Acquiring an AI business – Not your typical tech transaction

Author(s): Michael Fekete, Sam Ip, Shira Dveris

Dec 13, 2021

The race for leadership in artificial intelligence (AI) technology picked up speed in the past year. Multinational and domestic players were relentless in their quest for game-changing AI-enabled solutions and the accompanying talent, including leading data scientists and experienced machine learning engineers. While development and refinement of AI technologies still represent an important growth area for these issuers, growth through acquisition was certainly a noticeable trend in 2021 – one that we anticipate will continue to intensify in the coming years.

The past year also brought into focus how the acquisition of an AI company raises unique risks for purchasers that, in many respects, differ from those in transactions involving traditional technology and software companies. These risks require a re-thinking of legal due diligence and allocations of risk in purchase agreements. Given the novelty of these types of transactions, expertise in AI acquisitions is still developing and purchasers should ensure that they have professional advisors with the best experience and knowledge available to protect themselves.

Why acquisitions of AI companies require new thinking   

The starting point for providing effective legal advice on any M&A transaction is to understand the client’s business rationale for making the acquisition. Without insight into how the client values the target’s assets, it is challenging to ensure that client’s interests are protected through both the purchase agreement and the legal due diligence that informs it. In the case of all technology transactions, it is critical to understand how the target’s technology assets will be used by the purchaser.

For a traditional technology company, the primary strategic asset is usually its software. Understanding software-driven businesses and the risks that need to be investigated in these businesses in connection with an acquisition transaction is a well-trodden path. In undertaking due diligence with respect to software, it is typical for the purchaser and its professional advisors to undertake a deep dive into the target’s software code and software development practices, focusing on intellectual property and data security. This often includes assessing the target’s use of open source software and the presence of software bugs and security vulnerabilities. These considerations are also reflected in multiple elements of the transaction purchase agreement, including in software-focused representations and warranties, indemnities and closing conditions.

Unlike a software business, the core value of an AI company is often found in the company’s rights to datasets and the proprietary models that are used to ingest and analyze the data. It is the combination of data and these models that enables computers to mimic human intelligence and learn over time as they train themselves to perform increasingly complex tasks. Although an AI company may have developed proprietary software, such as a user interface to present the analysis performed by the company’s models, the code for the software usually performs a function that is only ancillary in its value to the company’s primary business.

Understanding the different drivers of value for AI companies is critical in the context of M&A transactions as these drivers change the nature of the purchaser’s focus. Similarly, advisors seeking to protect their purchaser clients need to appreciate this distinction in order to provide the right advice.

When assessing an AI target from a due diligence perspective, the purchaser and its advisors must adopt an approach that reflects the value of the target’s dataset and proprietary models. Rather than emphasize looking at software development and data security issues, purchasers must expand their focus to include the target’s rights to own and use data, the target’s ownership of proprietary models and improvements, the “outputs” of the models and the company’s practices to train, improve, test, maintain and explain such models. Investigating complex datasets and models from a legal diligence perspective requires a thorough knowledge of the construct and use of these assets in a manner that differs significantly from traditional technology acquisitions. Given the rapidly expanding uses of AI, knowledge of privacy considerations is also critical.

Once a purchaser and its advisors have sufficiently assessed the underlying assets and risks, and conducted thorough diligence on the target, these findings must be appropriately reflected in the transaction purchase agreement. It is important that a purchase agreement for an AI business be tailored to address AI and its unique attributes and risks. Although each transaction needs to be considered individually, there are a number of key considerations that should be addressed.

In particular, definitions require careful crafting to ensure that the agreement sufficiently captures relevant characteristics of artificial intelligence. For example, definitions focused on “AI technology” should be drafted sufficiently broadly to capture both techniques that enable computers to mimic human intelligence, including deep learning, machine learning and algorithms that make use of or employ neural networks, statistical learning algorithms or reinforcement learning, and software and hardware used to train, test and deploy the AI solution.

In preparing representations and warranties regarding the business, the purchaser should seek comprehensive disclosure and protection through reps and warranties that, among other considerations that may be identified through diligence, address

• ownership of, and rights to use, AI models and datasets, including those that are both owned and licensed
• the quality of the company’s datasets, including the degree of completeness, consistency and accuracy
• the company’s practices relating to the testing, improvement and development of AI models
• responsible use and ethical design of AI, including testing for bias or other harmful impacts
• use of facial recognition or other high risk use-cases that leverage AI
• the allocation of AI-related liability in agreements with suppliers and customers
• compliance with laws and industry standards and practices applicable to AI

These representations and warranties require particular attention to ensure that all aspects of the AI business are subject to comprehensive disclosure. Appropriate care needs to be taken in preparing indemnification provisions in the purchase agreement to appropriately balance liability. Particular consideration should be given to whether the quantum of the “hold-back” should be increased or the timeframe for paying out the hold-back extended. 

Looking to the future

Over the past year, sophisticated purchasers have demonstrated their willingness to invest time and resources into following best practices for AI acquisitions as they seek to grow their ownership of these assets. This includes having their professional advisors explore the nuances of AI as part of the diligence process and tailoring purchase agreements to reflect their findings. We expect this trend to continue, given the significant growth in AI-focused businesses, the significant growth in AI M&A and ongoing demand and competition for assets and talent. We also expect that purchasers of AI companies will increasingly seek to engage sophisticated legal counsel who have a comprehensive understanding of AI and how to protect their clients’ interests.