While most are not as newsworthy as AI that summons your car or delivers your package, AI services are already widely used behind the scenes by businesses worldwide. The agreements underlying these services mostly go unnoticed. Thus far they have simply applied well-accepted cloud services terms, but as the reach of AI services expands, new contract structures will be needed.

This article proposes that professional services and particularly IT outsourcing agreements provide a better model for these more analytical AI services.

Cloud Services Set the Stage for AI

In the early days of the cloud computing revolution, opinions were divided about whether the new centrally-hosted solutions that took control out of the hands of customers should be thought of as another form of software solution or as an IT outsourcing relationship. These contracting perspectives represented opposite ends of a spectrum in many respects and where there was disagreement on the contract model, cloud adoption was much slower.

In spite of some holdouts, notably within the financial services industry,1 the debate has long since been settled. Large enterprises across all industries rely on cloud services, with the 2019 market for public cloud services estimated to exceed $200 Billion.2

The geographical reach and scale of leading providers has reduced costs and increased the availability and security of most cloud services beyond the point of debate. With little or no need for negotiation, customers can gain quick access to commercial AI services through online subscription agreements providing:

  • Detailed availability and security commitments;
  • Reasonable IP infringement indemnity;
  • Clear delineation of ownership rights in IP and data; and
  • Limited contractual liability, typically tied to the fees paid over the subscription term.

While the services are designed to retain customers, price competition is the rule among suppliers, and the customer can generally take their data and their business elsewhere at the end of the term without penalty.

Commercial AI Services Today

Like cloud services, AI solutions are now available to serve a wide range of business functions. Some of the most common include industrial automation, customer relationship management, and logistics.3 It is therefore not surprising that cloud service providers have begun to offer AI services to their customers on substantially the same model as they offer cloud services.

The model (based roughly on software licensing) still makes sense where the AI focuses on process improvement, performing some identified task or function, in an incrementally better way (call this “Process AI”). Process AI can label or sort images, convert them to text, perform translations, and process vast amounts of data to deliver tailored results more cheaply than other methods.

For example, by using AI to deliver improved search results to its customers with (apparently) very little patience for taking time to discover entertaining content, Netflix reportedly saves $1 Billion annually in potential lost revenue.4

In these use cases, the task is generally low risk, and the user can directly see the outcome and assess its value. Thus the agreement structure of a standard cloud service, with very low liability on the provider but also relatively low risk and discernible value for the buyer, makes perfect sense. This is where the commercially available AI services market stands now, led by the large cloud platform and service providers offering standard cloud terms available with a click of the mouse.5

Evolving AI Services

While Process AI services are clearly providing value to users, principally by enabling better use of growing data sets in targeted ways, other AI services aim to do more. Emerging AI solutions promise to provide advice, interpretation, or find previously unknown solutions (call this “Advisory AI”). Instead of performing a defined function in a faster, more thorough or more detailed way, Advisory AI aims at making discoveries, performing deeper or new analyses or applying previously unknown or different approaches to data in order to yield new, or novel learnings or outcomes.6

Yet Advisory AI services offerings are fewer and the commercial models not well-established.

For Advisory AI solutions, the roles and expectations of the parties differ from Process AI, and the appropriate commercial terms simply cannot be adequately addressed in a typical cloud services agreement. Often, neither the exact processes or steps to be performed nor the outcomes are known in advance, and applications of Advisory AI services may give rise to greater risk or liability, or encounter ethical issues or biases.

Not surprisingly, much of this type of AI work remains in the areas of research and development, and if they make a solution available at all, providers of Advisory AI solutions may offer them for evaluation or trial use only. In search of a contract model that works for Advisory AI, the earlier debate over how best to structure a cloud services agreement (software license or professional service?) is once again useful.

Commercial Advisory AI Agreements

Logically, Advisory AI offerings resemble human-delivered professional or advisory services more than software solutions. Thus, as these offers mature, the interests of the provider and customer should be better served by adopting some of the elements of an outsourcing or similar professional services structure rather than a SaaS type contract, as suggested in the table below.

