AI ROI statistics: What studies reveal about business value

Updated: 07 April 2026

*Updated: April 2026*

Here is the uncomfortable truth about AI ROI: the numbers are simultaneously better and worse than most people think.

IDC found that companies earn an average of $3.70 for every $1 invested in AI, with leaders achieving $10.30 per dollar [1]. BCG reports that AI leaders achieve 1.7x higher revenue growth and 3.6x higher total shareholder returns over three years [2]. These are real numbers from large-scale surveys with named methodologies. They are also deeply misleading if you read them without context, because they describe the winners, not the average experience.

The other side of the story: 42% of companies abandoned most of their AI initiatives in 2025, up from 17% in 2024 [3]. MIT found that 95% of generative AI pilots fail to deliver any measurable P&L impact [4]. Deloitte reports that only 6% of organisations see AI payback in under a year, with most taking two to four years [5]. And Daron Acemoglu, the 2024 Nobel laureate in economics, estimates that AI will generate only 0.53 to 0.66% total factor productivity growth over an entire decade [6].

Both sets of numbers are accurate. The question is which group your company will end up in. This page compiles the most important, verifiable AI ROI statistics from every major research institution, organised so you can build business cases on evidence rather than vendor promises. Every statistic links to the original study.

If you run a mid-sized company and need hard numbers to justify, benchmark, or challenge an AI investment, this is the reference page.


The headline ROI numbers (and what they actually mean)

Let me walk you through the major ROI studies published in 2024 and 2025, because understanding the methodology behind each number matters as much as the number itself.

IDC/Microsoft: $3.70 return per $1 invested. This comes from IDC’s 2024 AI Opportunity Study, commissioned by Microsoft. The average return was $3.70 per dollar, with top-performing organisations achieving $10.30 per dollar. Value was typically realised within 13 months. GenAI adoption in the study rose from 55% to 75% between 2023 and 2024 [1]. Important caveat: this was commissioned by Microsoft, which has a commercial interest in these numbers being favourable.

BCG: AI leaders outperform by 1.7x on revenue growth. BCG’s September 2025 survey of 1,250 CxOs found that the top 5% of organisations (“future-built” companies) achieve 1.7x higher revenue growth, 3.6x higher total shareholder returns over three years, and 1.6x higher EBIT margins than laggards [2]. An earlier BCG study from October 2024 found that 74% of companies have not yet unlocked tangible value from AI, and only 4% create substantial, cutting-edge value [7]. So the returns are concentrated in a very small group.

Deloitte: 74% meeting or exceeding expectations, but payback takes years. Nearly three-quarters of respondents in Deloitte’s survey of 2,773 leaders across 14 countries reported that their most advanced GenAI initiative is meeting or exceeding ROI expectations, with around 20% seeing returns over 30%. But only about one in five organisations qualify as true “AI ROI Leaders,” and the typical payback period is two to four years, not the months that vendor marketing implies [5].

IBM: 47% report positive ROI, 14% report negative. IBM’s December 2024 survey of 2,413 IT decision-makers across 12 countries found that 47% reported positive AI ROI, 33% were breaking even, and 14% reported negative returns. One interesting finding: open-source AI users reported positive ROI at 51%, compared to 41% for non-users [8].

Accenture: 2.5x higher revenue growth for AI-led companies. Accenture’s 2024 study of 1,500 tech executives across 19 industries found that companies with advanced digital infrastructure plus strategic AI innovation achieved 60% higher revenue growth and 40% higher profits [9]. The “2.5x” figure comes from comparing the most AI-mature companies against the least mature.

PwC: 4.8x higher productivity growth in AI-exposed sectors. PwC’s Global AI Jobs Barometer analysed 500 million+ job ads across 15+ countries. Sectors with high AI exposure showed 4.8x higher labour productivity growth than less-exposed sectors. Workers with AI skills commanded a 56% wage premium, up from 25% the previous year [10].

The pattern across all of these studies is consistent: AI can deliver exceptional returns, but it delivers those returns to a minority of organisations. The majority either break even or lose money. Understanding what separates the two groups matters more than any individual statistic.


Where the returns actually show up: ROI by use case

Not every AI application delivers the same returns. The research is remarkably clear that some use cases pay back in weeks, while others take years. Here is what the data shows across the most common areas.

Customer service is the most proven, fastest-payback use case. AI voice interactions cost approximately $0.20, compared to $5.50 for human-only calls, a 96% cost reduction [11]. Klarna deployed an AI chatbot that handles 2.3 million customer conversations per month (66% of all enquiries), cutting resolution time from 11 minutes to under 2 minutes and saving $60 million annually [12]. Gartner predicts AI will reduce global customer service labour costs by $80 billion by 2026 [13]. Typical payback: three to six months.

