The AI-First Company: A Size-Segmented Implementation Guide

In April 2016, Google’s CEO Sundar Pichai said there would “soon be a transition from mobile first to AI first.”¹ Ten years later, many companies call themselves AI-first. Few get much out of AI.

How many companies use AI at all depends on who is asked. In the EU, 20% of enterprises with ten or more employees used any AI technology in 2025.² In the United States, the Census Bureau’s business survey, which is designed to represent all firms, puts the figure at about 18–20%.³ In a survey run with central banks in the United States, the United Kingdom, Germany and Australia, 69% of close to 6,000 executives said their firm used AI. About nine in ten said AI had made no difference to their firm’s productivity or employment over the past three years.⁴ McKinsey’s 2025 global survey found that 88% of organizations used AI in at least one business function. Only about 6% were “AI high performers”: companies that put 5% or more of their EBIT (operating profit) down to AI.⁵

What AI-first means depends on the size of the company. A ten-person firm, a 150-person manufacturer, a 600-person regional bank and a global enterprise have different budgets, different people to draw on and different risks. This guide uses four size bands. The two smaller ones follow the European Union’s definition of small and medium-sized enterprises: micro and small businesses (1–49 employees), medium-sized businesses (50–249), mid-market companies (250–999) and enterprises (1,000+). Every figure links to its source. A figure that comes only from the company itself, or from the vendor that sold the tool, is labelled as such.

---

What AI-first means, and how it differs from using AI

A company that uses AI adds it to work that otherwise stays the same: a chatbot on the website, a tool that drafts emails. An AI-first company changes the work itself, so that AI does part of it and people do the rest. Take two marketing teams. In the first, staff use ChatGPT to draft blog posts, and nothing else changes. In the second, the team has rebuilt how it produces content. AI does the research, the first drafts and the versions for each channel. People choose the topics, edit and check quality.

Three related terms come up often:

• AI-native: a company built with AI as its product, such as OpenAI, Anthropic or Midjourney.
• AI-enabled: a company that adds AI to some tasks without changing how it works, such as a bank’s chatbot or a retailer’s demand forecasting.
• AI-augmented: staff use AI to get more of their own work done, and stay in charge of it.

AI-first asks the most of a company, because it changes processes, roles and rules, not only tools.

Two consultancies offer rules of thumb. BCG describes three levels of ambition: Deploy ready-made tools, for a 10–15% productivity gain; Reshape whole functions around AI, for 30–50%; and Invent new products and services built on AI.⁶ The percentages are BCG’s estimates. BCG’s “10-20-70 rule” says that in an AI programme, 10% of the effort goes into the algorithms, 20% into technology and data, and 70% into changing how people and processes work.⁷ McKinsey’s Rewired research says the companies that succeed employ their own digital specialists instead of relying on outside ones.⁸

A public “AI-first” memo tells you what a chief executive intends, not what happened. In April 2025, Shopify’s CEO told teams that before asking for more staff they had to show why AI could not do the job.⁹ The same month, Duolingo’s CEO announced an AI-first plan that included phasing out contractors for work AI could handle. After a public backlash, the CEO said the memo had not given enough context.¹⁰

---

Micro and small businesses (1–49 employees)

Small firms are the least likely to have started. In the EU, 17% of small enterprises (10–49 employees) used AI in 2025, against 55% of large ones. Eurostat does not count firms with fewer than ten employees.² In the euro area in late 2025, 41% of micro firms used no AI at all, against 13% of large firms. Yet at every size, about 7–8% of firms made significant use of AI.¹¹ In the United States, bank records show that 17.7% of small firms had paid for an AI service by December 2025.¹² When the Federal Reserve asked small firms with employees, 46% said they used AI.¹³

Few small firms employ data or IT specialists. At this size, AI-first means running your main work on ready-made tools, not building your own systems. The advantage is speed. With little old software to replace and only one or two people to agree, a small firm can change a process in weeks.

