Your AI Is Only as Honest as Your Culture

Your AI Is Only as Honest as Your Culture

Data tells you what’s happening. Psychological safety decides whether you find out.

This post shares my new white paper on AI and psychological safety: why the two are one system, why the problem is sharper in regulated sectors, and four moves leaders can make now.

Download the white paper (PDF, 6 pages)

Executive summary

Leaders feel pulled in two directions. AI programmes demand more data and more measurement. Enterprise Agility asks for emotional intelligence, trust and psychological safety.

I believe the tension is real but misdiagnosed. The divide is not data versus EQ. It is data used for control versus data used for learning. When data is used for control, people manage the numbers and bury bad news, so the AI learns from a sanitised version of reality.

This paper explains why psychological safety shapes how good data-based decisions get made, why the problem is sharper in regulated sectors, and four moves leaders can make to treat data and safety as one system.

The green dashboard

It is the monthly business review. The coffee is going cold, and the slides are being discussed. The AI-generated dashboard shows green across the board. Delivery on track. Risks contained. Customer sentiment stable.

Everyone from operations in the room knows it should be showing amber. One or two of them suspect it should be red. Nobody says so.

Nobody is lying, exactly. The data is accurate as far as it goes. It just leaves out everything people have learned not to say. And the AI, doing exactly what it was built to do, has summarised that silence beautifully.

That is the problem I want to explore here. Your AI is only as honest as your culture.

The pull leaders feel

It strikes me that business leaders are being pulled in two directions simultaneously.

From one side comes AI. Every AI programme I see wants more data. More instrumentation, more dashboards, more decisions grounded in evidence rather than gut feel. Boards want to know the AI strategy. The message to leaders is simple. Be more data-driven.

From the other side comes Enterprise Agility. Here the emphasis is on emotional intelligence (EQ), trust and psychological safety. Leaders are asked to energise teams and stakeholders, lead with intent rather than instruction, and create space for people to speak up.

Many leaders feel they have to pick a side. Are you the hard-nosed, numbers-first executive or the people-first coach?

I want to be clear that the pull is real. Tracking work can feel like surveillance to the people being tracked. And talking about psychological safety on Monday, then rolling out productivity telemetry on Tuesday, sounds hollow largely because it is.

The misdiagnosis

Here is where I think we get it wrong. The divide is not data versus EQ. It is control versus learning.

Agile has never been anti-data. Scrum is built on empiricism. Lean tells us to go and see. Inspect and adapt is about as evidence-based as management gets. What agile pushed back on was using data to judge people instead of improving the system they work in.

That is the real distinction. When data is used for control, the number becomes a target and a verdict. It ends up in performance reviews and league tables. When data is used for learning, the number becomes a question. It starts a conversation, and the people doing the work own it.

Data used for control Data used for learning
The number is A target and a verdict A question
Who owns it Management, from above The team doing the work
What people do Manage the measure Improve the system
What bad news does Goes underground Surfaces early
What the AI learns from A sanitised version of reality What is actually happening

Table 1. The same data, used two ways.

The economist Charles Goodhart gave us the classic warning. Put simply, once a measure becomes a target, people start improving the measure instead of the thing it was meant to measure. Make velocity a target and story points quietly inflate. Make average handling time a target and customer calls get cut short.

Same data. Different use. Opposite outcome.

Safety is how good data gets made

In the 1990s, Amy Edmondson studied medication errors in hospital teams. She expected the better-led teams to make fewer mistakes. The data said the opposite. They reported more.

When she dug deeper, she found the explanation wasn’t that good teams were careless. It was that people in those teams felt safe enough to report errors. In the teams where people were afraid, mistakes still happened. They just stayed hidden.

That finding has stayed with me because the implication is so uncomfortable. A low-safety organisation does not have less bad news. It has less visible bad news.

Without psychological safety you do not get less data. You get worse data.

Now add AI. Almost every agentic AI tool in the enterprise learns from or summarises organisational information. Status reports, risk logs, tickets, meeting notes. If fear has already sanded the rough edges off that information, the AI inherits that modified dataset. Then it hands the result back to leaders with polish and authority.

In the early days of computing, I remember talking about garbage in, garbage out. It’s an old idea, but what is new is how confident the AI-generated output looks. A human status report that feels a bit too rosy invites a raised eyebrow. A crisp AI summary looks objective, and that makes it harder to challenge.

