Hello Reader!

The problem with provocative statistics.

MIT dropped a stat this week that made headlines and rattled markets:
​“95% of companies are seeing zero return from GenAI.”

Let’s pause right there.

Because we’ve read the study. And while the branding of the report screams MIT, the methodology reads more like a slide deck written to impress a cynical boardroom instead of academic research.

Here’s what we found:

The sample? Small.
The sourcing? Murky.

Most of the data comes from desk research. No transparency on who was interviewed or how they were selected. There were 153 conversations at 4 conferences they supposedly surveyed at. One of our team joked that a master's student would be grilled for submitting this as thesis work.
And they’d be right.

Even the Financial Times said the report “reads like something from the research page of a large consultancy.” Ouch.

The definition of ‘zero return’? Completely missing.

What counts as a return?

Is it time saved? Better decisions? Less churn? Happier customers?

If you can’t define what counts as success, you can’t credibly report on failure. And yet, here we are.

So what is true?

Plenty of GenAI projects do fail. But not for the reasons MIT suggests.

They fail because organisations start too big.They throw a whole lot of money at platforms no one uses.

That’s why we recommend a different path: start with AI literacy, build momentum through automation, and only then scale into larger AI projects.

Here’s what we’ve seen actually works:

1. Train your people before anything else.​
   When people understand AI and have used it in their own role, that's when ideas start to surface. Real ones. About how to work differently, smarter, better. Allow them to flow.
2. Show, don’t tell.​
   People adopt when it directly solves their problem. Let them try AI. Give them time to master it. Ethan Mollick states that everyone needs at least 10 hours using a model to get to know it well.
3. Execs need to walk the walk.
   When leaders use AI themselves, it signals this is a shift and a priority. Companies with visible exec buy-in are 1.7x more likely to succeed with GenAI. We're currently delivering Csuite Copilot Training in boardrooms all over Ireland. Book a call to discover more.​
4. Shadow AI is not a threat.
   It's a clue. And it shows you where AI is already helping and who's ready to lead the charge. But don’t leave it unchecked for too long: without the right guardrails, data security risks can creep in. 😨
5. Invest in one small win.
   Pick one task your team hates doing. Use AI to fix it. One thing the MIT reesarch shows is that external builders are more likely to succeed than internal teams. So bring them in if you need to.

The most important thing: just start.

ChattingGPT Podcast

What We’re Consuming This Week

🥊 Elon Musk tried to rope in Zuckerberg to take over OpenAI for $97.4bn​
​A joint bid to take over OpenAI. Wasn't it only last year they were challenging each other to cage fights?

⚡ Sam Altman warns of a bubble, while talking up a bubble​
Last week he warned of an AI bubble where someone will lose a "phenomenal amount of money," while casually mentioning that OpenAI would "spend trillions on datacenter construction" and serve "billions daily."

🤖 Dario Amodei speaking sense on how to properly value frontier model companies​
On how he treats models as a company... so that the same time as you're reaping the benefits from one company, you're founding another company that's much more expensive and requires more R&D. View YouTube short here.​

Our Autumn Courses Are Filling Fast

Our popular courses are back - and places are filling fast. Whether you’re looking to master the essentials, get hands-on with Copilot, design agentic strategies, or unlock the power of automations, we’ve got a course tailored for you.

• ​AI Core Skills (11 Sept)​
• ​Copilot Masterclass (25 Sept)​
• ​AI Agentic Strategy for Marketing (9 Oct)​
• ​AI Automations (4 Nov)​
  ​

See you next time!

The AI Institute Team