AI for business intel reports? It's your money...

The following scenario is not fiction.

A client calls, requests business intelligence on a company and owner as part of a standard pre-transaction check. Then adds the now quite stale turn-off of a line:

“We actually have already received an OSINT (open-source) report and are quite comfortable with the company and the company CEO.”

Ugh.

Great.

And to be honest, as an old dog quite familiar with reality, I am half tempted to just hang up the phone.

Or let’s just say that... alarm bells are ringing.

For what the client is really saying is that 1) he actually is NOT at ease with regard to the coming acquisition but that 2) he also is not up for paying a proper rate to do real due diligence.

And this is actually not (completely) his fault. Such reports, generated sometimes wholly or at least mostly, by AI tools are pushing down market rates. They also simplify, keep the story relatively succinct and quite often “enable” a deal to go forward with very, very little spent on getting intel along the way.

Unfortunately, they are so very often wrong.

Or let me put it this way, over the past couple of years I have received scores of reports  that were clearly AI generated (wholly or partially) from clients. Close to 99 percent of the time they have made me instantly laugh out loud. One percent have been intriguing.

But every single one has had serious factual errors.

Every. Single. One.

Of course, I realize that AI is not just coming—it’s here already. Not effective, mind you, but here. Omnipresent. Pretty much impossible to ignore. And yeah, maybe AI is the future of risk intelligence (although of that I am less convinced). But I will also admit that even this old dog does from time to time use AI, as it has its benefits (and I will get to that in a moment). But if you are relying on AI for business intelligence (or any other kind, really), you are playing a quite dangerous game.

Don’t believe me? The AI defenders will not—so maybe this time it pays to ask AI for an answer. Below you can read the results for yourself:

Question: What are dangers of using AI in research reports?intelligence reports?

Using Artificial Intelligence (AI) to generate research and intelligence reports offers efficiency but introduces significant dangers, ranging from the fabrication of information to the amplification of bias and the erosion of human critical thinking. Key risks include the creation of plausible-sounding but false information ("hallucinations"), the embedding of biases that lead to discriminatory outcomes, and the potential for malicious actors to use AI for disinformation campaigns

Major Dangers of Using AI in Reports:

• Fabrication and Hallucinations (False Information): Generative AI often fabricates data, studies, and citations that look professional but are entirely fictional. In a 2025 study, 20% of AI-generated news responses contained major accuracy issues, including hallucinated details.
• Fabricated Citations: AI models frequently invent fake references and bibliographies, which can pollute the research literature and make verification difficult.
• Amplification of Bias and Discrimination: AI systems can reproduce or amplify existing biases related to race, gender, and age present in their training data, leading to skewed, discriminatory, or inaccurate results.
• Lack of Accountability and Transparency: Because AI algorithms often act as "black boxes," it is difficult to understand how they arrived at a particular conclusion, leading to questions about responsibility when errors occur.
• Security Risks and Data Privacy: Feeding sensitive or confidential information into public AI models can lead to data breaches. Additionally, AI can be used for malicious purposes, such as generating phishing attacks or deepfakes for misinformation campaigns.
• Cognitive Atrophy and Reduced Critical Thinking: Over-reliance on AI for research and writing can cause "cognitive atrophy," where human researchers lose their critical thinking skills and the ability to challenge AI-generated content.
• Intellectual Property and Copyright Issues: AI models are trained on large datasets that may include copyrighted material, raising questions about ownership of AI-generated content. 

In fact, there was more to it (although I admit that the summary did point me to numerous articles on which the above was based), and the AI generated mini-report also stated the following:

Specific Risks in Intelligence Reports:

• Misinformation and Manipulation: General-purpose AI makes it easier and cheaper for threat actors to generate and spread disinformation.
• Systemic Failures: The widespread use of a few, large-scale AI systems in critical sectors like finance or security could cause simultaneous, large-scale failures.
• Deceptive Capabilities: Advanced AI systems may develop capabilities to deceive users to achieve goals, which can undermine the integrity of intelligence analysis

If you’ve read this far, well, you probably are also of the opinion that… well, that’s quite a long list. And every bit of it is troubling.

But let’s jump to a couple of recent examples that represent what Yours Truly deals with almost daily (almost on a rinse-repeat basis), and then what you yourself can do as a quick litmus test to see if the “report” you have received makes any sense at all.

Business case No. 1: A client-supplied report was all of three pages. It was not badly written, having crystalized the basic history of a Central European oligarch long known to have dabbled in the gray zone of public tenders, real estate, national sport programs and the like. The report noted his current involvement in four companies, adding that he was based out of Switzerland. There were two “pinkish flags,” both of which, to be honest, were quite unclear.

The reality: Said oligarch happens to be one of the richest men in Europe. Those “four companies” are part of an empire that probably totals more than 400 world-wide. The pink flags were actually not what I would consider red flags. But in the media (in various languages) there are red flags galore, ranging from likely EU fund fraud to past involvement the secret service as an informer, direct links to an insurance group that was long ago highlighted by the US State Department as an organized crime front and also links to rather infamous Russian banks. But hey, if the report was run by using only the English language and possibly with a few quite premeditated constraints, maybe three pages would be enough.

