AI matters for strategy because it changes how much search, representation, and aggregation can happen before a decision. A management team can consider only so many options, hold only so much information in mind, and deliberate for only so long. AI can move some of those limits outward.
That does not make AI the strategist. It changes the system that does strategy.
The bottleneck
Strategy has always been shaped by bounded rationality. Firms search locally because they cannot inspect every possible move. They use simplified representations because the world is too large to model completely. They rely on meetings, hierarchy, consensus, delegation, and committees because knowledge and preferences are spread across people. AI does not remove these limits, but it can make some work that was too slow, expensive, or cognitively demanding routine enough to include.
What AI changes
Search: AI can generate, cluster, criticize, and compare far more alternatives than a team would normally produce. More options are not automatically better. Someone still has to define the objective, screen the options, and decide when the search has gone far enough.
Representation: Strategy depends on models of customers, competitors, capabilities, technologies, and possible futures. AI can build richer and more frequently updated representations. A richer model can still be wrong: it can include too much, omit the decisive variable, or give a false sense of precision.
Aggregation: Organizations already combine partial judgments through voting, averaging, hierarchy, consensus, and delegation. AI adds model ensembles, simulated stakeholders, creator–critic workflows, adversarial reviews, and human–AI forecasts. Aggregation decides which signal counts, whose objection blocks action, and when disagreement becomes noise.
These changes reinforce one another. A broader search needs a representation to decide where to look. A representation often combines many partial views. Aggregation can improve a representation or distort it.
What we know so far
The strongest evidence comes from bounded tasks where strategic judgment can be scored.
- In Artificial intelligence and strategic decision-making, GPT-3.5-generated accelerator applications scored 0.14 standard deviations higher than entrepreneur-written versions and were five percentage points more likely to receive an acceptance recommendation. The model’s business-plan evaluations correlated 0.52 with the average scores of venture-capital and angel investors.
- In The strategic foresight of LLMs, models and humans ranked live Kickstarter technology ventures before outcomes were known. The best LLM correctly ordered about 79% of venture pairs; the strongest MBA-trained investor reached 67%. The tested human–AI combinations did not improve on the best standalone model.
The right interpretation is bounded. These studies show that frontier models can perform well on specific strategic tasks with clear information and measurable outcomes. They do not show that AI can run companies, choose objectives, or take responsibility for consequences.
Capability is not authority
The useful question is not whether AI can “do strategy” in general. That question is too broad to test. The better question is which strategic tasks can be specified, measured, compared, and governed.
The introductory essay to the Strategy Science special issue develops this distinction as two ladders: a Causal Ladder for what kind of reasoning a task requires and a Delegation Ladder for how much authority an organization should grant.
Capability is performance on a task. Delegation is a grant of authority. They are not the same.
A model might forecast well and still remain advisory because the stakes are high, the objective is contested, or the downside is hard to reverse. A narrower system might receive more autonomy because feedback is fast, errors are bounded, and accountability is clear.
The right comparison is between decision systems: old process versus redesigned human–AI process, judged on a task that can actually be evaluated.
What remains human
People still frame the problem, choose objectives, define unacceptable risks, supply proprietary context, judge whether evidence applies, resolve value conflicts, and own the consequences. Organizations still decide who has authority, who can object, and which errors are tolerable.
The point is not to keep people generically “in the loop.” The work is to decide where the loop belongs: before the model, after the model, around the model, or in some narrow settings, at a monitoring distance.
AI does not make strategy automatic. It makes weak strategy processes more visible. If a firm cannot state what good judgment would look like, what evidence would change its mind, or who owns the decision, adding AI will mostly produce more analysis.
The careful claim
AI can expand the cognitive work of strategy, especially where tasks can be specified, information can be standardized, and outcomes can be measured. The strategic problem is to design, test, and govern human–AI decision systems while keeping accountability attached to the people and organizations that act.
Where to read next
- For the theoretical argument, read Unbounding rationality: Why AI is a fundamental issue for strategy.
- For a shorter practitioner-facing version, read AI is revolutionizing strategic decision-making.
- For evidence on generation and evaluation, read Artificial intelligence and strategic decision-making.
- For prospective evidence on foresight, read The strategic foresight of LLMs.
- For a broader route through the research, use the reading map or the research index.