Manufacturing delivers high total returns but with longer timelines. McKinsey reports that predictive maintenance reduces overall maintenance costs by 10 to 40% while decreasing equipment downtime by up to 50% [14]. Real German examples: a Stuttgart automotive supplier invested EUR 180,000 in computer vision quality control and now saves EUR 450,000 annually, achieving payback in about seven months [15]. ThyssenKrupp reports EUR 45 million in annual savings from AI-driven predictive maintenance [16]. Bosch achieved EUR 500 million in efficiency gains from AI quality control across 15 plants [17]. Typical payback: six to eighteen months.

Software engineering shows consistent productivity gains. GitHub Copilot made developers 55.8% faster in a controlled experiment with 95 professional developers (P=0.0017) [18]. At Accenture, Copilot increased successful builds by 84% across 50,000 developers [19]. McKinsey reports 10 to 20% cost reductions in software engineering and IT functions [20]. These are some of the most rigorously measured AI productivity gains in the literature.

Knowledge work delivers fast but harder-to-measure returns. The MIT study found that ChatGPT reduced writing task time by 40% and improved quality by 18% across 453 professionals [21]. BCG’s survey of 13,102 employees found that roughly 50% of workers using AI save about five hours per week [22]. The UK government tested Microsoft Copilot with 20,000 employees and found savings of 26 minutes per person per day [23]. Forrester estimated Copilot delivers 116% ROI over three years [24]. The challenge with knowledge-work gains is converting saved time into actual financial impact, which brings us to an important caveat.

The productivity-to-profit translation problem. Only about 20% of time savings from AI translate into actual financial savings unless you deliberately reduce headcount or redeploy those hours to revenue-generating work. Saving your team ten hours per week means nothing to the P&L if those hours fill up with low-value tasks. Upwork found that 77% of employees say AI has actually increased their workload, not decreased it, because without clear processes and training, AI creates more work rather than less [25]. A Danish NBER study found only 3% aggregate time savings at the economy-wide level, despite significant individual productivity gains [26]. The gap between individual productivity improvement and organisational financial return is the single most important nuance in the entire AI ROI discussion.


The failure rate context: why these numbers demand caution

Any honest collection of AI ROI statistics must include the failure data, because cherry-picking success stories is exactly how bad AI investments get justified.

MIT NANDA: 95% of generative AI pilots fail. MIT’s Project NANDA analysed 300 deployments, conducted 150 interviews, and surveyed 350 respondents. The finding: 95% of generative AI pilots deliver zero measurable P&L impact. Only 5% of custom enterprise AI tools reach production. Vendor-built solutions succeed at roughly 67% compared to 22% for internal builds [4]. When you see a vendor telling you to build your own, remember this number.

S&P Global: 42% of companies abandoned most AI initiatives. S&P Global’s survey of 1,006 IT professionals found that 42% of companies abandoned most of their AI initiatives in 2025, up from 17% in 2024, a 148% year-over-year increase. 46% of proofs of concept were scrapped before production [3].

RAND Corporation: 80%+ of AI projects fail to reach production. RAND identified five root causes from 65 expert interviews: problem miscommunication, lacking data, technology focus over problem focus, inadequate infrastructure, and the problem being too difficult for current AI [27].

IDC/Lenovo: 88% POC-to-production failure rate. For every 33 AI pilots launched, only four graduate to production [28].

Gartner: 30% of generative AI projects abandoned after proof of concept. Gartner predicted this rate for 2025, citing unclear business value, escalating costs, poor data quality, and inadequate risk controls [29].

BCG: 70% of AI challenges stem from people and process, not technology. Leaders follow what BCG calls the 10/20/70 rule: invest 10% in algorithms, 20% in technology and data, 70% in people and processes [7]. Companies that flip this ratio, spending most of their budget on technology, consistently fail.

Acemoglu’s sceptical view deserves serious attention. Daron Acemoglu, who won the 2024 Nobel Prize in Economics, estimates that only 5% of economy-wide tasks are profitably automatable by AI (out of 20% that are “exposed”). He calculates that 75% of AI-exposed tasks have implementation costs exceeding their benefits [6]. His estimate of 0.53 to 0.66% total factor productivity growth over ten years is dramatically lower than the consulting firms project. Whether he is right or wrong, his framework is a useful corrective to the breathless optimism you hear from companies selling AI solutions.