The strongest evidence supports starting with one clearly defined task. In a field study of more than 5,000 customer-support agents who were given a generative AI assistant, the agents resolved about 15% more issues per hour. The least experienced agents gained the most.¹⁴ Giving staff a tool is not enough on its own. A large study of Danish workers found that AI chatbots saved their users about 3% of their working hours on average. The chatbots made no measurable difference to what those workers earned or how many hours they worked.¹⁵ Taken together, the two studies say that AI pays where a task is frequent and clearly defined and a person checks the result. It pays much less where staff are simply given access.

When nobody checks the advice, the people who most need help can lose. In a randomised trial, 640 small-business owners in Kenya were given a GPT-4 business mentor on WhatsApp. On average it made no difference to revenue or profit. Owners who were struggling at the start did nearly 10% worse with it. Owners who were already doing well may have gained over 15%.¹⁶

Small firms use AI for practical engineering and operations, not only for writing. Hrag Kalebjian, owner of Henry’s House of Coffee in San Francisco, used ChatGPT to write Arduino code for a small device that sprays a fine mist on the shop’s coffee grinders to reduce static.¹⁷ Aragon AI, which sells AI-made headshots, says it reached $10 million in annual recurring revenue in about two years with a team of nine (company-reported).¹⁸ Firms like Aragon are still rare. On Product Hunt, a website where new products are launched, solo founders launched more products after ChatGPT arrived, but teams hold more and more of the top rankings.¹⁹

Most of the tools small firms use fall into four groups. Prices change often, and several vendors that once let anyone sign up online now sell only to larger customers. Check current prices before you set a budget.

Group	Typical tools	What to check
Writing and research	ChatGPT, Claude, Gemini, Microsoft Copilot	Business plans with data-protection terms; whether your data is used to train the model
Automation	Zapier, Make, n8n	Limits on tasks or runs; whether you can run it on your own server (n8n)
Customer service	Chatbots in help-desk tools, ManyChat	Hand-over to a person; records of conversations
Marketing and e-commerce	Canva, Shopify’s built-in AI, Klaviyo	Whether AI features are included or cost extra

A good first project at this size starts with one process that clearly costs time or money, such as answering routine enquiries, drafting quotes or processing invoices. Measure how long it takes now. Use a ready-made tool, with a person checking what it produces. Use it more widely only when the numbers improve.

---

Medium-sized businesses (50–249 employees)

About three in ten medium-sized firms use AI. In the EU, 30.4% of companies with 50–249 employees used AI in 2025.² In the United States, the Census Bureau’s business survey shows similar rates for firms with 100–249 employees.³ Across seven OECD countries, 31% of small and medium-sized firms said they used generative AI.²⁰

A firm of this size can give one person responsibility for AI. Usually that is an operations or IT lead with time and budget set aside for it. The firm rarely has a data-science team. So the practical route is to buy tools, set them up and set the rules for using them, not to build. BCG’s 10-20-70 rule matters most at this size.⁷ Choosing the tools is the easy part. Results depend on changing the process, training the people who use it and measuring what changes.

If you operate in the EU, the law now applies to you too. Under the EU AI Act, the bans on certain uses of AI have applied since February 2025, and the rules for general-purpose AI models since August 2025. The Act first required every company that deploys AI systems (uses them in its own business) to make sure its staff had sufficient AI literacy. In May 2026, the EU institutions agreed an amendment called the “Digital Omnibus”. It postpones most duties for the high-risk systems listed in Annex III to December 2027. It also softens the literacy duty: firms must now support their staff’s AI literacy, not guarantee a particular level.²¹ At a minimum, a medium-sized firm should have a short written AI-use policy, a list of the AI systems it uses and basic training for staff.

---

Mid-market companies (250–999 employees)

Mid-market companies have real budgets and fewer layers of approval than large enterprises. In RSM’s 2026 survey of middle-market firms in the United States and Canada, 86% said they had partly or fully built AI into their operations. Some firms said their pilots had gone only moderately well or worse. The obstacles those firms named most often to rolling AI out more widely were data quality (53%) and connecting AI to their existing systems (47%).²² RSM defines the middle market by revenue, not by headcount, so its firms overlap with this size band but do not match it.

