If one of your New Year’s resolutions is to finally put AI to work for your business, let me share a few lessons from my own experience. I hope they will prove useful.
One important caveat: I am a marketing professional using AI, not an AI specialist. So, this is a practitioner’s point of view. And on a very specific subject: brand in asset management.
Together with my colleague Markus Kramer, we recently gave AI a go to help us analyse the brand architecture of 630 asset managers worldwide, as part of the Responsible Investment Brand Index.
Did it work? The answer is much less straightforward than I had thought.
Here are the key takeaways.
Tip #1: Focus on a few things AI can genuinely improve
When I started, I expected significant productivity gains and assumed AI would help automate almost everything. The reality is more nuanced. AI excels at certain tasks and is far less effective at others. The key is to identify where AI truly adds value and where a simple Excel formula does the job just as well. AI may be smart, but it is not particularly fast, so it should be used where the impact really matters.
Tip #2: Remember what LLMs are built for
They are called Large Language Models for a reason. They are impressive at writing, summarising, and analysing text. They are far less impressive when it comes to analysing structured data or identifying complex quantitative patterns.
Tip #3: Stay in command
Just as aircraft still have human pilots more than 50 years after autopilots became standard, you must remain in control. AI models can lack contextual understanding, and it is your role to recognise when this happens and steer the model back on course.
Tip #4: Everything starts with the prompt
The quality of the output depends directly on the quality of the prompt. This is where most of your time should be invested. There are plenty of resources available online with advices on how to prompt, but the real progress comes through incremental refinement and iteration.
Tip #5: The prompt belongs to the business
The best AI experts do not know your business well enough to write the prompt for you. They can help with structure, optimisation and improvements in efficiency, but the business expert must remain in charge of defining what needs to be asked and why.
Tip #6: Be very specific in your prompts
LLMs often want to please you and provide an answer at all costs. To balance this, you need to be very specific in your prompts, not only regarding the information you are looking for, but also the format in which you want the answer.
Tip #7: Choose tools for integration
Make a conscious choice between ChatGPT, Gemini, Copilot, Claude, or any other model but not based on the countless articles comparing marginal differences between them. What matters most is how smoothly the tool integrates with your existing systems. A poor interface will cost you far more than the slight theoretical advantage of one model over another.
Tip #8: Ask AI to justify and explain its answers
LLMs are fascinating because of the answers they provide. They become even more intriguing when you ask them to explain their “thought processes” behind these answers. The justification can sometimes be lengthy, but its structure and robustness are undeniable. In my own research project, which involved analysing the brands of over 600 asset managers across 15 dimensions, the depth and structure of the reasoning provided by AI was truly invaluable, and far beyond the capabilities of humans.
Experience (still) beats AI
I started this journey full of hope and dreams of dramatic productivity gains. I did achieve some, but my expectations were probably too high. What I gained instead was a far more robust and structured analysis than before. When it comes to comparing 630 companies using the exact same factual framework, machines outperform humans by bringing objectivity and comparability to an entirely new level. Even on a subject considered as subjective as brand.
However, my work also confirmed a simple truth: while AI can speed up analysis and improve comparability, experience remains crucial, and the skills developed over time are now more important than ever for understanding context and improving the reliability of the results.
I hope some of you will find these thoughts useful. For the record, this article was written by a human being – namely, me. Happy New Year!
