Talking about AI with boards (without the "left behind" spiral)

Nearly every board AI conversation I’m in starts the same way.

“We’re worried about being left behind.”

That’s a completely natural initial reaction right now. People who are working at the edge of AI capabilities are struggling to keep up with all the change.

The real questions people should answer are:

• What will actually change?
• Where is the biggest opportunity?
• What’s the threat?
• How long until we see ROI?

If you can answer those clearly, you don’t need big claims.

You just need to turn AI from a topic into a set of business-centric decisions.

The problem: most people lead with noise

AI is big, shiny and often seems magical.

There are: Demos. Tools. Pilots. Workshops. Agents. Another vendor.

A lot of activity. Not much clarity.

Just noise.

The board doesn’t need a tour of the models.

It needs a view on:

• what changes,
• how you’ll govern it, and
• how you’ll know it’s working.

Start with the bet (not the tool)

A strategic bet isn’t a gamble. It’s a testable hypothesis about behaviour. The simplest way to talk about AI with a board is to reduce it to the following:

1. Which workflow are we changing? (Be specific.)
2. Baseline: what does it cost today? (time, money, risk, delay)
3. Hypothesis: what do we believe will change?
4. Signal: what outcome will move if it’s working?
5. Decision rule: what do we do if it works (scale?), and what do we do if it doesn’t?
6. Guardrails: what are the boundaries? (data, escalation, audit)

If you don’t know this, you really just have a brainstorm.

A note on ROI (because boards will ask)

Be realistic about timeframes. Most organisations see the first returns in phases:

• Weeks: assistive use inside a team (speed + consistency)
• Months: workflow changes with human-in-loop controls (throughput + quality)
• Quarters: deeper operating model shifts (roles, handoffs, systems)

If you present “transformational ROI” with no timeline, you’re basically asking the board to fund belief.

Where the opportunity usually is

The honest answer is: where the business has friction.

In most organisations, the first real gains aren’t glamorous. They’re operational:

• customer support triage / faster resolution
• internal knowledge retrieval (stop reinventing answers)
• sales/admin workflows (drafting, summarising, routing, QA)
• finance/ops tasks that are repetitive, rules-based, and error-prone

AI works best where the system is already defined and the constraint is time, throughput, and consistency.

But note: if the underlying process is broken, AI will scale the mess faster.

What the threat actually is

There are threats. Most board decks ignore them (the old elephant in the room thing), which is why boards stay nervous.

The threat isn’t competitors will use AI. The threats are more boring — and more dangerous:

• Execution risk: pilots everywhere, adoption nowhere
• Quality risk: plausible errors entering real workflows
• Data/vendor risk: unclear boundaries, hidden dependencies
• Reputational risk: one visible failure breaks trust

A board doesn’t need certainty, but it does need to see that you’re addressing them rather than hoping they don’t happen.

Here are a few examples of what can go wrong:

1. Apple Intelligence notification summaries paused (January 2025)

What happened: Apple paused AI notification summaries for news/entertainment apps in iOS 18.3 after widely reported inaccurate summaries. (BBC, 2025)

Failure: “Authoritative” summarisation of third-party content without sufficient accuracy guarantees.

2. Deloitte Australia refunded government work after AI-generated errors (July 2025)

What happened: Deloitte Australia agreed to a partial refund after a government report contained apparent AI-generated issues including fabricated references; a revised version was issued later. (The Guardian, 2025)

Failure: AI-assisted drafting without robust citation verification and QA for a formal, high-stakes deliverable.

There are plenty more examples. The point isn’t to panic. The point is to treat quality and trust as first-order design constraints.

You can read more AI fails here.

Final thought

The fear of being left behind is a signal. Not a plan.

A good board conversation turns that signal into:

one or two clear bets,

outcome-linked signals,

realistic timelines,

and guardrails that match the risk.

Then you can move from “AI strategy” as a topic to “AI” as part of the operating system of the company.