MIT NANDA’s 2025 GenAI Divide report is often quoted about firms of this size. It is a preliminary report, not a measured study. It rests on a review of more than 300 public AI initiatives, 52 interviews and 153 survey responses collected at four industry conferences, and it uses two different definitions of success.²³ It said mid-market firms took about 90 days to go from pilot to full use, against nine months or more at large enterprises. It also said tools bought from specialised vendors succeeded about 67% of the time, far more often than tools companies built themselves. The report itself says those rates “may reflect organizational capabilities rather than implementation approach alone.” See the claim check on the report’s 95% figure.

Whether this size band has an advantage is disputed. In January 2025, researchers linked to BCG and its Henderson Institute argued in Fortune that mid-sized companies may have a “Goldilocks” advantage in AI. They also noted that mid-sized companies were behind: 28% reported mid-to-high maturity in generative AI, against 46% of large companies.²⁴ BCG’s 2026 study of the midmarket found that large companies were more likely to report significant revenue gains from AI. Large companies also spent a larger share of revenue on AI and hired AI specialists faster.²⁵ Deciding quickly is an advantage only for a company that then acts on its decisions.

Companies of this size usually organize AI in three stages. First, a named executive and a small group drawn from several departments set the priorities. Next, each business unit gets its own AI lead, and a central team looks after shared tools, data and rules. Finally, each department uses AI tools made for its own roles as part of normal work, and measures the results regularly.

---

Enterprises (1,000+ employees)

In a large company, AI-first means reorganizing operations, strategy and culture around AI. The gap between using AI and making money from it is widest here. McKinsey’s 2025 survey (see the start of this guide) found that the few high performers were nearly three times as likely as other companies to have rebuilt individual workflows from the ground up.⁵ Companies are also managing more AI risks than before: four on average in 2025, against two in 2022.⁵

Volkswagen has stated its aim plainly. In 2025 it announced plans to invest up to €1 billion in AI by 2030 under the principle “no process without AI”. It said it expected efficiency gains and avoided costs of up to €4 billion by 2035.²⁶ Banks show what this looks like at scale. JPMorgan Chase raised its target for business value from AI to $1.5 billion by the end of 2023, from uses in risk, marketing, customer service and fraud prevention.²⁷ Bank of America’s virtual assistant, Erica, had handled more than 3.2 billion client interactions by March 2026.²⁸ Goldman Sachs’s CEO, David Solomon, said in January 2025 that AI could draft 95% of an IPO prospectus in minutes. That work once took a team of six people two weeks.²⁹

UnitedHealth Group, the US health insurer, reported about 1,000 uses of AI in everyday operation in 2025.³⁰ On its April 2025 earnings call, its then chief executive said an AI agent had sent 26 million customer calls to the right adviser more accurately than before, and that AI would route more than half of the company’s calls during 2025.³¹

The bigger the company, the more it needs clear rules and named owners for AI. McKinsey’s March 2025 research found that a CEO who oversees how AI is run is one of the factors most closely linked to higher profit from AI. In 28% of organizations using AI, the CEO was responsible for it.³² McKinsey’s later survey found that just over half of organizations using AI had suffered at least one negative consequence. The most common was inaccurate output.⁵

---

What fails, and which companies backed down

Most AI projects still never reach everyday use, or never pay back. Gartner, an IT research firm, predicted that at least 30% of generative AI projects would be abandoned after the proof of concept by the end of 2025.³³ In S&P Global’s 2025 survey, 42% of companies said they had abandoned most of their AI initiatives, up from 17% a year earlier.³⁴ MIT NANDA’s preliminary report said about 95% of organizations saw no measurable effect on profit and loss from their generative AI pilots. That figure rests on 52 interviews and 153 conference survey responses. The report itself calls its figures “directionally accurate based on individual interviews rather than official company reporting.”²³

Several well-known AI-first moves have been partly reversed. Klarna said in 2024 that its AI assistant was doing the work of 700 customer-service agents. In 2025 its CEO said cost had been “a too predominant evaluation factor” and that quality had suffered. Klarna would hire more human agents.³⁵ Starbucks retired an AI tool for counting stock after about nine months, following complaints that its counts were wrong.³⁶ In July 2025, Commonwealth Bank of Australia said 45 customer-service jobs would go after it introduced an AI voice bot. The Finance Sector Union challenged the decision. Within weeks, the bank said its assessment “did not adequately consider all relevant business considerations”. It said the jobs were not redundant, and apologised.³⁷

The cases in this guide split along one line. The companies that got results changed one specific process and measured the outcome. The companies that failed rolled AI out widely on orders from the top, judged it mainly on cost and did not check quality.

