WHERE AI ACTUALLY PAYS OFF

A Sector-by-Sector, Department-by-Department Analysis of AI Cost and Return

Ask a chief financial officer whether AI is paying off, and the honest answer is: it depends entirely on where you look. A single company-wide figure the median 10 per cent return documented across finance functions, or the widely quoted $3.70 return per dollar invested obscures a far more useful pattern hiding underneath it. When investment and return data are broken apart by sector and by department, a consistent structure emerges: money spent on high-volume, operational workflows tends to show a measurable benefit sooner than money spent on judgment-heavy, strategic work.

Sector-by-Sector: Spend Intensity and Return

At the sector level, spending intensity in 2026 and the estimated value of that spending come from two different research programmes, so read them side by side rather than as a single number. BCG’s AI Radar tracks what companies say they plan to spend this year; McKinsey Global Institute’s 2023 economic-potential analysis estimates the productivity value available if generative AI were fully deployed across each industry – a ceiling, not a guarantee, and not all of it has been realised yet.

Figure 1: Planned 2026 AI spend as a share of revenue, by sector . Source: BCG AI Radar 2026

Claimed vs. Measured Return

Sitting alongside the potential-value estimates above is a separate, frequently repeated claim: that generative AI returns $3.70 for every dollar invested, rising to $10.30 for top adopters. That figure comes from a real 2024 IDC study, but the study was commissioned and sponsored by Microsoft, a company selling AI products, which is a reasonable basis for treating it as optimistic rather than neutral. Independent survey data collected specifically from finance functions, by contrast, puts the realistic median closer to a 10 percent return, or roughly $1.10 for every dollar spent.

Figure 2: Vendor-reported return multiples (IDC/Microsoft, 2024) versus independently surveyed finance-function returns (BCG, 2025–2026).

Department-by-Department: Where the Independent Evidence Is Strongest

The most defensible department-level evidence comes directly from McKinsey’s Global Survey on the State of AI, which asks respondents which specific business functions are producing measurable financial impact, rather than asking about AI in the abstract. Across the 2025 survey (1,993 respondents) and the 2026 update, the same pattern holds: cost reductions cluster in operational functions, while revenue gains cluster in customer-facing and development functions.

Figure 3: Functions where cost and revenue impact from AI are most frequently reported. Source: McKinsey Global Survey on the State of AI, 2025–2026.

One of the few department-level findings backed by a controlled, peer-reviewed study comes from software engineering. In a Microsoft Research experiment, 95 professional developers were randomly assigned to complete an identical coding task with or without GitHub Copilot. Developers using Copilot finished 55.8 per cent faster than the control group – a statistically significant result (p = 0.0017) – making it one of the best-evidenced single AI-productivity claims available.

Reading Vendor-Reported Figures With Caution

Beyond this independently gathered data, a large volume of department-specific ROI figures circulates in industry commentary: claims of roughly $3.50 returned per dollar spent on AI customer service, 15–25 per cent improvements in sales win rate, or several hours saved per lawyer per week. These numbers are worth noting as directional signals, but each originates from a company selling the tool being measured, with an obvious commercial interest in a favourable outcome. They have not been independently replicated in the way the McKinsey survey data or the Microsoft Research study have, and this article treats them as illustrative claims rather than settled findings.

Why the Pattern Repeats

Three structural factors help explain why operational, high-volume functions appear more quickly and consistently in cost-and-revenue-impact data than strategic or judgment-heavy ones:

• Volume compresses the time required to detect a signal: Supply chain, service operations, and manufacturing generate thousands of similar transactions every week, so an AI system’s impact on cost or throughput becomes statistically visible within a single reporting cycle.
• Judgment-heavy work resists quick attribution: Strategy, corporate finance, legal, and HR involve professional judgment, sign-off chains, and regulatory exposure. AI augments the person doing the work rather than replacing a discrete transaction, so value appears as freed-up hours and better decisions, real but harder to attribute to a line on the P&L within a single year.
• Integration depth varies sharply by function: Functions such as marketing and sales more often adopt AI through off-the-shelf, per-seat tools, whereas software engineering, supply chain, and finance increasingly require integration with core systems of record – a slower and more expensive path to measurable value.

What This Means for Sequencing Investment

None of this argues for deprioritising finance, legal, or HR – McKinsey’s own data shows that strategy and corporate finance are among the functions most often producing reported revenue gains. The more useful conclusion concerns sequencing and expectation-setting. Programmes in supply chain, service operations, manufacturing, marketing, and software engineering have the strongest independently documented evidence of financial impact and are reasonable places to demonstrate value first. Programmes in legal, HR strategy, and other judgment-heavy back-office work deserve continued investment, but boards should be told upfront that the evidence trail for measurable return is thinner and the timeline is longer, rather than discovering that gap after a year of unmet expectations.

References

• BCG: “As AI Investments Surge, CEOs Take the Lead on Decision Making and Upskilling Themselves.” bcg.com, January 2026.
• Center for CFO Excellence (BCG): “How to Get ROI from AI in the Finance Function.” bcg.com, 2025–2026.
• McKinsey Global Institute: “The Economic Potential of Generative AI: The Next Productivity Frontier.” mckinsey.com, June 14, 2023.
• McKinsey & Company.:“The State of AI in 2025: Agents, Innovation, and Transformation” and “The State of AI: Global Survey 2026.” mckinsey.com.
• IDC (sponsored by Microsoft): “2024 Business Opportunity of AI: Generative AI Delivering New Business Value and Increasing ROI.” 2024.
• Peng, S., Kalliamvakou, E., Cihon, P., and Demirer, M. “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot.” Microsoft Research, 2023.

Disclaimer: This article draws on publicly available data from named research organizations (BCG, McKinsey, Microsoft Research) and vendor-published sources as of September 2026. Vendor-reported figures reflect company claims and haven’t been independently verified; all figures may be revised as new studies emerge. This is not investment, legal, or financial advice  consult original sources for full methodology before relying on any single statistic.