Deal Structuring for Agentic Managed Services

Why are Agentic Managed Services Different?

Managed services have long been performed by provider personnel leveraging various software to automate some activities, most notably SaaS tools, to increase productivity. What makes AI agents different from prior automation efforts is that the generative models underlying the AI agents are non-deterministic: the same input may produce different outputs. Behavior may also change as models, data, prompts, and upstream systems change, without any deliberate modification. As a result, performance cannot be fully specified in advance, and testing before go-live does not assure that the agents will behave the same way months later. That characteristic, rather than autonomy as such, drives the key changes in deal structure.

Under an agentic managed service deal structure, AI agents receive work, plan steps, retrieve information, call APIs, update systems of record, generate communications, escalate exceptions, and in some cases, take consequential actions at speed and scale without a person’s approval. Representative engagements are emerging across business functions.

The characteristics of agentic AI drive substantial differences across all three elements of a managed services deal structure: the services, the performance standards, and the charges. The variability of AI agents, together with their greater autonomy, scale, and speed, changes how the parties can best define, measure, govern, price, and allocate responsibility for the services. It also calls for forms of development, support, supervision, correction, and management that differ from those used for human personnel. Because the future capabilities and costs of AI agents are uncertain, detailed contractual plans and roadmaps are difficult to produce. That favors a more flexible deal structure, with hybrid delivery, a focus on outcomes, and pricing that aligns the parties toward a successful transformation.

Despite those differences, the commercial model remains managed services. The company outsources responsibility for a function and expects the provider to deliver agreed outputs and service levels on an ongoing basis. The provider has substantial discretion over how it meets those commitments, subject to contractual guardrails. Charges vary based on volumes in a manner designed to reflect cost in a way that provides acceptable levels of profit for the provider and savings for the company. Exit and liability provisions protect the parties against risks of unforeseen problems.

How Should Agentic Managed Services Be Defined in Contracts?

Core Service Description

The service description should define the business or technology function the provider is responsible for performing, not the agents. Agents are delivery resources and will change over the term; the outsourced function and the required work and outcomes should remain the anchor.

In describing the services, the drafters should move away from the tasks that the company’s in-scope personnel perform today and toward the business results to be delivered. That will increase deal value by allowing the provider to re-engineer workflows to gain the cost savings and other benefits of agentic AI instead of replicating a human process with an AI agent. This may require a more strategic look at the in-scope function, focusing on why, instead of how, people perform in-scope tasks.

Ongoing Transformation

After a transition, the provider may go live performing the in-scope business functions with provider personnel. The provider then develops, tests, and begins to deploy AI agents with the company’s collaboration and approval. During the term, the provider may continue to automate additional tasks, increasing the share of the function performed by agents and reducing the share performed by provider personnel.

That progressive shift is what we would call an ongoing transformation. With an ongoing transformation, the operating model described in the documents at signing is not the operating model that will be running in year three. The changing mix of agentic and human labor potentially changes the pricing basis, the staffing commitments, the calibration of service levels, and the allocation of risk during the term.

A claims operation illustrates the point. At go-live, agents might perform intake and document extraction while adjusters decide every claim. Two years later, agents might decide claims below a value threshold and adjusters review a sample. The outsourced function may be the same, but the value-maximizing metrics, pricing and operational limitations, will be far different.

The Provider’s Responsibilities for the AI Agents

The service description should also define what the provider must do to operate the AI agents. Typical obligations include onboarding agents into production, validating them before release, monitoring their performance (including for model drift), maintaining prompts and workflows, managing permissions, correcting, reconfiguring, and tuning them (and, where the arrangement includes fine-tuning, re-tuning within agreed controls), handling exceptions, implementing approved changes, and sunsetting. Where the provider is expected to increase automation over the term in an ongoing transformation, that obligation should be stated expressly, distinguished from ordinary maintenance and support, and tied to pricing.

These responsibilities may also include compliance with contractual guardrails that can be amended only through change control. Those guardrails could include, for example, the categories of decisions agents may take; transaction value and volume limits; the data categories agents may access; the systems agents may write to; actions that always require human approval; actions that are never permitted; and decisions reserved to the company regardless of agent capability. For agents authorized to take consequential actions, those contractual guardrails may also address topics including segregation of duties, permitted counterparties or accounts, external communications, privileged or administrative actions, and the circumstances in which an agent may act or communicate on the company’s behalf.

However, as is common in managed services, the provider will have broad operational control within those guardrails. For example, the provider might have decision rights over the specific tasks that AI agents perform at a given time, escalation thresholds, sampling rates, and the current AI-and-human mix. This allows the provider to offer stronger performance commitments and lower pricing by allowing the provider to assume that it will have the control needed to implement cost-effective technologies within the guardrails.

What Performance Commitments Apply to Agentic Managed Services?

Core business service levels remain central to this hybrid deal structure. Availability, response and resolution times, processing timeliness, accuracy, backlog, throughput, and other outcome measures apply whether provider personnel or AI agents perform the services. Indeed, those service levels may become more important, because they measure the value the company receives regardless of whether a person or an AI agent performed the work. In addition, metrics directed at provider personnel, such as turnover, continue to matter, because people remain part of the delivery model.