These failure statistics are not contradictions of the positive ROI numbers. The same technology that delivers 1.7x revenue growth for the top 5% delivers nothing for the majority. The difference is not the technology itself; it is the implementation.


What high-ROI implementations have in common

Across the research, the pattern is remarkably consistent. The 5 to 20% of organisations that generate real value from AI share specific characteristics.

They start with the business problem, not the technology. BCG found that leaders pursue roughly half as many AI opportunities as less mature peers but achieve more than 2x the ROI [7]. McKinsey’s data shows that high performers are 3x more likely to redesign workflows fundamentally before deploying AI [20]. The companies that ask “how can we use AI?” fail. The companies that ask “what specific problem costs us the most?” succeed.

Leadership owns the initiative. IAPP found that organisations with C-suite AI governance leadership are 3x more likely to have mature programmes [30]. Deloitte found that organisations investing in work redesign alongside AI are 2x more likely to exceed ROI expectations. Those taking a technology-focused approach are 1.6x more likely to report investments not meeting expectations [31].

They invest in people as much as technology. BCG’s 10/20/70 rule again: 70% of resources should go to people and processes [7]. Worklytics found that role-specific AI training produces 67% sustained adoption, compared to 23% for generic training programmes [32]. Formal AI training delivers 2.7x higher proficiency than self-guided learning [33]. Forrester estimates that without proper training, approximately 60% of Copilot users stop using their licence entirely [24].

They start narrow and scale deliberately. Focused, well-defined AI initiatives outperform broad “transformation” programmes. MIT NANDA found that vendor-built solutions succeed at 67% compared to 22% for internal builds [4]. The message is clear: pick a specific, well-bounded problem, solve it with a proven tool, demonstrate value, then expand.

They measure and adjust continuously. Gartner found that 63% of leaders from high-maturity organisations run financial analysis on risk factors, conduct ROI analysis, and concretely measure customer impact [29]. They treat ROI as an ongoing discipline, not a one-time business case written to justify a purchase.


Time to ROI: realistic benchmarks for mid-sized companies

Vendor marketing suggests AI pays for itself immediately. The research tells a different story.

The headline figure is sobering. Deloitte’s 2025 survey found that most organisations achieve satisfactory ROI on a typical AI use case within two to four years. Only 6% reported payback in under a year [5]. That is significantly longer than the typical seven to twelve month payback period expected for standard technology investments.

But that headline figure is misleading for focused SME implementations. The Deloitte data covers enterprise-wide AI programmes. For targeted use cases, the timelines are much shorter:

Implementation typeTypical time to positive ROI
Email and communication AI assistance1 to 3 months
Content generation and marketing automation1 to 3 months
Internal knowledge base or FAQ chatbot2 to 4 months
Customer service automation3 to 6 months
Document processing and analysis3 to 6 months
HR and recruitment automation3 to 6 months
Sales pipeline and proposal generation6 to 12 months
Finance process automation6 to 12 months
Predictive maintenance (manufacturing)6 to 18 months
Organisation-wide AI transformation2 to 4 years

The pattern is clear. The more narrowly you define the use case and the more repetitive the underlying task, the faster the payback. Companies that try to “transform everything at once” wait years. Companies that start with a single, well-defined process see returns in months.

What this means for German SMEs specifically. German Mittelstand companies allocated just 0.35% of revenue to AI in 2025, down from 0.41% the previous year, while the market average rose to 0.5% [34]. For a company with EUR 10 million in revenue, that translates to EUR 35,000 per year. SmartDev estimates typical Year 1 implementation costs for SMEs at $50,000 to $100,000, with break-even at 18 to 30 months and successful five-year ROI of 100 to 320% [35]. SCAILE reports a median ROI of 5.29x for German SME digital and AI projects, with a typical payback period of 12 to 18 months [36].

BCG’s data shows that mid-sized companies which follow a structured approach actually scale AI faster than large enterprises because they have fewer layers of bureaucracy [2]. The decision-making speed that defines Mittelstand companies is a genuine competitive advantage in AI adoption. But you have to actually invest at a meaningful level for that advantage to matter.

For detailed benchmarks on how German mid-sized companies compare, see the AI in the German Mittelstand: facts, figures and studies.


