McKinsey’s “Growth favors the bold: AI as force multiplier” argues that companies capture meaningful growth from AI when they set larger ambitions, redesign commercial workflows, and scale integrated systems rather than collect isolated tools. It challenges seven familiar assumptions, illustrates its case with companies including Reckitt, DBS, and Salesforce, and asks CEOs to concentrate resources on a few valuable domains.
Much of that diagnosis is sound. Automating a weak process rarely repairs it, a technology function cannot own commercial transformation alone, and adoption depends on changes to incentives, skills, governance, and day-to-day work. The article is also more credible when it describes AI as an operating-system change rather than a productivity accessory.
But the force-multiplier metaphor contains a warning the argument does not pursue far enough: a multiplier is indifferent to the quality of what enters it. AI can amplify sound pricing judgment, useful customer insight, and disciplined experimentation. It can also amplify a mistaken market thesis, a convenient proxy, an aggressive sales incentive, or a biased historical pattern. Faster decisions are valuable only when the decision logic deserves to travel faster.
That matters because the evidence presented is more persuasive about association than causation. AI leaders report stronger growth, and individual companies attribute large revenue, margin, or conversion gains to transformation programs. Yet ambitious, well-capitalized companies with capable management are also more likely to adopt AI deeply, redesign workflows, and measure results. They may outperform partly because of those prior advantages. Case figures rarely reveal the counterfactual, the full cost of organizational change, unsuccessful experiments, or how much improvement came from ordinary process discipline bundled into the AI program.
The call for bolder goals can make this problem worse. An aspirational target may force useful rethinking, but it can also turn an uncertain model into a machine for pursuing the wrong number. A pricing engine asked to maximize near-term revenue may erode trust. A propensity model can repeatedly ignore less visible customers. Personalized discounts may lift conversion while teaching buyers to wait for offers. When the objective is incomplete, continuous optimization does not discover wisdom; it exploits the omission.
The article’s reassurance that “good enough” data is often sufficient to begin also needs a sharper boundary. A company may not need a perfect data lake, but data quality is not merely an infrastructure threshold. Missing consent, unstable definitions, selective outcomes, feedback loops, and histories shaped by previous human bias affect what a commercial model learns. Starting with imperfect data can be sensible in a reversible pilot. Embedding that data into pricing, targeting, coaching, and resource allocation across an enterprise is a different risk.
Scale raises the stakes further. Reusable models and standardized decision components reduce cost and spread successful practices, as the article says. They also turn a local error into a systemic one. Human oversight offers limited protection when employees face high recommendation volumes, cannot inspect model reasoning, or are evaluated against the system’s preferred action. A person clicking “approve” is not necessarily exercising judgment.
The addendum is that boldness should describe the quality of the test, not the size of the promise. Before scaling an AI growth engine, leaders need a defensible commercial thesis, causal measurement, explicit harm and margin constraints, named owners for model errors, and kill criteria that survive executive enthusiasm. Pilots should be reversible, comparisons should include the true transformation cost, and customer outcomes should sit beside revenue metrics. AI can multiply a strong commercial system. Until a company proves which kind of system it has built, acceleration is not evidence of progress.