• Businesses are rapidly adopting AI, driven by improving technology, decreasing costs and a retreat in comprehensive AI regulation.
• Companies face increasing risks if they delay AI implementation, including “shadow IT” and falling behind competitors who began experimenting sooner.
• Successful AI integration requires clear goals, strong organizational buy-in, secure enterprise tools, quality data, targeted training, flexible approaches to legal and compliance hurdles and well-developed pilot programs.

Over the past several months, we have observed a significant increase in AI adoption among our clients. In this Debevoise Data Blog post, we provide five reasons why some businesses are accelerating their use of AI and eight factors that impact whether those efforts succeed.

Five Reasons Why Businesses Are Accelerating AI Adoption

AI Is Pretty Good Now and Probably Will Continue to Improve

Starting with the release of reasoning models and deep research tools earlier this year, we have seen significant improvements in the capabilities of the frontier AI models. Although they still make mistakes, fabricate citations and omit important facts and arguments, the best AI models are now able to add significant value to many work-related tasks, including brainstorming, conducting complex research, summarizing, translating and preparing first drafts of various documents. We expect their performance to continue to improve and for hallucinations and error rates to continue to decrease.

AI Is Getting More Secure and Cheaper

Over the last year, the costs of enterprise-grade AI (i.e., GenAI tools licensed to the business that provide an acceptable measure of confidentiality and cybersecurity and do not train on client data) have fallen significantly. For instance, OpenAI’s users have seen prices drop roughly 90% from about $30 per million input tokens for GPT‑4 down to $3 per million for GPT‑4o over the last 16 months. Although increased token consumption for certain complex tasks, like legal analysis or agentic workflows, may partly offset these savings, the overall trend remains clear: AI capabilities are becoming more financially accessible, enabling broader adoption across industries.

Regulation Is Lighter Than Anticipated

As we discussed in a recent blogpost, there has been a general retreat from efforts to enact comprehensive AI legislation in favor of more targeted regulations addressing specific harms like nonconsensual sexualized deepfakes and mental health chatbots for minors—important laws that fill actual gaps in the existing legal landscape but which are unlikely to impact most businesses. In addition, the recently released Executive Orders and the White House’s new Winning the Race: America’s AI Action Plan are parts of a comprehensive framework that aims to continue the shift in U.S. federal government policy towards promoting AI adoption and competitiveness.

Slow Adoption of AI Creates Risk of Off-Platform Use

In a recent blog post, we wrote that one of the biggest AI risks for companies today is moving too slowly with AI adoption because powerful AI tools are readily available to employees as consumer products on their personal devices. Although it violates company policy, many employees will use consumer AI for work-related tasks if there is no company-approved enterprise AI tool available, a phenomenon known as “shadow IT.” So not allowing employees to use AI for work is no longer the lowest-risk option.

Recognition That AI Adoption Takes Time and the Risk of Getting Left Behind

Many companies have sat on the sidelines, waiting for AI to “get good,” figuring that when that time came, (1) they would know it, and (2) they can adopt AI quickly. But some companies are starting to question those assumptions. First, the point at which AI becomes “good enough” to create meaningful value is not a one-size-fits-all threshold and may be difficult to discern from the outside looking in. It is often difficult to know whether adopting AI can provide significant value in a particular business without trying it in several well-constructed pilots, which takes time. Companies that wait too long to start that process may find it difficult to catch up with peers that started to adopt AI earlier.

Eight Hallmarks of Successful AI Adoption

Companies that decide to push forward with AI adoption inevitably encounter organizational hurdles that can slow down their initiatives. A recent Fortune article cited an MIT study for the proposition that 95% of AI pilots fail. As Nathanial Whittemore observed, the study says more about common organizational struggles with change management than about AI itself.

Permission to Fail

• Many companies that adopt AI successfully treat it like a Research & Development program, not a traditional software implementation, such that some degree of failure should not only be tolerated but expected.

Senior Leadership Buy-In

• Clear support from senior leadership for AI adoption can be helpful to secure the necessary resources and to get buy-in from group leaders, especially because of the likelihood of some wasted time and money.

Team/Group Leadership Buy-In

• Support is also needed from the leaders of the relevant business units and corporate functions, so they can give their teams the time and resources necessary to succeed, as well as the assurance that they are not just training their machine replacements. Ideally, they can be role models for AI use for their teams.

Good Tools

• Most company AI use cases that add significant value require access to secure enterprise AI tools. Vetting and onboarding those vendors takes time but unlocks many options for use that are not available from consumer AI tools.

Good Context (i.e., Documents/Data) for the AI to Use

• For many use cases, cutting edge AI alone is not good enough to solve the problem. But if that AI has access to highly accurate and relevant data (i.e., context), the chances for success increase significantly. Finding that data and making sure the company has all the necessary rights to use the data to achieve its goals can be complicated and take time, but it is often worth the effort.

Training

• Providing AI tools to employees who are not sure how to use them, or what they are allowed to do with them, often results in little use or use for low-value tasks only. Proper AI training is an ongoing project, which means that companies need AI professionals who can answer employees’ questions, work with them to design and implement pilot programs and share successes and failures efficiently across the organization.

Creative and Flexible Legal, Compliance and Risk Groups

• The risks associated with adopting AI (e.g., cybersecurity, privacy, bias, regulatory compliance, quality control) are real. But those are reasons to adopt AI thoughtfully and gradually, not reasons to forego AI adoption entirely. In the absence of a clear legal prohibition, creativity from legal, compliance and risk teams allows them to evaluate the actual (not hypothetical) risks of particular use cases, identify ways to lower those risks without significantly undermining the value of the use case and then balance those risks against the upside of the use case and the downside of off-platform AI use.

Well-Designed Pilots

• As discussed above, AI tools have improved dramatically in the last six months, and hallucinations have significantly decreased, which has led to a rise in overreliance because it is hard to find mistakes with a low error rate. Companies that have successfully adopted AI have been able to structure their pilot programs with ongoing quality controls that are effective at catching infrequent mistakes and omissions that occur over the long term.

This publication is for general information purposes only. It is not intended to provide, nor is it to be used as, a substitute for legal advice. In some jurisdictions it may be considered attorney advertising.