Business strategy with AI

AI can generate market ideas, business models, pricing suggestions, and competitive analyses within minutes. That is exactly what makes it strategically useful — and dangerous.

A convincingly written strategy is not yet a viable business. A model can invent plausible customer segments, inflate market sizes, underestimate competitors, or derive a price from a feature list rather than from real customer value. The better the language, the easier it becomes to mistake polish for economic substance.

The professional question is therefore not: Which business strategy does the AI recommend?

It is: Which economic assumptions in our project are real, which are hypotheses — and what evidence would have to exist before we should risk money, time, and reputation on them?

AI accelerates strategy work. Industry knowledge determines which questions matter. Market evidence determines which answers survive.

A business strategy is not an idea list

Strategy means choice.

A project decides for whom it creates value, which problem it solves better than alternatives, how that value is delivered, how the customer perceives and buys the solution, and how the resulting activity becomes economically sustainable.

A list of 30 monetization ideas is therefore not a strategy. A Business Model Canvas is also, at first, only a structured representation of assumptions.

Strategy emerges when options are deliberately chosen against each other and then tested against reality.

The first resource is industry knowledge

People who know an industry see things that public data rarely captures: which problems customers actually consider urgent, where theoretical processes differ from daily practice, which budgets truly exist, who formally decides and who can effectively block, which quality attributes are simply expected, which regulatory or organizational frictions delay buying decisions — and which “solution” may be technically elegant but operationally unusable.

This knowledge is not an old-fashioned counterweight to AI. It is high-value context.

The more strategically AI is used, the more valuable genuine domain expertise becomes as a filter against plausible but economically irrelevant suggestions.

AI does not replace industry knowledge — it multiplies it

With strong context, AI can structure industry knowledge, form counter-hypotheses, compare market segments, research competitors, and expose transferable patterns from adjacent markets.

Without that foundation, it easily produces generic strategies: subscription, freemium, platform, community, marketplace, premium service, API, consulting, licensing.

All of those models can be useful. Merely naming them tells us nothing about which one works in this project.

AI is therefore often more valuable for systematically exploring and falsifying business hypotheses than for “inventing the business model.”

Map the market landscape before designing the project strategy

For large strategic questions, a two-stage research process is more robust than one giant prompt.

Stage 1: Map the landscape

First investigate the market independently of the team’s favorite product: customer segments, recurring problems, existing solutions, price points, distribution and sales channels, barriers to entry, relevant regulation, business models, successful and failed providers, technological shifts.

Stage 2: Compress it onto the specific project

Only then ask: Which of these patterns actually apply to our concrete initiative?

This sequence reduces the risk that research becomes a machine for collecting arguments in favor of an idea the team already loves.

AI needs real cases, not invented industry folklore

Strategy research should not stop at claims such as “companies in this industry typically use X.”

Important statements should, where possible, be anchored in real cases: Who actually does this? For which customer segment? At what price? Through which channel? What is included in the offering? Since when? What credible signals exist for demand or economic viability?

The instruction to AI should therefore not merely be “give me examples,” but show evidence for the examples.

The market does not begin with TAM

Many strategy decks start with an enormous Total Addressable Market. That can look professional while saying surprisingly little about an early project.

The more important starting point is the reachable problem space.

A market is not attractive because it is large. It becomes strategically interesting when an identifiable segment has a meaningful problem and an accessible solution can create perceivable value.

Bottom-up questions are often more informative: How many realistic customers can we reach within a year? How many of them actually have the problem? Who owns the budget? How long is the sales cycle? What delivery capacity do we have? Which conversion assumptions are plausible?

The customer is rarely a single person

Especially in B2B contexts, multiple roles commonly exist:

AI can model these roles quickly. Only conversations, process knowledge, and real sales data reveal whether the map is correct.

A strategy often fails not because the user sees no value, but because the buyer sees no budget justification or the blocker sees unacceptable risk.

Problem before solution

A strong AI project does not begin with “we will build an agent.” It begins with a testable problem.

A useful problem has at least four properties:

Relevance — it affects an important outcome.

Frequency — it occurs often enough to matter economically.

Intensity — today’s solution consumes time, money, quality, or risk.

Dissatisfaction — current alternatives do not solve it well enough.

This shifts strategy away from technology and toward economic friction.

Jobs, pains, and gains are observations, not imagination

The Value Proposition Canvas usefully separates customer jobs, pains, and desired gains.^[4]

The critical addition is that these elements should be observed and tested, not merely filled in by an AI.

A model can produce an excellent initial hypothesis:

“Marketing teams need to localize campaigns faster, suffer from approval loops, and want consistent quality.”

That becomes strategically valuable only when we know: Does the problem occur frequently? How is it solved today? What does the status quo cost? Who actually suffers from it? What happens if nothing changes?

Demand is an evidence ladder

Not every positive reaction carries the same weight.

A practical evidence ladder can look like this:

AI can organize this evidence. It must not translate interest into willingness to pay.

A value proposition is not a feature list

“Faster,” “smarter,” “AI-powered,” “automated,” or “multi-agent” are not yet value propositions.

