Apparently I Trendslop Too
All major frontier models lean toward the same strategic advice. My own preferences are surprisingly close. What this means for strategy work with AI.
Originally published on LinkedIn.
AI as business strategy advisor has entered the mainstream. Adventurous CEOs have been experimenting with it for a while. Consultants are fully embracing it. We at KRS certainly use it as well for our strategy work. Now a recent HBR study found that this might be misguided as all major frontier models systematically lean toward the same kind of strategic advice. The authors call it "trendslop", a strong bias towards popular management trends. Awkwardly, skimming the study, I noticed my own preferences closely match those of the LLMs. Apparently I trendslop too.
The authors tested seven core business tensions. On six of them, the models all leaned strongly toward the same side: differentiation over commoditisation, augmentation over automation, long-term over short-term performance, decentralisation over centralisation, incremental over radical innovation, collaboration over competition. Only exploitation vs exploration split the field. The article calls these strategies "aligned with modern managerial buzzwords and trends". That hurts. So, where do my personal biases or preferences come from?
Where my preferences come from
Take differentiation vs commoditisation. Without any specific product or service in mind, my intuition draws me toward differentiation. "Competing on price is a losing battle." That impulse probably comes from working mainly with businesses in the past that struggled against price-based competitors and tried to protect their margins through performance differentiation or services. A well-known story in European industrial companies facing low-cost competition from abroad. I am aware, of course, that successful businesses built on price competitiveness exist, but my reflex dismisses them as exceptions, always on the lookout for profit-improving differentiation. Whether my bias is representative or just trendslop with footnotes is hard to tell from the inside.
On automation vs augmentation, I am immediately drawn to augmentation. I have seen my fair share of AI initiatives, dozens of them across very different organisations, and the pattern is consistent: full automation works for very narrow, well-defined tasks. Low ambiguity, little uncertainty. The moment context, judgement, complexity, or edge cases enter the picture, augmentation under human supervision wins. My intuition tells me that there is much more opportunity in augmentation than in automation, at least today. Same lean as the LLMs. This time, though, I am willing to defend the position. I would argue this holds for the majority of firms today.
I could go tension by tension and find arguments for the lean. But that's not where I want to go. The authors show that adding detailed context does shift some of the LLMs' answers, sometimes substantially: centralisation and innovation biases largely dissolve, others soften. But for collaboration and augmentation, the lean persists even with detailed context. The question I wonder about is whether there is a good reason for that. Put differently: if we sampled successful companies' strategic choices on those persistent tensions, would the distribution be balanced, or would it lean toward the poles the LLMs prefer? That is the question I am leaving this article with.
What this means for strategy work with AI
Independent of whether there is some truth behind these specific biases, I have little doubt about inherent biases baked into today's LLMs. Therefore, the article's practical guidance on how to work with LLMs for strategy work is very solid. 1. Provide detailed, relevant information as context (our favourite topic at KRS). 2. Ask for benefits and risks of strategic counterpositions. 3. Ask for examples of companies that succeeded or failed with each option. This points to a broader principle for working with AI: the objective is to use the AI to surface strong positions and arguments you would not have produced on your own. Then you work with AI to analyse, evaluate, and weigh your options. Finally, you, and you alone, decide and take responsibility for it. That is strategic-thinking augmentation vs automation right there.