Where’s the AI-based research in thought leadership today?

Despite plenty of findings about AI, there is almost no sign of AI being used in the research itself.

Recent inspiration

There are some examples of AI-based research, most notably this recent fascinating study from HBR, which uses machine learning to analyse around 1.4m job posts from an online freelancing platform to explore how demand for certain kinds of jobs changed between July 2021 and July 2023.

The report highlights how volumes of freelance writing and coding work, among other things, dropped fairly sharply following the introduction of ChatGPT (yes, many people should be worried).

It’s an excellent example of what a great AI-based B2B thought leadership piece could look like – but it’s an academic study, rather than a corporate research piece. This is by no means a comprehensive review, but concrete examples of B2B research using AI are certainly rare. This Counterparty Radar piece uses machine learning to scrape and aggregate trading data across around 2,000 funds.

Microsoft’s 2023 Will AI Fix Work? study analyses aggregate user behaviour across its Office 365 suite, but essentially as a side-note to a wider survey. This earlier report from PwC uses research that scanned some 2.2m companies across the UK to determine their usage of AR/VR technology. While these pieces are interesting, the most notable conclusion is the widespread absence of AI as a research input.

Learning from academia

All this is in stark contrast with the scientific community, where there is already a significant archive of AI-based research papers. These range from the profound to the whimsical. At one extreme is this example, which uses AI to help uncover links and relationships across 1,000 different scientific papers, discovering structural parallels between biological materials and Beethoven’s 9th Symphony (who knew?), among other things.

Far more usefully, this paper shows how AI can help detect and diagnose dementia and do so better than human-only diagnosis. Most notably, Sir Demis Hassabis and John Jumper of Google DeepMind were two of three joint recipients of a Nobel Prize in chemistry in 2024, thanks to their work in predicting the structure of every known protein using the company’s AlphaFold suite of AI tools. 

Naturally, GenAI is used significantly around the edges of B2B thought leadership – much of which (creating social media posts, generating survey questions, report title options, etc.) is not publicised, for obvious reasons. But more often than not, thought leadership reports remain survey-based. (There are, of course, many excellent examples of non-survey based research, such as econometric indexes.) 

This isn’t the first instance of new technology apparently being underutilised. A decade or so ago, there was a proliferation of reports about the rise of big data and how it was going to change businesses. At the same time, it was nigh-on impossible to find reports actually drawing on analysis of big data to derive their findings. 

Why might this be? 

First, and most obviously, surveys work really well for B2B thought leadership.

Many studies are inherently forward-looking. (See the WEF example mentioned above, which looks at the future of jobs. Polling companies to ask how their job plans are changing is a clearly relevant way to work this out, especially when paired with backward-looking real-world jobs data.) 

Second, original data or research about the issues under discussion can be scarce

The Payhawk report mentioned above, for example, explores the technology gap within companies’ finance technology stacks. It’s exceedingly unlikely that data exists on a topic like this, so conducting a survey, along with a number of one-to-one interviews, is an obvious choice for exploring it.

Third, it’s difficult to get AI-based research right.

Designing a credible study based on non-survey data poses a considerable challenge, and many organisations lack the skills and experience to pull it off. ChatGPT and other GenAI tools can deliver impressive insights, but they need to be tightly prompted and very carefully considered before they go into a report. 

Delivering something that’s actually newsworthy is an even higher-order challenge. To take the HBR example above, this draws on expertise in identifying a suitable idea, finding and accessing relevant data sources, dealing with data extraction and structuring issues, the ability to analyse large data sets, and so on. 

Finally, building on the prior point, there is also a trust gap when it comes to AI.

Are the findings accurate and defensible? Are there any risks to overcome, such as possible hallucinations in the data? 

Once the technology is more dependable and autonomous, AI-based reports could be a future trend. For now, though, expect AI to continue to be more often the subject of reports, rather than the source of their insight.