---

AI in seven industries

Technology and software companies come closest to the pure form of AI-first, because AI is often the product. Stripe built a foundation model for payments: one large AI model trained on tens of billions of its transactions. Stripe says the model raised its detection rate for card-testing attacks on large businesses by 64% almost overnight (company-reported).³⁸ In a card-testing attack, criminals try stolen card numbers with small payments.

Marketing teams took up generative AI early. Adore Me, a lingerie brand, used Writer to produce product descriptions and stylist notes. The time it spent each month on product descriptions for online marketplaces fell from 20 hours to 20 minutes (vendor case study).³⁹ Bayer Australia used Google Cloud machine learning to predict when searches for colds and flu would rise, and adjusted its search ads to match. It reported an 85% rise in click-through rate (the share of people who see the ad and click it) and a 33% lower cost per click than the year before (Google case study).⁴⁰

Retailers use AI from the first contact with a customer to the warehouse. In a 2026 survey of retail and consumer-goods companies by NVIDIA, which makes the chips most AI runs on, 91% said they were using or assessing AI.⁴¹ McKinsey estimates that generative AI could create $240–390 billion in economic value for retailers. That is a forecast, not a measured gain.⁴²

Manufacturers use AI mostly in engineering, maintenance and quality. Siemens added AI agents to its Industrial Copilot in 2025. One customer, thyssenkrupp Automation Engineering, is rolling the Copilot out worldwide and reports better code quality and faster development (Siemens press release).⁴³ Rolls-Royce uses digital twins of aircraft engines: software copies of real engines, fed with data from them. It says this has extended the time between maintenance for some engines by up to 50%.⁴⁴ In the EU, manufacturers that use AI use it most often for marketing and sales (30%) and for business administration (27%), according to Eurostat.²

Financial services firms spend heavily on AI and must test and approve their models under strict rules. In NVIDIA’s 2026 survey of more than 800 people working in financial services, 65% said their company was actively using AI, up from 45% a year earlier.⁴⁵ Klarna (see above) shows the risk of judging AI that serves customers mainly on cost.

Healthcare shows why independent evidence matters. An AI medical scribe listens to a consultation and drafts the doctor’s notes. The vendors often promise large time savings. Two randomised trials found smaller ones. At UCLA Health, one of the two tools tested cut the time doctors spent writing notes by 9.5%. The other made no difference beyond chance. Measures of burnout improved a little. Doctors used the tools in only 33.5% and 29.5% of visits.⁴⁶ At the University of Wisconsin, a scribe cut time on notes by 0.36 hours a day and was used for 38% of notes.⁴⁷ At The Permanente Medical Group, Kaiser Permanente’s group of doctors, AI scribes saved an estimated 15,791 hours of note-writing over about the first 15 months.⁴⁸ The scribe company Abridge raised $300 million from investors in June 2025.⁴⁹

Professional services firms took up generative AI quickly, but the gains vary widely. In a BCG survey of professional-services workers in the United States, carried out in late 2024 and published in 2025, 95% used generative AI at least once a month. Yet 53% said it saved them less than a quarter of their time.⁵⁰ McKinsey says 72% of its own staff use its internal AI tool, Lilli, and send it more than 500,000 prompts a month (company-reported).⁵¹ In the American Bar Association’s 2024 technology survey, 30% of lawyers said they used AI tools, up from 11% the year before.⁵² Harvey, which sells AI to law firms, says most tasks it measured were 60–90% faster with its tools (vendor claim).⁵³ The risks are well documented. As of 16 October 2026, one researcher’s database listed 2,199 court decisions involving material that AI had made up, 871 of them involving lawyers.⁵⁴ For small law firms, time saved has not yet become money earned. In a survey by Clio, which sells software to law firms, 86% of solo lawyers and 78% of small law firms had not changed their prices because of AI (vendor survey).⁵⁵

---

Tools and systems at each size

What a company runs depends on its size. Micro and small businesses use online subscription tools that need little technical skill. Medium-sized firms add automation tools and the AI features inside the business software they already pay for. Mid-market companies combine the AI from their main software vendors with a few specialised tools and direct connections to AI models (APIs), on top of one shared store of company data. Enterprises build their own: data platforms, teams that run and monitor models, and models trained on their own data, in the cloud and in their own data centres.

