Are companies’ AI promises real or BS?

This week, I spoke at a major annual valuation conference in Singapore hosted by the CFA Society of Singapore and IVSC.

For my keynote, I wanted to answer a question that’s been on my mind, and probably yours too: how do we tell whether a company’s announced AI initiatives are real or BS?

So I went to the data and ran a full study...with the ultimate goal being a framework to better evaluate these AI initiatives from a shareholder perspective.

I scraped every single earnings call from the S&P500 from the past 10 years to compare what was said about AI vs what happened afterward.

This week, I want to share the key findings from that study.

(one of the results was quite surprising imo)

Here are four takeaways I think are worth your attention.

1. AI now comes up on roughly 2/3 of earnings calls

In our sample of S&P 500 earnings calls, executives discussed AI on about 4% of calls in early 2016.

By Q2 2026, that figure was about 68% (!)

Of course, most of this growth turned parabolic post-GPT release few years ago.

So at this point, hearing management mention AI gives you very little basis for choosing one business over another.

2. Execs pump up AI talk heading into planned share sales

I lined up earnings calls around the dates when company officers and directors adopted 10b5-1 trading plans.

There was strong evidence AI talk ramped up heading into share sale dates and declined in the quarters after insiders sold shares.

aka - company insiders on average were potentially using the AI narrative to pump their price.

So one part of diligence I'd add to the list if a company announces major AI initiatives is to check whether there are planned share issuances or insider 10b5-1 trading plans.

3. Companies talk about AI products 3x as much as cost savings

Products and pricing accounted for about 40% of the classified executive AI discussion, versus 13% for cost savings.

My view is that the product side often deserves more attention.

“We’ll use AI to cut costs” can be a lazy answer to a much bigger strategic question.

There is a limit to how much of an existing expense you can remove. A better product can reach new customers, support a higher price or make a previously uneconomic service worth offering.

For example, helping a software company serve smaller customers without an expensive implementation team could open a market it previously couldn’t profitably reach.

4. Product promises had more than 2x the reported success rate

By eight quarters after a promise was made, estimated reported delivery was about 63% for product commitments and 28% for cost-saving commitments.

Taco Bell gives you a concrete example of an AI rollout: in November 2024, its parent reported voice AI in more than 300 U.S. stores and over two million cumulative orders.

The hard part is turning saved time into lower spending.

Suppose AI saves a support team ten minutes per case.

The team can provide better service or handle more customers. But if staffing stays the same, payroll stays the same too. But now there’s also a software bill.

To turn that into savings, management has to change staffing, responsibilities or the way work is organized.

This is where internal politics matter.

Asking employees to redesign a process that could eliminate their own jobs creates an obvious incentive problem.

My takeaways

How has this study impacted my framework for investing?

Anytime an executive talks about AI initiatives, I look for:

• Any planned share sales or issuances (for incentive misalignment)
• Checking if it's product vs cost focused
• Checking team's shipping velocity track record
• DDing jobs boards for the type of AI jobs they are hiring for

Personal

I just hired three new people to help scale our video production.

That means even better YouTube content for you. If you haven’t subscribed yet, check out my YouTube channel.