Agentic delivery may also warrant additional measures directed at agentic performance where the outcomes of the function cannot be effectively measured. In a managed service, the measures most worth considering are those that capture burdens or risks that the outcome metrics do not, such as human-handoff or intervention rates and timeliness; rework rates; and false-positive and false-negative rates in high-consequence areas. Many of these may be better used as reported and monitored metrics that inform governance than as service levels carrying credits, and each would require technical and operational input as to practicality and value.

Service-level credits can continue to operate in the conventional way for measurable service-level failures, but remediation plans implemented by the provider may involve increased monitoring or human oversight, restriction or rollback of affected AI agents, or temporary manual processing. These operational remedies may be particularly important where a performance issue requires immediate correction rather than compensation through service-level credits alone.

What Pricing and Charging Methodologies Fit Agentic Managed Services?

Agentic AI is expected to create substantial business value and to support win-win deals. The managed services providers’ economies of scale, technical skill, experience with agentic AI, access to scarce talent, and ability to leverage solutions across many companies should allow them to price below the cost of similar internal solutions while achieving desired profit margins. Their willingness to invest in company-specific solutions has long made them a preferred path for business process innovations.

Unlocking this value requires pricing methodologies that address the substantial uncertainty about the future capabilities and costs of agentic AI. The pricing methodology must address both the companies’ concern about settling for too little of what might be dramatic cost savings and the providers’ concern about agreeing to share cost savings that may not actually be technically feasible. It also must address the risk of investing in emerging technologies that might shortly be made obsolete by a next generation.

The pricing methodology should be designed to work in an evolving delivery environment, in which human performance of the services declines, agentic performance increases, and the remaining human effort shifts toward oversight. Depending on the arrangement, pricing for the ongoing services may for example include a combination of fixed managed-service charges, transaction or outcome-based pricing metrics, AI-platform or consumption charges, charges associated with human intervention or operational support, and pass-through charges, which might include token costs and other payments to tool and model providers. Prices for specific services might change based on the provider’s mix of human and agentic labor in delivery or the effect on the company’s retained costs and operational performance.

The pricing methodology will be key to aligning incentives on the speed of agentic transformation. There are many alternatives for any specific transformation. For example:

• Provider-led: The contract could state that the provider decides whether, when, and how to invest in the transformation, bears most if not all of the costs, and obtains most if not all of the benefits. This works particularly well if the transformation changes only the provider’s ways of working without adverse effect on the company. The freedom to decide the pace of transformation allows the provider to offer pricing that assumes an optimal pace of transformation.
• Company-led: The contract could give the company an option to “own” the transformation, with set terms for sharing the costs and benefits of the transformation. This allows the company to preserve the control that it would have in an internal deployment and to know in advance what portion of the benefits it will share with the provider.
• Jointly Agreed: The transformation could occur only after signature of a project addendum describing a transformation plan and the sharing of costs and benefits from the transformation. This provides the company with control, and allows the sharing arrangements to be created when the implementation teams are aligned on plans, but limits both parties’ ability to model what will result.

We see sharing of costs and benefits as an incentive for collaboration. A party may not know what it will be able to do at signing because the implementers have not yet been made aware of the project or because the facts will evolve over time. However, the transformation might clearly require extensive collaboration. The traditional contractual response is a dependency: the company commits to provide specified resources, information, decisions, or approvals, and the provider is excused from its performance obligations to the extent the company fails to do so. A dependency remains useful, but it operates only after a failure and does no more than allocate responsibility for it. Sharing costs and benefits aligns joint incentives to reduce costs and increase benefits and provides a flexible platform for agile innovation that is superior to relying on a dependency that merely excuses performance. This may apply to costs incurred and benefits achieved by both parties.

The parties may also be able to identify material assumptions underlying the pricing, including anticipated transaction volumes, AI consumption, the expected mix of agentic and human performance, tool and model costs, and the level of human review and intervention. If so, the pricing methodology can specify which changes to those assumptions will change the charges and how the changes in charges will be calculated.

The pricing should include adjustments in the event that the benefits actually delivered over the term by advancements in agentic AI platforms are substantially larger or smaller than contemplated at contracting. In longer-term arrangements, benchmarking, continuous-improvement, productivity, and technology-refresh mechanisms may also help to account for potentially rapid changes in AI costs and achievable automation levels.

Takeaways

Agentic managed services are a new generation of managed services outsourcing. The company is buying a service delivered by AI agents and supported, monitored, and managed by provider personnel, so performance is the result of both agentic and human action. The provider still assumes ongoing responsibility for a defined business or technology function. What is new is the delivery mechanism.

The best deal structures for that new delivery mechanism will reflect the variability, uncertainty, and other characteristics of agentic AI. Instead of defining services based on what the provider’s people will do, services will be defined by what the provider will get done. Instead of reflecting human inputs, performance commitments will reflect the value of outputs and the control over agentic variability. Instead of compensating, more or less directly, for time and materials, pricing will align incentives for agentic transformation. There is work to be done, but the success in creating deal structures for earlier delivery mechanisms lights the way for this next era.