Habr AI: LLMs Can Take Over Routine Business Research, but Not Strategy

A long analysis of the future of product and marketing research boils down to a simple idea: LLMs are not equally useful for every type of research work. The more rigidly the frame is set in advance (what to measure, how to ask, and how to interpret the answer), the easier the work is to automate.

Three levels of research

The author suggests viewing business research not as a set of separate methods but as work with a "knowledge matrix": the system of distinctions through which a company describes its market, its users and its product. At the first level, the matrix is already built, and the researcher simply fills it with data. At the second level, the frame is refined as the work proceeds: categories change, segments are reassembled, and the user behavior model is gradually fitted to reality. At the third level, the subject of analysis is the frame itself: how a company defines value, problem, loyalty or success in the first place.

The text's main conclusion follows from this logic: the question is not whether "an LLM can do research at all" but which level a specific task sits on. If a business needs to measure predefined parameters, a model will handle it much sooner. If the team needs to rethink the very concepts it uses to describe the market, current language models alone are not enough. That is why the debate about fully replacing researchers looks too crude here: different classes of tasks will be automated at different speeds.

Where LLMs are already strong

The most obvious candidate for automation is first-level research. Here the company already has ready-made metrics, question templates and interpretation rules. Essentially, the job is not to find a new research logic but to execute a formalized procedure quickly. That is why the author believes many such tasks could in principle have been automated earlier, and LLMs merely sharply cut the cost and lower the barrier to entry.

• sales funnels, NPS, CSI and other metrics with fixed calculation rules
• one-off A/B tests comparing predefined options
• pricing studies such as Van Westendorp, Gabor-Granger and conjoint approaches
• structured CustDev interviews and usability tests with a rigid script
• feature prioritization through models such as Kano, MaxDiff and TURF

"LLMs do not create a new class of capabilities; they only remove the

costs of formalization."

Second-level tasks are more nuanced. A prompt or RAG alone is not enough here: the model must not only process responses but also gradually refine the very set of distinctions on which the analysis is built. The article names LoRA and Representation Engineering among suitable approaches. These are methods that change a model's weights or activations and thereby let you adjust its semantic field. In other words, the author allows that LLMs may help with segmenting complex audiences, building decision-making models and refining research categories, but this is no longer a "chat with documents"; it is a deeper tuning of the system.

Where the limit lies

The main constraint begins at the third level, where research must neither fill in nor refine an existing frame but take it apart and reassemble it. These are tasks in which the team asks not "why is NPS falling" but "what exactly do we call loyalty, and why do we consider it important." This also covers studies of brand language, cultural codes, organizational discourses and the strategic concepts through which a company sees its problems and opportunities at all.

In the author's view, the current LLM architecture hits a fundamental limit here. A model can generate interpretations, argue with itself in a multi-agent setup and even use self-reflection, but all of this remains work within a single system of distinctions. Such a loop can improve an answer, but it does not turn the model itself into an object of lasting transformation. So the model can support a researcher, suggest moves and speed up analysis, but it does not replace a human where the research lens itself has to be rebuilt.

What this means

The practical conclusion is blunt: most of what is called product and marketing research today will become an automatable service. Human value will shift to where the job is not just to count, compare and code answers but to change the language in which a problem is posed, notice hidden frames and combine business, culture and strategy into one research picture.

Hence the author's forecast: instead of rigidly separated roles such as UX researcher, CX manager or marketing analyst, demand will grow for curator-researchers who can manage an ensemble of AI tools.