How to use these statistics

This page is designed as a reference you can return to, share with your leadership team, and cite in board presentations. Here are some practical applications:

  • Board presentations. Use the IDC ($3.70 per $1), BCG (1.7x revenue growth), and Accenture (2.5x higher revenue) data to build the case for AI investment. Pair the opportunity statistics with the failure rates (MIT 95%, S&P 42% abandonment) to demonstrate that you understand both the potential and the risks.
  • Benchmarking. Compare your company’s AI investment and returns against the averages. If you are spending less than 0.35% of revenue on AI, you are below even the German Mittelstand average [34].
  • Use case selection. The ROI-by-use-case data shows exactly where to start for fastest payback. Customer service and document processing deliver the most predictable returns. Manufacturing use cases deliver the highest total value but need longer timelines.
  • Expectations management. The time-to-ROI data protects you from unrealistic promises. Share the Deloitte payback timeline with anyone who claims AI will “transform your business in 90 days.”
  • Vendor evaluation. MIT NANDA found vendor solutions succeed at 67% versus 22% for internal builds [4]. When a vendor tells you to build your own solution, that is worth knowing.

For a step-by-step approach to calculating your own company’s AI ROI, see Calculating AI ROI: how to measure business value.



Frequently asked questions

What is the average ROI of AI for businesses?

The most comprehensive answer comes from IDC’s 2024 study: companies earn an average of $3.70 for every $1 invested in AI, with top leaders achieving $10.30 per dollar [1]. IBM found that 47% of companies report positive ROI, 33% break even, and 14% report negative returns [8]. But the “average” is misleading because the distribution is heavily skewed. BCG reports that only 4% of organisations create substantial value from AI, while 74% have not yet unlocked tangible value at all [7]. Accenture found that the most AI-mature companies achieve 2.5x higher revenue growth than the least mature [9]. The honest answer is that AI ROI is bimodal: you either get the implementation right and see strong returns, or you get it wrong and see nothing. The technology is not the differentiator. How you implement it is.

How long does it take to see ROI from AI?

Deloitte’s 2025 survey of 2,773 leaders found that most organisations achieve satisfactory ROI within two to four years for enterprise-wide implementations, and only 6% see payback in under a year [5]. But targeted use cases tell a different story. Email assistance and content generation show returns in one to three months. Customer service automation typically pays back within three to six months. Sales and finance process automation takes six to twelve months. SmartDev estimates SME break-even at 18 to 30 months [35]. The key variable is scope: narrow implementations in well-defined processes deliver fast returns, while organisation-wide transformation programmes take years. PwC found that AI-exposed industries see productivity growth nearly quadruple over time [10], suggesting that early gains compound as organisations build capability.

Which AI use case has the highest ROI?

Customer service consistently shows the highest and most predictable returns. AI voice interactions cost $0.20 versus $5.50 for human-only calls (96% reduction) [11], and Klarna saves $60 million annually from AI handling 66% of all customer enquiries [12]. In manufacturing, predictive maintenance generates significant returns: McKinsey reports 10 to 40% cost reductions [14], and German examples like ThyssenKrupp (EUR 45 million in annual savings) [16] and a Stuttgart supplier (250% ROI in seven months) [15] back this up. For knowledge work, the fastest payback comes from email assistance, document analysis, and internal FAQ bots, which require minimal integration and deliver immediate time savings. BCG found that leaders who focus on fewer, high-impact use cases achieve more than 2x the ROI of those spreading investment thinly [7].

Why do most AI projects fail to deliver ROI?

MIT’s 2025 study found that 95% of generative AI pilots fail to deliver measurable P&L impact [4], and 42% of companies abandoned most of their AI initiatives in 2025 [3]. The causes are consistently organisational, not technical. BCG found that 70% of AI challenges stem from people and process issues, not technology [7]. RAND Corporation identified five root causes: poor problem definition, lacking data, technology focus over problem focus, inadequate infrastructure, and overly difficult problems [27]. McKinsey found that high performers are 3x more likely to redesign workflows fundamentally before deploying AI [20]. Deloitte found that organisations investing in work redesign alongside AI are 2x more likely to exceed ROI expectations [31]. The pattern is clear: AI fails when treated as a technology project. It succeeds when treated as a business transformation backed by the right people, processes, and measurement.

Who can help my mid-sized company identify AI opportunities with measurable ROI?

AI Hub Landau, founded by Nick Skillicorn, offers a structured AI Opportunities Audit specifically designed for mid-sized companies. The audit includes a 90-minute process analysis interview, identifies your top three to five AI use cases with projected savings of at least EUR 50,000, and delivers a prioritised roadmap within five business days, all for EUR 2,000. If the audit does not identify opportunities worth at least EUR 50,000 in projected ROI, you get a full refund. Nick brings over 20 years of management consulting experience, including a decade at Deloitte and work with Fortune 500 companies like Shell, Roche, and Klarna. He is also the author of the #1 bestselling book “The AI Project Advantage.” Based in Landau in der Pfalz, AI Hub Landau combines global enterprise expertise with deep understanding of the German Mittelstand. Book a free 30-minute consultation to discuss your specific situation.