An economic value proposition answers:

Which relevant outcome improves for which customer compared with which status quo?

For example:

“Reduces manual pre-screening of incoming tenders from two hours to twenty minutes while keeping every decision traceable in the project dossier.”

Technology has now been translated into an outcome.

Customer value must become quantifiable

Not every benefit can be translated exactly into euros. But economic strategy still needs a plausible value logic.

Typical value dimensions include labor time saved, additional capacity, higher conversion or close rate, lower error cost, lower regulatory or operational risk, shorter time-to-market, higher quality, reduced external purchasing cost, better utilization, and avoided opportunity cost.

The question is not only “What can the solution do?” but which economic state changes because of it?

Price does not come from token cost

A frequent AI-business mistake is technology-centered pricing:

“The API call costs €0.40, so we will sell the output for €2.”

That confuses production cost with customer value.

Price is influenced by at least four forces: perceived customer value, the alternative cost of the status quo, competitive and reference prices, and the provider’s own cost and margin structure.

Model cost matters — but it does not define willingness to pay by itself.

Willingness to pay has to be tested

Customers can confirm a problem and still refuse to pay.

Strategy should therefore include early tests such as price interviews, concrete sales conversations, paid pilots, landing pages with real price positioning, offer variants, pre-orders where appropriate, and comparison with alternatives customers already pay for.

The most valuable question is often not “Would you use this?” but:

“How do you solve this today — and what budget already flows into that solution?”

Revenue model and pricing are two different decisions

The same service can be monetized in different ways: a one-time project fee, subscription, usage-based pricing, seat-based pricing, license, retainer, success fee, service plus software, freemium with paid expansion, or internal cost-center allocation.

The revenue model describes how money flows. Pricing determines the amount and structure of the price.

Both must fit usage behavior, procurement logic, and the underlying cost structure.

A business model means creating, delivering, and capturing value

A viable architecture connects three layers:

Value Creation

Which capability actually creates the benefit?

Value Delivery

How does the service reliably reach the customer — including sales, onboarding, integration, support, and quality assurance?

Value Capture

How is part of the created value converted into revenue and contribution margin?

Many AI projects are impressive at the first layer and unfinished at the other two.

Cost structure: the invisible half of strategy

AI products incur much more than model cost.

A realistic cost map can include model and API usage, search and data providers, hosting and storage, local hardware and depreciation, development and maintenance, observability and evals, human review time, support, sales, data licensing, compliance and legal review, integrations, outage and fallback capacity, and refunds or remediation.

Cheap inference can still produce an expensive service when review, support, and customer acquisition dominate the economics.

Variable AI costs change project economics

Traditional software can have very low marginal cost per additional user. Agentic AI can behave differently.

Long contexts, multi-agent runs, deep research, image/video generation, and tool-heavy execution can produce substantial variable costs.

Strategy therefore needs metrics such as Cost per Task, Cost per Successful Task, Cost per Customer, Human Review Minutes per Task, Retry Rate, Failure Cost, and a Provider/Fallback Premium.

Only then does “AI is cheap” become a testable statement.

Contribution margin before scaling fantasies

A simple baseline equation is:

Contribution margin per unit = selling price − variable cost per unit.

If a service produces €100 in revenue but uses €35 in model/tool cost, €25 in human review, and €15 in support, €25 remains to cover fixed costs and profit.

Growth amplifies a bad model. If each additional unit has negative contribution, scaling is not success.

Break-even is a project parameter

For simple models:

Break-even quantity = fixed costs ÷ (price per unit − variable cost per unit).

This is not a complete financial plan. But it forces price, cost, and sales volume into the same reality model.

That is why KfW pairs business planning with financial, liquidity, and profitability planning for startup financing.^[2]

AI value often comes from the process, not the model

Current enterprise research shows a recurring pattern: economic value does not automatically follow from access to a capable model.

McKinsey identifies workflow redesign as a particularly important factor associated with EBIT impact.^[7] The OECD emphasizes complementary capabilities beyond models and compute, including data, digital infrastructure, management capability, skills, and finance.^[6]

For projects, the implication is simple:

“We use a better model” is rarely a complete business strategy.

The real question is which process now works differently.

Core use case before AI decoration

An AI feature can generate attention without carrying the business model.

Every strategy should therefore ask: Is AI constitutive of the customer value? Or does it mainly improve convenience? Would the customer still buy the solution without AI? Which part of the value comes from data, workflow, distribution, or expertise? What remains valuable if a competitor gets access to the same base model tomorrow?

These questions also matter in funding and investor conversations.

The dangerous “wrapper moat”

If the only competitive advantage is “we use Model X with a strong prompt,” the advantage can disappear quickly.

More durable differentiation can come from proprietary or hard-to-access domain data, deep workflow integration, superior distribution, trust and reputation, regulation-ready processes, network effects, workflow lock-in without abusive dependency, proprietary evaluation and quality data, specialized user experience, customer relationships and service capability, and intellectual property.

Not every company needs a technological fortress. But every strategy needs an answer to: Why should this value not become immediately interchangeable?