Businesses Should Integrate AI, but Shouldn't Rely on It

Its pretty clear that most or all businesses should integrate AI into their workflows. Its great technology for what it does, it can help cut costs, and it can speed up timelines tremendously in some cases, especially for routine tasks. I recommend it, I use it for our work- but there's a catch.

While I could be wrong about this, from what I've seen, AI won't be fully replacing anyone who cares about quality any time soon. And even if it did, it will still likely be used more effectively by experts. AI is probably overhyped.

I've heard it can replace analysts. I've seen it choose the wrong formulas. I've heard it can replace writers. I've seen it write generic garbage. I've heard it can do research. I've seen it collect information from unverified secondary sources that sell services. I've heard it can do anything. I've seen it produce terrible outcomes.

Again, I could be wrong and we monitor this over time as we look to further implementing the tools, but the biggest problem seems to be this: 'black box' understanding. Machine learning (ML) models are best used for singular tasks with plenty of historical data to learn from. Once the scope of duties starts to expand, the models break down (for example a multiple linear regression model can become underfitted to expand its uses- or overfitted and become useless for predictions). And if they don't have historical data to work with, they can't learn well, which also makes them relatively bad at predicting unexpected turning points (though some experts contend this isn't true- check out Machine Learning for Asset Managers, López de Prado, 2020).

Most of the ML models we use for analysis are proprietary and they do fairly specific things. For example, K-Means cluster analysis for industry groupings.

So, we know its not very good at what it does, why do big companies implement it on massive scales? These big companies experience problems with it all the time- I've seen it. Companies are likely doing what their shareholders tell them to. One potential problem with public companies is the short-sightedness of shareholders, whom are overly focused on quarterly earnings despite them having basically no correlation with value (see Investment Valuation, Damodaran, 2025 (updated)). Companies are likely being pushed to cut costs, and some of them think they can replace their entire departments and I look forward to doing business with them.

AI has appeared to be generally, very useful for a few things: efficient searching and reference materials (what is the DuPont ROE decomposition formula? what are the different methods for valuing private companies?), specific tasks (write the code to read an Excel file in Python, pull out the risk factors mentioned in this company's 10-K report), and generic tasks (write my blog, please). Even then, you'll still have to proofread everything, which means you need to be an expert.

You should absolutely use AI, it can cut costs, improve productivity and, as it improves, our team will become even better at what we do, but I advise businesses against relying on it for department needs.

If you found this helpful or enjoyable, please consider subscribing to our newsletter for more.  

You can reach out to us here for consulting services or here to discuss booking a workshop to learn useful skills and help you get more familiar with us.