Where AI Actually Creates Value

← Back to Intelligence

McKinsey's foundational economic analysis estimates that roughly 75% of generative AI's $2.6 to $4.4 trillion annual value potential concentrates in just four functions: sales and marketing at approximately $1.2 trillion, software engineering at approximately $1.2 trillion, product R&D at roughly $0.4 trillion, and customer operations at roughly $0.4 trillion. But the functions that capture the most aggregate value are not necessarily the functions that deliver the highest return on investment for any individual company. The distinction matters because it determines where a specific organization should direct its first AI investment.

Capgemini's 2025 AI in Business Operations survey provides the clearest function by function ROI breakdown available. The average ROI across all AI investments in business operations is 1.7 times, but it varies significantly by function. Customer operations leads at 2.1 times ROI. People operations, covering HR functions, achieves 1.7 times. Finance and accounting hits 1.5 times. Supply chain and procurement also reaches 1.5 times. The Grant Thornton CFO Survey of 2025 found that 77% of CFOs report at least 2 times ROI from generative AI investments, though BCG's finance specific research found that the median finance team ROI is only 10%, well below the 20% target most organizations set. One third of finance leaders report limited or no gains.

The research reveals a consistent pattern: the highest returns come from operational AI, not analytical AI or customer facing tools. Parabola's State of AI in Operations 2025 report found that 98% of operations teams use or experiment with AI, but only 10% have scaled AI as a core function. Data entry and cleaning delivers the highest ROI, cited by 41% of operations teams. Data centric work represents 87% of priority use cases. The "boring" work of processing, cleaning, and routing information is where the money is.

Specific use cases make the abstract concrete. In invoice processing, organizations report 250 to 300% ROI in the first year with payback in 60 to 90 days. Per invoice cost drops from $12 to $16 manually to $2 to $4 with AI. A Fortune 500 retailer cut cost per invoice from 15 euros to 3 euros, an 80% reduction, and captured 2.3 million euros in early payment discounts annually. Straight through processing rates reach 85 to 92% compared to 20 to 30% for manual processing.

In demand forecasting, accuracy improves from 55 to 65% to 82 to 91% at the SKU level. DP World reports up to 50% reduction in forecasting errors. Safety stock levels drop 20 to 35% without increasing stockouts. Transportation costs drop 12 to 22%. Manufacturers report 150 to 250% ROI within 18 months.

In fraud detection, the financial sector saved $120 billion in 2025 according to TechBullion. Visa prevented nearly $40 billion in fraudulent transactions globally using AI. JP Morgan saves $2 billion annually across fraud detection, coding, and operations automation. DBS Bank generated $1 billion in value from AI across 370 use cases. The U.S. Treasury prevented $4 billion in fraud and recovered $1 billion in check fraud.

In legal document review, Forrester and LexisNexis found in house legal teams achieved 284% ROI. Hartwell and Associates reported $520,000 in annual savings with 70% document processing time reduction and three times faster contract review. The CLOC industry survey found 63% average time savings across contract review processes. Compliance checking automation rates reach 85 to 95%.

In recruiting and screening, AI tools deliver 340% ROI within 18 months. Screening cost drops from $2.92 to approximately $1.00 per candidate, a 96% reduction. Time to hire decreases by 50 to 75%. In manufacturing quality inspection, AI achieves 99.2 to 99.5% defect detection accuracy compared to roughly 80% for human inspectors over a full shift. One precision manufacturer achieved 450% ROI in five months on a $52,000 investment, recovering $380,000 per month in costs.

The research draws a clear line between visible AI and invisible AI. Visible AI includes chatbots, copilots, and productivity assistants. These deliver 5 to 10% individual productivity improvements. Invisible AI includes back office automation, agentic workflows, and process level intelligence. These deliver 20 to 50% efficiency improvements by eliminating entire workflows rather than assisting individual tasks. BCG's research found that leaders generate 62% of their AI value from core business processes, not from front facing tools. Agentic AI already accounts for 17% of total AI value in 2025, projected to reach 29% by 2028. By November 2025, 80% of Fortune 500 companies were building active AI agents.

Industry specific patterns reinforce the operational focus. In healthcare, NVIDIA's 2025 to 2026 survey found 70% of healthcare organizations actively use AI, with 85% reporting increased revenue and 80% reporting reduced costs. In financial services, 90% of institutions now use generative AI for fraud detection. In manufacturing, AI powered quality inspection delivers 37% defect reduction across the industry. In retail, AI demand forecasting reduces stockouts by 65% and improves delivery reliability by 40%.

The companies that fail to capture AI value share a common trait: they invest where AI is visible rather than where AI is valuable. BCG's "Widening AI Value Gap" report from October 2025 found that most companies spread investment across 6.1 use cases with moderate impact. Leaders focus on 3.5 use cases and anticipate 2.1 times greater ROI. Leaders allocate over 80% of AI investments to reshaping key functions. Laggards allocate the majority to smaller productivity initiatives. 60% of companies fail to define and monitor financial KPIs for AI value creation. They literally cannot measure where value goes.

The process maturity of the organization determines whether AI creates value or creates cost. SpryFox's July 2025 industry survey found that fewer than 8% of AI failures are due to inadequate technology. 54.8% trace to stakeholder management issues, including lack of executive sponsorship and unrealistic expectations. 43.5% trace to poor data fundamentals, including inadequate data infrastructure and governance. Accenture's findings confirm that only 16% of organizations have achieved the highest level of AI ready operations, but those that have are 3.3 times more likely to successfully scale high value use cases, achieving 2.5 times higher revenue growth and 2.4 times greater productivity improvements.

McKinsey's 2025 data delivers the key finding: the single factor with the strongest correlation to EBIT impact from generative AI is whether the organization has fundamentally redesigned workflows around AI. Not which model they use. Not how much they spend. Whether they changed how work actually gets done. The companies layering AI onto broken processes will get marginally faster broken processes. The companies redesigning processes around AI capabilities will compound their advantage over every competitor who did not.