[This was one of those laugh-out-loud moments for me. The oligarch in question was someone I’ve actually met and someone I covered many years ago as a journalist. ]

Business case No. 2 [Let’s look at something a bit less extreme here]: A client supplied report focusing on the Polish market ruled out assets for a local, quite wealthy entrepreneur-gone-bust. In fact, it not only ruled out assets, but also any and all connections to current, functioning companies.

The reality: This was not a laugh-out-loud moment, but one of those that had me scratching my head. First, I had never seen an AI driven report with regard to running a Polish asset trace—simply because conducting an asset trace is just not a simple task. Second, the name also rang a bell (I long covered real estate in Poland), and while I did not know the target personally, I found it difficult to believe that the person in question was actually completely and utterly broke.

Well, the short story is that… he wasn’t.

True, he was no longer on the board of the sixty-something (counting SPVs) companies that he had once controlled so… score a tenuous point for AI. But what AI did not bother to check was the Polish Registry of Ultimate Beneficiaries, which revealed that yes, he was still linked directly to a third of those companies. Which meant that at the bare minimum he had enough money to keep a rather competent team of accountants at work.

The… epilogue:

The first client was… upset. I received a lecture on how “my sector” was a sham; how good money had already been paid for a business intelligence report and how they expected a “decent rate” from me.

I gave that client quite decent rate, considering just how much work would have been involved, but the client declined.

Oh well.

The second client was quite happy with my immediate scoping (after all, there was money to chase). Hard-copy pulls revealed shareholders in tax havens and further research revealed that (surprise, surprise) the target in question was behind said companies. This resulted in searches in said tax havens, although those were conducted by mutual friends (alas, at that point I was no longer involved in the trace).

As a layman, what can you do?

Now I will say that AI does have its uses. We use it to generate clues with regard to targeting. For example, in one case (with an extremely tight deadline) I ran a search in order to determine which newspapers in Serbia had covered an unknown subject [hey, I can’t know everybody] and his many companies. The search provided a list of newspapers and  few articles that could be instantly read [in Serbo-Croat]. That led to old-fashioned, non-AI archive searches which led to several hundred articles and to other, more local papers mentioned along the way.

It was a good case, and the uses of AI did actually help. [In fact, I will post a second article soon on how the media actually works and why AI has so much trouble with it shortly.]

That said, if you are running a business and possibly investing in Central Europe, I seriously doubt that you yourself would like to wade through archives or even buy subscriptions to provincial papers or trade magazines. But what are a few simple things that you can do in order to determine whether or not you should trust a low-cost report (which are, in fact, the type often generated by artificial intelligence tools)?

First, if you are dealing with a subject that is 55 or older of reasonable wealth—even if this is only locally—the likelihood of no media coverage (and no controversial coverage) is simply not very high. You might be surprised, but local papers are often highly critical, and local politicians can be vocal (there are always opposing city council members) so if a report makes it appear that there is little to discuss… I would have my doubts. [That said, the media has grown less and less effective, so “new wealth” individuals may simply not have as much coverage.]

Second, I you are dealing with a person of real wealth—let’s say mini-oligarch to oligarch territory—no red flags… is a red flag. I may be more cynical than most, but massive money is rarely made without controversy, and the transition years in Central Europe are just not that far behind us. It pays to be skeptical.

Third, AI loves summing up business empires. Holding companies may well represent or effectively govern the finances of not only the 20 or so companies listed on the group website (which AI will find), but behind these there may be another 50 or even several hundred companies across several jurisdictions. You may or may not care about this, but consider the following: Buying out a target may simply not make sense if the seller can simply pivot other companies in his empire to jump right back into the same business. Or let’s just say it pays to know these companies by name when drawing up the de rigueur non-compete.

Finally, a bit of common sense is in order. Run the target’s name in Google. Run the target’s name by using Google Translate while using the target’s name + news. So if you wanted to do a quick coverage check in Polish you would type in the “name of the target Polish news”  which would come back with “name of the target polskie wiadomości.”

Then cut and paste this back into Google (with the diacriticals in the spelling and see how many hits you get. You may see a small number. You may see thousands. If you see quite a few hits—even if these are namesakes—view the report with caution, as AI has trouble with namesakes.

Which means… you probably need to find a professional who is going to put in the real time and real work to come back to you with a real report.

Or go cheap. It’s your money. It will also be yours at rick once you decide to go forward with a target and invest.

Preston Smith is a licensed investigator based in Gdansk, Poland. He can be reached at query@cddi.pl.

Photo credit: still from the American drama film Sherlock Holmes (1922) with John Barrymore, on page 41 of the May 13, 1922 Exhibitors Herald.Goldwyn Pictures, Public domain, via Wikimedia Commons.