Vendors are also building AI into existing plans and raising prices. Google put its Gemini AI into all Workspace business plans and raised the price of Business Standard by $2 per user per month.⁵⁶

The parts are the same at every size. Data sits at the bottom. AI models, and the work of running them, sit above it. At the top are the applications people use: assistants, agents and tools made for one industry. What changes with size is how much of each part a company owns and how much it rents.

Read return-on-investment figures with care. IDC did research for Microsoft, which sells AI tools. Companies told IDC they got an average return of $3.50 for each $1 invested in AI in 2023, and $3.70 for each $1 invested in generative AI in 2024. Business leaders reported these figures themselves; nobody measured them.⁵⁷ People’s own estimates can be wrong in either direction. In a randomised trial, experienced software developers took 19% longer on tasks when they used AI. Afterwards they believed AI had made them about 20% faster.⁵⁸ Across the US workforce, workers’ own estimates of the time generative AI saves them added up to about 2% of all work hours in April 2026.⁵⁹ The independent evidence in this guide, from field studies and from the high abandonment rates, points to returns that are smaller and far more uneven.

---

Who does what: roles and leadership

AI-first needs new responsibilities more than new job titles. Someone has to own each process that AI supports, from start to finish. Someone has to keep the prompts, settings and connections between systems working as the tools change. Someone has to answer for risk and legal compliance. In a small firm, one person often does all three. In an enterprise, these become separate roles, often in a central AI team.

Microsoft, which sells AI tools, describes three phases in its 2025 Work Trend Index. First, AI works as an assistant to each employee. Next, AI agents join teams as “digital colleagues” that take on specific tasks. Finally, people direct agents that run whole processes.⁶⁰ That is Microsoft’s forecast, not a measured result.

Leaders matter as much as the technology. McKinsey found that AI high performers were about three times as likely as other companies to have senior leaders who show ownership of and commitment to AI.⁵ In practice, that means leaders who use the tools themselves. They present AI as help for staff, not a replacement for them. And they treat what AI produces as a draft that a person answers for.

---

What to do, by company size

• Micro and small businesses: start with one costly, repetitive process. Use ready-made tools, with a person checking what they produce. Before you do more, measure the time saved, and whether sales or fees moved.
• Medium-sized businesses: name one person responsible for AI. Put a short AI-use policy and staff training in place. Buy and set up tools rather than building them.
• Mid-market companies: use your ability to decide quickly. Start with the AI in your main software and a few specialised tools. Build shared data and rules for AI early.
• Enterprises: change the workflows instead of adding tools to them. Put senior leaders visibly in charge. Run AI as part of how the company operates, not as an IT project.

At every size, buying the technology is the smaller job. Most of the work is changing how people and processes work (BCG’s 10-20-70 rule, above). Companies that treat AI-first as a software purchase tend to join the majority whose projects stall. The companies that gain, in the evidence so far, are the ones that changed how the work gets done.

Sources

1. 9to5Google, reporting Sundar Pichai’s remarks and 2016 Founders’ Letter, 28 April 2016: https://9to5google.com/2016/04/28/founders-letter-2016-pichai/ ↩

2. Eurostat, “Use of artificial intelligence in enterprises,” 2025 data: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises ↩ ↩² ↩³ ↩⁴

3. US Census Bureau, Business Trends and Outlook Survey, May 2026: https://census.gov/library/stories/2026/05/ai-use-businesses.html ↩ ↩²

4. I. Yotzov, J. M. Barrero, N. Bloom and others, “Firm Data on AI,” NBER Working Paper 34836, 2026: https://www.nber.org/papers/w34836 ↩

