI deployment; 27% name governance, security, or compliance, ahead of skills shortages and technology limits.
But I believe the problem is how governance is run, not governance itself. Think of AI governance as setting expectations and managing accountability in a working relationship. It removes uncertainty and provides clarity.
Settled rules about data, ownership, and human review take the debate out of most requests, speeding up approvals. I would rather spend a few weeks setting those boundaries than spend the year approving things one at a time.
WHAT YOU GET FOR YOUR TROUBLE
I understand if this feels both oversimplified (Trustworthy data! Keep a human in the loop!) and cumbersome (Fix your framework! Sort! Audit! Document!). But it really comes down to a bit of up-front effort for a significant payoff that can arrive faster than you might think.
IT professionals using AI at the most advanced level said in our survey that they save an average of six hours a week—double what their peers at the earliest stage save—and half say AI is freeing them up for more complex, strategic work.
Trusted data and a human in the loop let you widen an AI program with confidence instead of widening it and hoping. It’s also what allows you to confidently invest in more comprehensive, end-to-end automation systems that can improve flow without extra headaches, because you’re building on solid ground.
So, if you are deciding where to put resources in the next quarter, I would skip a bolted-on tool. Test all this with one workflow you already depend on. Reconcile what your systems say about the assets it touches. Then decide who signs off on what. Do that a few times and you’ll have both things in place: a record you trust and people positioned where it counts. While others keep bolting on and reacting, you’ll be setting up for fewer headaches, streamlined operations, and more confident decisions.
Melissa Puls is chief marketing officer and SVP of customer success and renewals at Ivanti.
