Most leadership teams are not short of data. They are short of time, and short of confidence that the numbers in front of them still describe the business as it is this week. That gap is where artificial intelligence has started to earn a place at the top table.
AI for business leaders is less about dashboards than most vendors suggest. It is about shortening the distance between a question and a defensible answer, so a pricing call, a hiring freeze or a market entry decision can be made in days rather than quarters. McKinsey’s 2026 State of AI survey of 1,719 leaders found that around half say AI helps them make better decisions. The same survey found that only 37 percent could attribute any EBIT impact to AI, a figure that was flat on the previous year. Both things are true at once, and the distance between them is what this article is about.
At iSpark we spend a lot of time with executive teams who have bought the tools and not yet changed the decision. Below you will find what AI actually changes about strategic planning, where it helps and where it quietly misleads, a documented example from a US business, the mistakes that waste a first year of investment, and a sequence you can start on this quarter.
What AI Actually Changes About Executive Decision Making
Three things change, and none of them is that the machine decides.
The first is synthesis. A question that used to need two analysts and a fortnight, such as why margin fell in one region and held in another, can be answered in an afternoon because the model reads across finance, CRM and operational systems at once. The second is option count. Leaders normally compare two or three scenarios because building each one is expensive. When scenario generation becomes cheap, you compare twelve, and the outlier nobody would have staffed sometimes wins. The third is the record. A well built decision system writes down what it considered, what it weighted and what it set aside, which is more than most executive committees produce today.
That last point is the one leaders underrate. Our work on AI for executive and strategy teams almost always starts with a stop list rather than a build list, because value appears when a small number of recurring decisions get a repeatable evidence base, not when every meeting acquires a chatbot.
Where AI Driven Decision Making Earns Its Keep
Not every decision benefits. The ones that do share a shape. They repeat, their outcomes are measurable, and the underlying data already exists somewhere in your estate.
| Decision type | What AI contributes | What stays with people | Realistic payback |
|---|---|---|---|
| Pricing and discounting | Elasticity estimates and deal level guidance | Channel and relationship consequences | 3 to 6 months |
| Demand and capacity planning | Forecasts with stated uncertainty | Supplier terms and commercial commitments | 6 to 9 months |
| Portfolio and investment choices | Scenario generation and downside modeling | Risk appetite and sequencing | 9 to 18 months |
| Market entry | Evidence gathering at speed | Judgement on fit and timing | Case by case |
| Cost and productivity reviews | Pattern finding across systems | The people consequences | 3 to 9 months |
The pattern holds across all of them. AI business insights are an input to a judgement, never the judgement. A model can tell you a 4 percent price rise is defensible in one segment. It cannot tell you what your largest account will do about it at renewal.
The honest pros and cons
- Pro: faster evidence, wider option sets, and a written trail of reasoning.
- Pro: consistency, because the same question stops producing three different answers.
- Con: confident output on thin data, which is harder to spot than an obvious error.
- Con: running costs that keep growing after the pilot budget closes.
A Real Example: Moderna and Clinical Dose Decisions
The challenge. Moderna, the Massachusetts based biotechnology company, had a decision bottleneck rather than a data bottleneck. Choosing a vaccine dose to carry into late stage trials means reading thousands of pages of clinical data under time pressure, and the reading capacity of the study team set the pace.
The solution and implementation. Rather than buying one system, Moderna rolled out ChatGPT Enterprise broadly and let staff build their own assistants, supported by an internal AI academy. Within two months employees had created more than 750 custom assistants, and roughly 40 percent of weekly active users had built one themselves. One of them, Dose ID, reviews and visualizes clinical trial data against standard dose selection criteria.
The outcome and business impact. Dose ID produces a rationale, cites its sources and generates charts for review, and it is positioned explicitly as a data analysis assistant to the clinical study team rather than a decision maker, as OpenAI’s account of the deployment sets out. The gain is not headcount. It is that a decision which used to depend on how much data one team could read now has a repeatable, auditable evidence base behind it.
Common Mistakes Leaders Make With AI in Strategic Planning
- Buying a platform before naming a single decision it will improve.
- Measuring adoption and licence usage instead of decision quality.
- Treating model output as a conclusion rather than as evidence to be challenged.
- Ignoring the fact that the data you need usually belongs to another function.
- Skipping the baseline, so nobody can prove anything improved afterwards.
Best Practices: A Short Checklist
- Name the decision, the owner and the current cycle time before any build starts.
- Record how the decision is made today. That is your baseline.
- Insist on sources and uncertainty ranges in every output.
- Agree in advance what evidence would change the answer.
- Review override rates quarterly. High overrides are information, not failure.
- Put running costs in the business case, not just build costs.
How to Get Started
- Pick one recurring decision that happens at least monthly and has a measurable outcome.
- Check the data actually exists and that your team can access it without a new programme.
- Run a narrow build of eight to fourteen weeks against the baseline you recorded.
- Put the output in front of the people who already make the call, and let them argue with it.
- Keep it only if decision quality or cycle time moved. Stop it if neither did.
Future Trends Worth Watching
Three shifts look durable. Agent based tools are starting to run scenario work end to end rather than answering one question at a time, and about 40 percent of larger enterprises now report scaling agents in at least one function. Evidence provenance is becoming a board expectation, which favours systems that can show their working. And operating cost is moving into planning conversations, with around a fifth of organisations already reporting that AI running costs constrain further use.
Key Takeaways
- AI shortens the path from question to defensible answer. It does not remove accountability.
- Decisions that repeat, with measurable outcomes and existing data, are where value appears first.
- Individual productivity gains are common. Enterprise financial impact is not, and the difference is workflow redesign.
- Baselines and override rates matter more than adoption statistics.
Frequently Asked Questions
What does AI for business leaders mean in practical terms?
It means using models to gather evidence, test scenarios and record reasoning, so executives reach defensible decisions faster without handing over judgement or accountability.
Do we need a data warehouse before we start?
No. Begin with one decision whose data already exists in usable form. A full warehouse programme usually delays visible value by a year or more.
How do we measure whether AI improved a decision?
Compare decision quality and cycle time against your current process on the same question, then track outcomes across two quarters rather than opinions after one demonstration.
Can AI replace executive judgement in strategic planning?
No. Models estimate and summarize. They cannot own accountability, read the political context inside a business, or carry the consequences of a decision.
What size of budget does a sensible first project need?
Most useful first projects are small, running eight to fourteen weeks around one defined decision, existing data, and a named owner able to act on the result.
Where to Take This Next
The businesses getting real value from AI strategy are not the ones with the largest licence counts. They are the ones that picked three decisions, measured them honestly, and redesigned how those decisions get made. That is unglamorous work, and it is the work that shows up in the accounts.
If you want an independent read on which of your decisions would benefit and which would not, iSpark runs fixed scope assessments that end with a recommendation, including a recommendation to stop where that is the honest answer. Start with a conversation about one decision that currently takes too long.
Published by iSpark.
