AI ROI by Industry: What McKinsey, BCG, and Real Deployments Show (September 2026)
ROI data for AI deployments is scarce, contested, and often vendor-sourced. The best available data comes from McKinsey's State of AI 2025 and Build for the Future research, BCG's AI adoption surveys, and public case studies from enterprise vendors. This page compiles the honest picture: where ROI is documented, where it is projected, and where the evidence is thin.
AI ROI varies widely by industry, and most published figures are vendor-sourced, so treat any single multiple with caution. McKinsey's economic-potential analysis puts the largest generative-AI value potential in financial services (2.8-4.7% of revenue), pharmaceuticals and medical products (2.6-4.5%), and retail and consumer packaged goods ($400-660 billion a year); software engineering shows a 20-30% productivity improvement. At the deployment level the best-documented signals are customer service ($150-500K saved per enterprise at 40-60% deflection), IT service management ($1M+ labour savings at Fortune 500 scale), and legal (4-6 hours saved per attorney per week). The honest caveat: only around a third of organisations have begun to scale AI across the enterprise, and most deployers are still early on the returns curve.
AI ROI by Industry at a Glance
| Industry | ROI signal | Source |
|---|---|---|
| Financial services | 2.8-4.7% of revenue (potential value) | McKinsey, economic potential of generative AI (2023) |
| Pharma & medical products | 2.6-4.5% of revenue (potential value) | McKinsey, economic potential of generative AI (2023) |
| Retail & consumer packaged goods | $400-660B annual potential value | McKinsey, economic potential of generative AI (2023) |
| Software engineering | 20-30% productivity improvement; GitHub Copilot 55% faster task completion | Multiple studies; Microsoft GitHub Copilot research |
| Customer service | $150-500K annual savings per enterprise at 40-60% deflection | Intercom Fin case studies |
| IT service management | $1M+ annual IT labour savings (20-40 FTE equivalent) at Fortune 500 scale | Moveworks deflection benchmarks |
| Legal | 4-6 hours saved per attorney per week ($600-1,500 in value) | Spellbook, CoCounsel |
| Sales | 15-25% improvement in win rates in documented cases | Gong revenue intelligence |
Macro bands are McKinsey potential-value estimates; per-vertical figures are documented deployment signals. Vendor-sourced figures vary widely between studies and are rarely independently audited.
The Macro Picture
McKinsey's economic potential of generative AI analysis (June 2023) estimates $2.6-$4.4 trillion in annual value potential across industries. Financial services leads: McKinsey estimates generative AI could add value equivalent to 2.8-4.7% of banking and financial-services revenue. Pharmaceuticals and medical products: 2.6-4.5% of revenue. Retail and consumer packaged goods: $400-$660 billion a year from personalisation, supply chain, and customer operations. Technology sector (software): 20-30% productivity improvement in software engineering tasks per multiple studies. These are potential-value ceilings; realised value in 2026 is a fraction of potential.
Vertical AI vs Horizontal AI ROI
Vertical AI deployments are widely reported to out-return horizontal deployments on equivalent domain tasks, but the published multiples are vendor-sourced, range over an order of magnitude between studies, and are rarely independently audited, so no single headline figure should be treated as definitive. The directional mechanism is well established: vertical AI requires less human correction, integrates with domain workflows, and surfaces domain-specific insights that horizontal models miss. The cost of vertical AI (higher ACV, longer integration) is offset by the accuracy premium in high-stakes domains. For low-stakes, broad tasks, horizontal ROI is lower but so is the cost, making the per-dollar ROI comparable.
