The AI strategy question most companies avoid

Most organizations developing an AI strategy answer a predictable set of questions. Where can we apply AI to existing workflows? What are the highest-impact use cases? How do we build the technical capability? What is the roadmap? These are legitimate questions, and spending time on them produces useful answers.

The question that gets avoided is different: given what AI makes possible, what will we stop doing?

It is the harder question. It has organizational and political dimensions that the technical questions do not. It requires making decisions about people, processes, and capabilities in ways that cause friction. It is easier to add AI to existing operations than to use AI as a reason to restructure them. So most organizations do the easier thing, and their AI strategies become addition strategies: AI added on top of existing operations, increasing complexity and cost rather than changing the fundamental economics of the business.

Why the subtraction question matters

The promise of AI, for most business applications, is not that AI does new things. It is that AI does existing things at lower cost, at higher scale, or with better consistency. A language model that drafts documentation is not doing something that humans cannot do. It is doing something humans currently do, at a cost structure that is fundamentally different.

This distinction matters for strategy. If AI reduces the cost of existing work, the strategic question is what to do with that reduced cost. There are two answers.

The first answer is to capture the cost reduction: use AI to do more of the existing work with fewer resources, reduce headcount or headcount growth, and take the savings as margin improvement. This is the restructuring answer.

The second answer is to reinvest the capacity: use AI to free up human capacity that gets redirected to work that AI cannot do. This is the reallocation answer.

Both answers are legitimate. The mistake is to choose neither: to deploy AI while maintaining existing staffing and workflows unchanged, producing net cost increase rather than cost reduction or capacity reallocation. This is the default outcome when AI strategy answers only the addition questions and avoids the subtraction question.

What avoidance looks like in practice

The organizations that avoid the subtraction question are usually not deliberately avoiding it. The avoidance is structural.

AI strategy is typically owned by technical teams: engineering, IT, data science. These teams are well-positioned to answer the questions about what can be built and how. They are less well-positioned to answer questions about organizational redesign, workforce planning, and which existing processes should be discontinued. Those questions cross into HR, finance, and line-of-business leadership in ways that require different authority and different conversations.

The path of least resistance is to build the AI capability and declare the strategy complete. The organizational implications get deferred to “phase two” or to the business units that will operate the AI systems. Phase two often does not happen on any defined timeline, and the business units inherit new AI tools without clear guidance about what changes.

The result is what it sounds like: new AI capabilities layered on top of unchanged operations. The AI does work; humans also continue doing the same work, or the freed capacity gets absorbed by growth rather than reallocated deliberately. The AI investment improves output quantity without improving unit economics.

The question in its clearest form

The subtraction question has a specific form: if AI reduces the cost of [task] by [amount], what changes about how we staff, price, or structure [the part of the business that depends on this task]?

This question is uncomfortable because it immediately implicates specific people and specific decisions. But it is the only question that connects AI investment to business outcome. An AI strategy that cannot answer it is not a strategy for changing the business with AI. It is a plan for spending money on AI.

The question has to be asked at the level of specific tasks and functions, not at the level of the organization in aggregate. “AI will improve our productivity” is not an answer. “AI will reduce the time our analysts spend on [specific work] by approximately [X]%, and we will use that capacity to [do more analysis / reduce team size / take on higher-complexity work]” is an answer. The specificity is what makes it actionable and what makes it uncomfortable.

What good AI strategy looks like when it includes the subtraction question

Organizations that genuinely answer the subtraction question tend to have AI strategies with different characteristics.

They have a portfolio of use cases with different business models. Some use cases are cost-reduction plays: AI does work that humans currently do, with net reduction in the cost of that work. Some are capacity-expansion plays: AI frees human capacity that gets redirected to higher-value work. Some are new-capability plays: AI enables things that were not economically feasible before. A good AI strategy is explicit about which use cases belong in which category and what the expected business outcome is for each.

They involve the people affected in the planning. Restructuring decisions made without involving the affected teams tend to fail in implementation, or to create organizational damage that offsets the economic benefits. Organizations that handle this well treat the employees whose workflows AI will change as participants in redesigning those workflows, not as recipients of decisions made elsewhere.

They have explicit decisions about what stops. If AI is going to do X, what is the process for winding down the human version of X? What is the timeline? What happens to the people doing it? Leaving these questions open is not the same as answering them generously. It is a form of organizational debt that eventually has to be paid.

They set performance baselines before AI deployment. The only way to measure whether AI has changed the business economics is to have a clear picture of what the economics were before. Organizations that deploy AI without documenting current performance have no way to know whether the AI investment is producing the expected return.

The objection worth taking seriously

There is an objection to the subtraction question that deserves a direct response.

The objection is that organizations should use AI to grow, not to cut: use AI-freed capacity to expand into new markets, take on more customers, and build new products rather than reducing headcount. This is a legitimate choice, and for many organizations it is the right choice.

But making it a legitimate choice requires actually answering the subtraction question. “We will use AI-freed capacity for growth” is only strategically meaningful if there is a specific plan for what the growth looks like, how the freed capacity maps to the growth activities, and what the timeline is. Without that specificity, it is not a growth strategy. It is a way of avoiding the subtraction question while sounding growth-oriented.

The organizations that successfully use AI to grow rather than cut have answered the harder version of the subtraction question: not “what will we stop doing” but “what will we do differently with the capacity AI frees, and how specifically will that produce growth.” That version is just as specific and just as uncomfortable as the cost-reduction version. It just has a different conclusion.

Starting with the question

The practical implication for organizations building an AI strategy is to add the subtraction question explicitly to the process and not to treat it as a downstream implication that will sort itself out.

The question can be introduced at any stage of strategy development, but it is most useful early, when use case selection is still open. At that point, the subtraction question shapes which use cases are prioritized: use cases with clear economic consequences for the business get preference over use cases with unclear consequences, because clear consequences are what a strategy can be built around.

The question also disciplines the ROI conversation. An AI use case whose ROI depends on cost reduction or capacity reallocation that has not been committed to by the relevant business leader does not have a real ROI. It has an optimistic scenario that may or may not happen depending on decisions that have not been made. Making those decisions, or acknowledging that they have not been made, is the practical output of taking the subtraction question seriously.

Most AI strategies will improve if they add one discipline: before finalizing any use case for investment, answer what changes in the business because AI does this work. Not what could change. What will change, specifically, and who has committed to making it change.