A new report from McKinsey & Company, Agents for Growth: Turning AI Promise into Impact, addresses a question on the minds of executives across all industries, including robotics and physical AI, Why are so many companies still struggling to generate real returns on their AI investments and what sets companies that are making real progress? The November 2025 report, authored by senior partners Greg Kelly and Steve Reis and partner Lisa Harkness, is specifically focused on marketing, sales and customer service — but its core argument about agentic AI is broadly relevant to anyone contemplating autonomous systems.
The Divide Between Hype and Results
The report opens with a sobering statistic: nearly 80% of organizations report no meaningful bottom-line gains from AI to date, mostly because of siloed pilot projects, weak underlying data, and a lack of governance. “That’s a pretty high failure rate given all the capital and attention thrown at AI adoption in the last several years. McKinsey’s explanation for this gap comes down to a fundamental distinction: most companies are still using generative AI as an assistive tool to help humans get tasks done faster, not deploying real AI agents which can act, decide and collaborate independently across an entire workflow.
That distinction is important because the report estimates that agentic AI will be responsible for more than 60 percent of the added value expected from AI deployments in marketing and sales going forward. Some Fortune 250 companies already report campaign creation and execution accelerating by a factor of 15 and McKinsey estimates fully scaled agent deployments could yield annual productivity gains of three to five percent while boosting growth by 10 percent or more.
Four Lessons from the Leaders
Based on McKinsey’s work across industries, the report synthesizes four lessons on what distinguishes successful agentic AI deployments.
First, focus efforts where the value really is — deploy agents to move the needle on core growth levers like personalization, conversion, and dynamic pricing, instead of diluting efforts. Consider, for instance, a European insurer that used AI agents to tailor campaigns across hundreds of microsegments, doubling or tripling conversion rates, and reducing customer service call times by a quarter.
Lesson two is to redesign entire workflows rather than just bolting agents onto existing processes. The report contends that point-solutions in isolation are of limited impact. It’s not much help improving how customers discover products if purchasing and fulfillment are still slow. That’s what a European insurer of the same stripe did: in 16 weeks the firm reconfigured its commercial model around a network of specialized agents—some centralizing policy knowledge, others grading sales calls, others handling systems integration—that together cut average call times by a quarter and dramatically increased the share of sales calls receiving quality review, from just 3 percent to 95 percent.
The third lesson is to think of AI agents less as software tools and more as team members, with defined roles, onboarding, and metrics for performance. Instead of tracking raw output like call volume, leading organizations are tracking the quality of conversations, the accuracy of task completion, and the speed at which agents learn from feedback. A US homebuilder did this well. They analyzed more than half a million sales transcripts to figure out what their best human sellers did differently, and built AI sales agents modeled on those patterns. In the end, they tripled their appointment conversion rates.
The fourth and final lesson is to redesign the organization itself, around agentic systems. The report notes that leading companies are moving away from the traditional siloed model in which marketing hands off to sales, which hands off to service, and instead building “agent factories”: centralized hubs that standardize how agents are built, deployed, and governed, preventing the redundant, ungoverned sprawl that McKinsey calls “agent chaos.” For example, global banks are restructuring due-diligence workflows with dedicated agent squads and a North American manufacturer has redesigned customer service so agents do diagnosis and data retrieval and humans focus on empathy and resolution.
Why This Matters Beyond Sales & Marketing
The report’s case studies come from digital growth functions, but the framing is worth sitting with for readers focused on robotics and physical AI. The same core tension – whether to treat AI as a bolt-on productivity tool or rebuild workflows and organizational structures around autonomous agents – maps directly onto the challenges facing humanoid and embodied AI deployments today. The report’s focus on governance, models of human-agent collaboration, and measuring agents against new kinds of performance indicators (as opposed to legacy KPIs) offers an early look at organizational challenges that physical AI companies will have to grapple with as their systems move from pilots to fleet-scale deployment.
The report ends on a forward-looking note, with questions that seem particularly pertinent to an industry that’s seeing autonomous systems gain more independence: as AI agents start to negotiate and act on behalf of companies, what will truly differentiate one company’s product or service from another when execution itself becomes a commodity? That question may be more pointed than for marketing and sales, for an industry built on autonomous embodied systems.