Per-Vertical ROI Data
Customer Service: Intercom Fin case studies show $150-$500K annual savings per enterprise deployment from deflection rates of 40-60%. ITSM: Moveworks deflection benchmarks show $1M+ annual IT labour savings at Fortune 500 scale (20-40 FTE equivalent). Legal: Spellbook and CoCounsel cite 4-6 hours saved per attorney per week on research tasks; at $300-$500/hour billing rates, this is $600-$1,500 in value per attorney per week. Sales: Gong's revenue intelligence platform reports 15-25% improvement in win rates in documented cases; at enterprise ACV, this is multi-million dollar annual impact. Engineering: in GitHub's controlled experiment, developers completed a coding task about 55% faster with Copilot (1h11m versus 2h41m for the control group); this is a single-task speed result rather than an all-round productivity multiple.
What ROI Are Companies Actually Getting?
Potential-value bands are ceilings, not realised returns, and the gap between the two is the central ROI story of 2026. BCG's Center for CFO Excellence surveyed more than 280 finance executives already working with AI and generative AI in March 2025 and found the median reported ROI was just 10%, well below the roughly 20% many teams target, with about a third reporting limited or no gains so far. This is one of the few named, non-vendor ROI figures in the market, and it points the same direction as McKinsey's scaling data: the technology's ceiling is high, but most deployers are still early on the curve. In BCG's survey, 44% of finance teams had moved past piloting into scaled deployment; the highest-return teams shared a consistent playbook of focusing on value over experimentation, embedding AI into a broader transformation, and scaling in sequence rather than launching many pilots at once.
The Honest Caveats
About a third of organisations have begun to scale AI across the enterprise per McKinsey's 2025 State of AI; nearly two-thirds have not yet begun enterprise-wide scaling, and only around 7% report AI fully scaled. Most pilots do not reach full production ROI. Common failure modes: insufficient data quality for training or retrieval, integration complexity underestimated, change management (getting users to adopt the AI) not resourced, and ROI measurement not defined before deployment. The organisations achieving the highest ROI (McKinsey's AI high performers) are far more likely to have dedicated AI transformation teams, executive sponsorship, and measurement frameworks defined before deployment. As BCG's finance-function survey shows, the median realised ROI so far sits well below the potential ceilings above.
Frequently Asked Questions
What is the ROI of AI by industry?
Per McKinsey's economic potential of generative AI analysis (June 2023): financial services leads with a value potential equivalent to 2.8-4.7% of revenue, pharmaceuticals and medical products 2.6-4.5%, retail and consumer packaged goods $400-$660 billion a year, and software engineering shows 20-30% productivity improvement. These are potential bands (ceilings), not documented returns; realised value in 2026 is a fraction of potential.
Does vertical AI deliver better ROI than horizontal AI?
Vertical AI is widely reported to deliver higher ROI than horizontal AI on equivalent domain tasks, but the published multiples are vendor-sourced and vary widely between studies, so no single figure should be treated as definitive. The reliable part is the mechanism: vertical AI requires less human correction, integrates with domain workflows natively, and surfaces domain-specific insights horizontal models miss.
How many AI pilots actually reach production ROI?
About a third of organisations have begun to scale AI across the enterprise, per McKinsey's 2025 State of AI; nearly two-thirds have not yet begun, and only around 7% report AI fully scaled. Common failure modes are data quality, underestimated integration complexity, and unresourced change management.
What ROI are companies actually getting from AI, versus the potential?
The potential bands (2.8-4.7% of revenue in financial services, hundreds of billions in retail) are ceilings, not realised returns. BCG's Center for CFO Excellence surveyed more than 280 finance executives working with AI and generative AI in March 2025 and found a median realised ROI of just 10%, below the roughly 20% many teams target, with about a third reporting limited or no gains. The gap reflects how early most deployers still are: McKinsey finds only around a third of AI pilots reach production scale.
Explore the Verticals
Other Cross-Vertical Topics
Sister Sites
- buildingeffectiveagents.com →[sister site]
Agent patterns reference: orchestrator, subagents, tool-use patterns
- whatisanaiagent.com →[sister site]
Glossary and definitions for AI agent terminology
Sources
All statistics cited on this page are tagged to source URLs on the sources index. Publication dates included for freshness verification.