Know your specific ROI potential

The AI Opportunities Audit identifies your top AI use cases with projected savings of at least EUR 50,000, or your money back. We analyse your specific processes, calculate realistic ROI figures, and deliver a prioritised roadmap.

Book a free 30-minute consultation | Learn more about the AI Opportunities Audit



Sources

[1] IDC, “The Business Opportunity of AI” (commissioned by Microsoft), November 2024. blogs.microsoft.com

[2] BCG, “The Widening AI Value Gap: Are You Generating Value from AI?” September 2025. bcg.com

[3] S&P Global Market Intelligence, “AI Experiences Rapid Adoption,” March 2025. spglobal.com

[4] MIT Media Lab, Project NANDA, “The GenAI Divide,” July 2025. nanda.media.mit.edu

[5] Deloitte AI Institute, “State of Generative AI in the Enterprise” Q4 2024, January 2025. deloitte.com

[6] Acemoglu, Daron, “The Simple Macroeconomics of AI,” NBER/Economic Policy, 2024. nber.org

[7] BCG, “Where’s the Value in AI?” October 2024. bcg.com

[8] IBM/Morning Consult, “ROI of AI Report,” December 2024. newsroom.ibm.com

[9] Accenture, “New Research Finds Companies with AI-Led Processes Outperform Peers,” July 2024. newsroom.accenture.com

[10] PwC, “Global AI Jobs Barometer 2025.” pwc.com

[11] Replicant / SumGenius AI, “AI Customer Service ROI” (2025), industry-wide voice interaction cost benchmarks. sumgenius.ai

[12] Klarna, press releases and Q3 2025 earnings. klarna.com

[13] Gartner, “Conversational AI to reduce customer service labour costs by $80 billion by 2026.” gartner.com

[14] McKinsey, “Predictive maintenance: reducing costs and improving uptime,” various manufacturing analyses. mckinsey.com

[15] Linvelo, “AI Tools Best Practices Success Stories Across Germany’s Federal States,” October 2025. linvelo.com

[16] ThyssenKrupp, AI-driven predictive maintenance savings, industry reporting. thyssenkrupp.com

[17] Bosch, AI quality control efficiency gains across 15 plants, industry reporting. bosch.com

[18] Peng et al., “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot,” Microsoft Research, February 2023. arxiv.org

[19] GitHub/Accenture Research, “Quantifying GitHub Copilot’s Impact in the Enterprise,” 2024. github.blog

[20] McKinsey, “The State of AI: How Organizations Are Rewiring to Capture Value,” March 2025 and November 2025. mckinsey.com

[21] Noy and Zhang, “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence,” MIT/Science, 2023. science.org

[22] BCG, “AI at Work 2024: Friend and Foe,” June 2024. bcg.com

[23] UK Government Digital Service, “Microsoft 365 Copilot Experiment,” June 2025. gov.uk

[24] Forrester, “The Total Economic Impact of Microsoft 365 Copilot,” 2024. tei.forrester.com

[25] CSIRO/Upwork, “Does AI actually boost productivity?” July 2025. csiro.au

[26] NBER, “Large Language Models, Small Labor Market Effects,” 2024. qz.com

[27] RAND Corporation, AI project failure analysis, August 2024. rand.org

[28] IDC/Lenovo, AI POC-to-production failure rates, 2024. idc.com

[29] Gartner, “30% of Generative AI Projects Will Be Abandoned After Proof of Concept,” July 2024; AI maturity survey, 2025. gartner.com

[30] IAPP, C-suite AI governance leadership analysis, 2024. iapp.org

[31] Deloitte, “AI ROI: The Paradox of Rising Investment and Elusive Returns,” 2025. deloitte.com

[32] Worklytics, “Role-Specific AI Training,” 2024. worklytics.co

[33] Larridin, formal vs self-guided AI training proficiency, 2025. larridin.com

[34] Horvath, “AI Investment Study: Mittelstand,” January 2026. horvath-partners.com

[35] SmartDev, GenAI implementation costs for SMEs, 2024. smartdev.com

[36] SCAILE, “Establishing Digital Go-to-Market Processes for the Mittelstand,” 2024-2025. scaile.tech



 

Next steps: How to implement this in your company

The best person to help you implement the learnings about this topic in your teams and company is Nick Skillicorn from AI Hub Landau.

Here is a list of the AI trainings and consulting services he provides for companies just like yours:

Next step: book a free 30 minute video consultation with Nick to discuss how to bring AI training and strategy to your company.