5. McKinsey, “The state of AI in 2025,” 5 November 2025 (archived copy; the live URL now shows the 2026 edition): https://web.archive.org/web/20251215/https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai ↩ ↩² ↩³ ↩⁴ ↩⁵

6. BCG, “CEO’s Guide to Maximizing Value Potential from AI in 2024,” July 2024: https://media-publications.bcg.com/BCG-Executive-Perspectives-CEOs-Guide-to-Maximizing-Value-from-AI-EP0-3July2024.pdf ↩

7. BCG, “5 Rules for Fixing AI and Machine Learning for Your Business,” 2022: https://www.bcg.com/publications/2022/5-rules-for-fixing-ai-and-machine-learning-for-your-business ↩ ↩²

8. McKinsey, “The rewired enterprise: How five companies built to outcompete”: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-rewired-enterprise-how-five-companies-built-to-outcompete ↩

9. Tobi Lütke, internal memo published on X, 7 April 2025: https://x.com/tobi/status/1909251946235437514 ↩

10. Fortune, “Duolingo CEO admits controversial AI memo did not give enough context,” 18 August 2025: https://fortune.com/2025/08/18/duolingo-ceo-admits-controversial-ai-memo-did-not-give-enough-context-insists-company-never-laid-off-full-time-employees ↩

11. Annalisa Ferrando, Sara Lamboglia, Judit Rariga and Maurice Schmidt, “Adoption and investment in AI across the euro area: insights from harmonised firm-level data,” ECB Occasional Paper 395, 2026: https://www.ecb.europa.eu/pub/pdf/scpops/ecb.op395.en.pdf ↩

12. JPMorganChase Institute, “Understanding the use of AI among small businesses,” April 2026: https://www.jpmorganchase.com/institute/all-topics/business-growth-and-entrepreneurship/understanding-ai-use-by-small-businesses ↩

13. Federal Reserve Banks, Small Business Credit Survey, “2026 Report on Employer Firms,” March 2026: https://www.fedsmallbusiness.org/-/media/project/clevelandfedtenant/fsbsite/reports/2026/2026-report-on-employer-firms/2026-report-on-employer-firms.pdf ↩

14. Erik Brynjolfsson, Danielle Li and Lindsey Raymond, “Generative AI at Work,” The Quarterly Journal of Economics 140(2), 2025: https://doi.org/10.1093/qje/qjae044. Earlier version: NBER Working Paper 31161: https://www.nber.org/papers/w31161 ↩

15. Anders Humlum and Emilie Vestergaard, “Large Language Models, Small Labor Market Effects,” May 2025 version (BFI Working Paper 2025-56; 2.8% average time savings): https://bfi.uchicago.edu/wp-content/uploads/2025/04/BFI_WP_2025-56-3.pdf. Revised as NBER Working Paper 33777: https://www.nber.org/papers/w33777 ↩

16. Nicholas G. Otis, Rowan Clarke, Solène Delecourt, David Holtz and Rembrand Koning, “The Uneven Impact of Generative AI on Entrepreneurial Performance: Evidence from a Field Experiment in Kenya,” Management Science, 2026: https://doi.org/10.1287/mnsc.2024.06909 ↩

17. Daily Coffee News, 18 October 2024: https://dailycoffeenews.com/2024/10/18/weekly-coffee-news-gpt-assisted-grinder-hack-an-espresso-smoker ↩

18. ARR Club, “Aragon AI at $10M ARR in 2 years with 9 people”: https://www.arr.club/signal/aragon-ai-arr-at-10m-in-2-years-with-9-people ↩

19. Hyunso Kim, Hyo Kang and Jaeyong Song, “Generative AI Fuels Solo Entrepreneurship, but Teams Still Lead at the Top,” arXiv 2605.10291, May 2026: https://arxiv.org/abs/2605.10291 ↩

20. OECD, “Generative AI and the SME Workforce,” 2025: https://www.oecd.org/en/publications/generative-ai-and-the-sme-workforce_2d08b99d-en.html ↩