  Software License = primarily a technology offering Outsourcing Transaction = primarily a professional service
Term  Fixed subscription Ongoing - multi-year
Upfront Cost  Low High
Time to Deploy/Change  Short Long
Ease of Change/Move Easy Difficult
Performance Standard Focused on defined performance or availability Multi-factor/custom
Liability on Provider  Limited/low Limited/higher
Fee Structure  Standard/metered Shared upside and downside
Agreement Form Standard/provider Custom
Model Applies to: Cloud Services, Process AI Advisory AI

If viewed through this lens, the parties agreeing to a deployment of an Advisory AI solution might be concerned with the following:

  • For the Provider:
    • A pricing model that retains Process AI elements such as initial engagement + platform access and usage-based charges, but also takes into account scale-up and learning time, as well as the value of specialized analysis and interpretation (akin to time and materials);
    • The concept of a success fee or other method of sharing in the upside generated by a successful engagement, to enable the Process AI provider to realize a desired return on their substantial up front development and ongoing investment;
    • Retaining rights to use and apply learnings and know-how gained during a customer engagement in the future, even if those learnings are unforeseen or unrelated to prior solutions or other prior output from the solution;7 and
    • A willingness to take on a greater measure of risk, particularly where the exact methodologies are not disclosed, and where the pricing includes a share in upside.
  • For the Customer:
    • More hands-on engagement in the deployment, both at start up, and throughout, to inform the tuning of the solution or the analysis of specialized customer data;
    • An element of performance-based pricing – e.g. fees related to value delivered;
    • A willingness to share in the value of efficiencies and learnings gained throughout the work; but also
    • An expectation that the provider will take on greater liability for bad outcomes, especially when they are in the best position to understand or foresee them, and when the provider shares in upside gain.


AI services have already become an important component of the IT portfolio for many large and small businesses. The easy application of existing cloud services agreements as a contractual structure for AI services has helped accelerate AI’s adoption. As the reach of AI expands, new structures are needed to drive adoption of a set of potentially valuable AI solutions.

The professional services/IT outsourcing framework provides a useful contractual model on which businesses commercializing and acquiring these Advisory AI services and their counsel can build.


1  Jamie Dimon, CEO JPMorgan Chase, Annual letter to shareholders 2018, wherein he committed the bank to be “all in” on cloud and AI. “We were a little slow in adopting the cloud . . . My early thinking was that it was just another term for outsourcing.”
2  Gartner, Sept. 12, 2018. https://www.gartner.com/en/newsroom/press-releases/2018-09-12-gartner-forecasts-worldwide-public-cloud-revenue-to-grow-17-percent-in-2019.
3  Nichols, Greg, ZDNet, November 15, 2019 https://www.zdnet.com/article/why-business-leaders-are-short-sighted-on-ai/?mod=djemAIPro.
4  Deloitte. Insights: State of AI in the Enterprise, 2nd Edition, 2018.
5  See, e.g., publicly available offerings from Google https://cloud.google.com/solutions/#smart-business-analytics-ai, AWS https://aws.amazon.com/machine-learning/#AI_SERVICES, Microsoft https://azure.microsoft.com/en-us/overview/ai-platform/. See also Deloitte, 2018; associating the accelerating use of cognitive technologies (“often focused on specific, job-related tasks”) in part with their growing availability through enterprise software and through as-a-service models, both of which can be deployed “with low initial cost and minimal risk.”
6  See Chow and Albert, blog post on DWT.com https://www.dwt.com/blogs/artificial-intelligence-law-advisor/2019/09/uspto-comments-on-ai-patent-applications.
7  In an outsourcing or management consulting engagement, the service provider will typically retain rights to “intellectual capital” in the form of the learnings their personnel gain through their work, which adds to their value and which they need to apply in future engagements.