Dear Reader,

To quote the comedian, Vic Reeves:

In so many conversations that I have with leaders, they want numbers to give them confidence that they are on the right path with AI. Yet, CFOs operate on the basis that words talk, numbers scream.

But is there any good data or is it all Lies, Damn Lies and AI Statistics, to paraphrase Disraeli (who was probably quoting someone else).

How does this show up?

Every month, reports claim "mind-blowing" capabilities and/or a jobs apocalypse.

This week it was Bill Gates' turn to contribute 6000-ish words. I haven't done a dissection of Bill's essay yet, but you can find my marking of Mark Zuckerberg's homework here. And my hot-take on the Pope's contribution here. In many of these tomes, there's little in the way of data or numbers.

We also have headline grabbing statements like "95% of enterprise AI pilots delivered no measurable impact on the bottom line according to MIT." And I've even used some of these myself in the past, mea culpa. But three months ago, I gave myself a self-imposed moratorium on bandying about random statistics as they didn't seem to have good provenance.

This week I've done a deep dive into some of the underlying studies and here are some examples of what is supported by evidence, rather than not.

The Downright Dubious

No one's getting ROI

"Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return. [...] Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact."

Source: The GenAI Divide: State of AI in Business 2025, Project NANDA, July/August 2025. Authors: Aditya Challapally, Chris Pease, Ramesh Raskar, Pradyumna Chari

Strangely, I couldn't find the report on the MIT website.

The sample size appears small (52 interviews, 300 projects analysed, 153 leaders surveyed) and the criteria was P&L impact at 6 months.

My Conclusion: I'm not sure this is telling us anything apart from mostly the answer to the question "Is AI delivering ROI?" is No One Knows.

My Advice: Stop focussing on other organisations. Ask yourself the question, "How might we test more things, better, faster, cheaper to improve our outcomes?" and AI's a fantastic prototyping tool.

The Caveated

"Nearly 40% of companies that measured AI cost savings landed below 10%, despite targeting 11% to 20%... yet 90% are increasing their budgets again."

Source: Bain & Company, Automation and AI Pathfinder Survey 2026​

There are 951 respondents, so a good sample size. It's not clear how robust the measurement of cost saving is vs self reporting and how many of the 951 measure their AI cost savings. It covers automation as well as AI (but in reality many AI projects are automation with a souped up label). The targeted savings are potential cost reduction as a percentage. It doesn't have enough context to understand ROI and the scope of what's being tackled - a small focused agent or a complete process redesign.

My conclusion: The survey has some great insights on where leaders think the obstacles are. And it's clear that this area is going to continue to gather investment. But there isn't enough information to decide if this is good money after bad, or a learning curve, or noise.

My Advice: Focus energy, time, money and people on using AI where performance improvement moves the needle on the business. Sometimes that might be cost. Sometimes that might be revenue growth, customer satisfaction or better decision making.

The Tip Top

Who's using AI?

It's a large sample size (3500). However, despite being published Feb 2026, the field work finished in May 2025.

My Conclusion: Most businesses are at the start of their journey, despite all the hype.

My Advice: Channel September's back to school energy and get started.

Where are those productivity gains?

One company, one AI tool. However, it covered over 5172 customer service staff, over 3 million chats and measured outcomes not sentiments from a survey. From 2020-21 fieldwork using a GPT-3 era model.

My Conclusion: There are productivity gains, but mainly for new joiners and lower skilled staff. The experienced staff have lower individual benefit, but their input is vital to train the tool. And the main productivity benefit may come from upskilling workers rather than outsourcing activity to the tool. Or put another way... Ooh, interesting. That said, we don't know how this picture might have changed with newer reasoning models.

My Advice: Like all good innovations, baselines and deciding what outcomes you are trying to influence are key to measuring any sort of improvement. And AI's an ingredient not a meal, so the outcomes you want from AI productivity gains are generally already in your business strategy.

In Summary

The "MIT" figure is everywhere and is widely debunked. I've seen few people quote the DSIT report, and yet it's telling us that most people are still in first gear when it comes to Agentic AI, including your competitors.

So, focus on yourself and your own organisation. Find things that move the needle on your business. Taking friction out is a great place to start.

If there's a PowerPoint floating round your business with an AI statistic that no one has checked, send it to me and I'll tell you if it's tip top, caveated or downright dubious. I always reply to replies.

Until next time,

Helen

P.S. I'll mark Bill's homework next week, via LinkedIn Live

If a friend shared this with you, and you want to join the crowd finding ROI from AI, hop on the list here.