21. Gibson Dunn, “EU AI Act Omnibus agreement: postponed high-risk deadlines and other key changes,” May 2026: https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/ ↩

22. RSM, “Middle market embraced AI; now comes the hard part,” 2026: https://rsmus.com/newsroom/2026/rsm-survey-middle-market-embraced-ai-now-comes-hard-part.html ↩

23. A. Challapally, C. Pease, R. Raskar and P. Chari, MIT NANDA, “The GenAI Divide: State of AI in Business 2025,” preliminary findings, July 2025. Copy hosted at mlq.ai, not on an MIT domain: https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf ↩ ↩²

24. Fortune, commentary by BCG-affiliated researchers, 10 January 2025: https://fortune.com/2025/01/10/ai-artificial-intelligence-goldilocks-mid-sized-companies ↩

25. BCG, “How midmarket firms can get value from AI,” September 2026: https://www.bcg.com/publications/2026/midmarket-firms-get-value-from-ai ↩

26. Volkswagen Group, “Boosting innovation, reshaping mobility: Volkswagen Group invests in AI,” September 2025: https://www.volkswagen-group.com/en/press-releases/boosting-innovation-reshaping-mobility-volkswagen-group-invests-in-ai-19852 ↩

27. Markets Media, “JPMorgan Chase raises target for value realised from AI,” May 2023: https://www.marketsmedia.com/jpmorgan-chase-raises-target-for-value-realised-from-ai/ ↩

28. Bank of America newsroom, March 2026: https://newsroom.bankofamerica.com/content/newsroom/press-releases/2026/03/bofa-ai-and-digital-innovations-fuel-30-billion-client-interacti.html ↩

29. Fortune, “Goldman Sachs CEO David Solomon on AI,” 17 January 2025: https://fortune.com/2025/01/17/goldman-sachs-ceo-david-solomon-ai-tasks-ipo-prospectus-s1-filing-sec ↩

30. Becker’s Payer Issues, “UnitedHealth has 1,000 AI use cases,” 2025: https://www.beckerspayer.com/payer/unitedhealth-has-1000-ai-use-cases-report/ ↩

31. UnitedHealth Group, first quarter 2025 earnings call remarks, 17 April 2025: https://www.unitedhealthgroup.com/content/dam/UHG/PDF/investors/2025/UNH-Q1-2025-Remarks.pdf ↩

32. McKinsey, “The state of AI: How organizations are rewiring to capture value,” March 2025: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value ↩

33. Gartner, “Gartner predicts 30% of generative AI projects will be abandoned after proof of concept by end of 2025,” 29 July 2024: https://www.gartner.com/en/newsroom/press-releases/2024-07-29-gartner-predicts-30-percent-of-generative-ai-projects-will-be-abandoned-after-proof-of-concept-by-end-of-2025 ↩

34. CIO Dive, “AI project failure rates are on the rise,” reporting S&P Global Market Intelligence, 2025: https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/ ↩

35. Entrepreneur, “Klarna CEO reverses course by hiring more humans, not AI,” 2025: https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396; CEO quote via Customer Experience Dive, quoting Bloomberg, 9 May 2025: https://www.customerexperiencedive.com/news/klarna-reinvests-human-talent-customer-service-AI-chatbot/747586/ ↩

36. Fortune, 28 May 2026: https://fortune.com/2026/05/28/starbucks-quietly-retired-ai-inventory-agent-barista-complaints-hallucinations/ ↩

37. Information Age (Australian Computer Society), “CBA reverses AI-driven job cuts, admits ‘error’,” 21 August 2025: https://ia.acs.org.au/article/2025/cba-reverses-ai-driven-job-cuts—admits—error-.html ↩

38. TechCrunch, “Stripe unveils AI foundation model for payments,” 7 May 2025: https://techcrunch.com/2025/05/07/stripe-unveils-ai-foundation-model-for-payments-reveals-deeper-partnership-with-nvidia ↩

39. Writer, “Adore Me customer story”: https://writer.com/blog/adore-me-customer-story ↩

40. Google, “Bayer: search trends and demand predictions”: https://business.google.com/us/think/ai-excellence/search-trends-demand-predictions-bayer/ ↩