What AI changes about leading

As AI makes analysis cheaper, producing the numbers stops being what sets a leader apart. Anyone can get a dashboard. Anyone can get a summary.

What becomes scarce is everything around the numbers. Judgment about what they mean. Knowing which question to ask next. Making sense of things when the data conflicts or runs out. And, above all, the trust that makes people bring you the uncomfortable truth before it shows up in the figures.

Those are EQ and human relationship skills. So AI does not reduce the value of emotional intelligence in leadership. It raises it.

I have argued before that there is no such thing as an AI transformation. There is only business transformation that uses AI as one performance improvement mechanism among others. The same logic applies here. You cannot bolt AI onto a fearful culture and expect better decisions. You get faster, more confident versions of the same blind spots.

The regulated-sector twist

In financial services and other regulated sectors, this problem has an extra edge. Accountability is often personal. In the UK, the Senior Managers and Certification Regime puts named individuals on the hook. Everything leaves an audit trail.

The natural human response is a defensive data culture. Reports get written for the file rather than for learning. The dashboard stops being a tool for improvement and becomes evidence for the prosecution. People choose their words carefully, and the AI reads exactly those words.

In a recent assignment, management felt that some outsourced teams were not performing. Contracts were being waved about and penalty clauses cited. The outsourced team was afraid to speak up for fear of contract termination and job loss. We created a digital twin of the software development process and quickly established that the supplier was not at fault: significant restrictions in the client-provided ecosystem were inhibiting team productivity. The team’s silence came from a defensive, servile culture, and we used the data to break that pattern.

AI governance can go one of two ways here. Done badly, it adds another layer of control and makes people even more cautious. Done well, it does the opposite. It sets out clearly what data is used for and what it will never be used for. It protects the people who raise issues. It keeps learning data separate from performance data. In creating the digital twin, we emphasised that it was measuring the processes, not the people.

Governance done that way is not a brake on AI. It is what makes the organisation’s data worth trusting in the first place.

Four moves for leaders

None of this needs a new framework. It needs leaders to treat data and safety as one system. These four moves are where I would start.

1. Let teams own their metrics first

The people closest to the work should use their data to improve their own work before anyone aggregates it upward. When a team chooses and owns its measures, it has a reason to hold itself accountable. When the measures arrive from above, the team has a reason to manage them.

In one transformation I worked on, leadership defined the required business outcomes but gave the team responsibility for creating the key result metrics for sponsors to sign off. The result was a higher level of commitment to the objective from each individual on the team.

2. Practise data humility in public

When a leader looks at a dashboard and asks out loud, “What might this be missing?”, something shifts in the room. It signals that the numbers are a starting point, not a verdict. It also gives everyone else permission to say what they know.

3. Make AI use transparent

People should know what is being measured, why, and what it will never be used for. Write it down. Stick to it. This is the point where AI governance and psychological safety meet, and it is where trust is either built or lost.

4. Reward bad news delivered early

The first person to say “this is amber” should be thanked, visibly and in front of their peers. Every leader says they want early warning. Teams watch what happens to the messenger and act on that, not on the words.

Back in the room

Go back to that monthly business review. Same coffee, same slides, same green dashboard.

This time the head of operations clears her throat. “I think this should be amber.”

The chair does not frown. She says, “Thank you. Tell me what you’re seeing.”

The next twenty minutes are the most useful of the meeting. The real risks come out. Someone notes that the AI summary needs the context it was missing. Nothing about the data changed. The culture did.

Data tells you what’s happening. Safety decides whether you find out.

So here is my question. At your next business review, who will be the first to say it’s amber? And what will happen to them when they do?

Download the full white paper

Take the full paper into your next leadership conversation. It includes the control-versus-learning table and the four moves in a shareable format.

Download: Your AI Is Only as Honest as Your Culture (PDF)

If you’d like to talk about how AI governance and psychological safety play out in your organisation, get in touch.

About the author

Jon Ward runs Beneficial Consulting, a consultancy focused on Enterprise Agility and AI governance. He is the author of The Supercharged Enterprise and Developing Agile Coaches.

Leave a reply

Your email address will not be published. Required fields are marked *

six − three =