41. NVIDIA, “State of AI in Retail and CPG,” 2026: https://blogs.nvidia.com/blog/ai-in-retail-cpg-survey-2026 ↩

42. McKinsey, “LLM to ROI: How to scale gen AI in retail”: https://www.mckinsey.com/industries/retail/our-insights/llm-to-roi-how-to-scale-gen-ai-in-retail ↩

43. Siemens, “Siemens introduces AI agents for industrial automation,” 12 May 2025: https://www.businesswire.com/news/home/20250512053219/en ↩

44. CIO, “Rolls-Royce turns to digital twins to improve jet engine efficiency,” 2021: https://www.cio.com/article/3620993/rolls-royce-turns-to-digital-twins-to-improve-jet-engine-efficiency.html ↩

45. NVIDIA, “Survey Reveals the Financial Services Industry Is Doubling Down on AI Investment and Open Source,” 22 January 2026: https://blogs.nvidia.com/blog/ai-in-financial-services-survey-2026/ ↩

46. Lukac and others, “Ambient AI scribes in clinical practice: a randomized trial,” NEJM AI, 2025: https://doi.org/10.1056/AIoa2501000 ↩

47. Afshar and others, “A pragmatic randomized controlled trial of ambient artificial intelligence to improve health practitioner well-being,” NEJM AI, 2025: https://doi.org/10.1056/AIoa2500945 ↩

48. American Medical Association, “AI scribes save 15,000 hours and restore the human side of medicine”: https://www.ama-assn.org/practice-management/digital-health/ai-scribes-save-15000-hours-and-restore-human-side-medicine ↩

49. Latham & Watkins, “Latham & Watkins advises Abridge in US$300 million Series E funding round,” June 2025: https://www.lw.com/en/news/2025/06/latham-watkins-advises-abridge-in-us300-million-series-e-funding-round ↩

50. BCG, “Gen AI in professional services,” 2025: https://www.bcg.com/publications/2025/gen-ai-in-professional-services ↩

51. McKinsey, “Rewiring the way McKinsey works with Lilli”: https://www.mckinsey.com/capabilities/tech-and-ai/how-we-help-clients/rewiring-the-way-mckinsey-works-with-lilli ↩

52. LawNext, “ABA tech survey finds growing adoption of AI in legal practice,” March 2025: https://lawnext.com/2025/03/aba-tech-survey-finds-growing-adoption-of-ai-in-legal-practice-with-efficiency-gains-as-primary-driver.html ↩

53. Harvey, “How Harvey saves lawyers time,” 31 July 2025: https://www.harvey.ai/blog/how-harvey-saves-lawyers-time ↩

54. Damien Charlotin, “AI Hallucination Cases” database, count as of 16 October 2026: https://www.damiencharlotin.com/hallucinations/ ↩

55. Clio, “Legal Trends for Solo and Small Law Firms 2026,” press release, 4 May 2026: https://www.clio.com/about/press/2026-solo-small-firm-report/ ↩

56. Google Workspace blog, announcement of Gemini in all Workspace business plans and the new prices, 15 January 2025: https://workspace.google.com/blog/product-announcements/empowering-businesses-with-AI ↩

57. Microsoft, “New study validates the business value and opportunity of AI,” 2 November 2023: https://blogs.microsoft.com/blog/2023/11/02/new-study-validates-the-business-value-and-opportunity-of-ai/ and “IDC’s 2024 AI opportunity study,” 12 November 2024: https://blogs.microsoft.com/blog/2024/11/12/idcs-2024-ai-opportunity-study-top-five-ai-trends-to-watch/ ↩

58. METR, randomised study of experienced open-source developers using early-2025 AI tools, 10 July 2025: https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/ ↩

59. Alexander Bick, Adam Blandin and David Deming, “Generative Artificial Intelligence, Time Savings: Employed Adults,” series RPSGENAITSALL, Federal Reserve Bank of St. Louis (FRED), April 2026 value: https://fred.stlouisfed.org/series/RPSGENAITSALL ↩

60. Microsoft, Work Trend Index 2025, “2025: The year the Frontier Firm is born”: https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born ↩