How States Will Actually Buy AI: Cooperative Vehicles and Modular Contracts

Why an AI purchase needs room to change

Many familiar purchases can be specified with relatively stable quantities and acceptance criteria. An AI-enabled service adds moving parts: model versions, consumption-based charges, data use, and performance that depends on the task and operating environment. A multiyear commitment needs a way to handle those changes without reopening every architectural decision.

The practical opportunity is to combine two established procurement approaches. Cooperative purchasing addresses repeated sourcing and negotiation effort. Modular contracting addresses the risk of committing the entire service to one indivisible delivery. Our analysis is that they work best together when an agency retains enough internal expertise to make acceptance decisions and manage the boundaries between components.

Cooperative purchasing shares work, not accountability

NASPO describes ValuePoint as cooperative contracting through competitive solicitations led by multistate sourcing teams. Its supplier guidance identifies participating addenda or chief procurement officer approval as mechanisms for purchasing. That is evidence of a procurement structure, not evidence that every AI offering on a cooperative vehicle is approved for every jurisdiction or use.

A shared evaluation can spare small jurisdictions from repeating parts of a complex review. It cannot establish that a model is appropriate for a local benefits decision, that the proposed data use is lawful, or that a particular integration meets the agency's security requirements. The buying agency still needs to establish what the shared review covered and where local work begins.

Before relying on a cooperative vehicle, ask:

• Authority: Is the jurisdiction eligible to use it, and what approvals or participating terms are required?
• Scope: Does the competed service actually cover the proposed AI capability and implementation work?
• Evidence: Which security, privacy, and performance claims were examined, and when?
• Local fit: Which decisions, data, users, and integrations were outside that examination?
• Ownership: Who is responsible for accepting the deployed system and monitoring it after purchase?

The value is shared capacity. A vehicle becomes less useful when its availability is mistaken for a complete suitability decision.

Modular contracting keeps the next decision open

NASPO's modular procurement primer describes breaking complex procurements into smaller, interoperable increments and discusses both benefits and implementation difficulties. Applied to AI, the objective is to make a meaningful component replaceable without forcing replacement of the entire public service.

Shorter phases alone do not create modularity. Several small contracts can still produce a tightly coupled system if they share undocumented interfaces or depend on proprietary formats. Conversely, a long-running service can retain flexibility if the buyer has enforceable rights, clear boundaries, and tested transition arrangements.

For an AI-enabled workflow, consider separating data access, model inference, business rules, human review, and outcome reporting where technically appropriate. These are design options, not a prescription to split every system into five procurements. Someone must own integration, incident response, and end-to-end performance across the resulting contracts.

Put portability into acceptance criteria

Terms and conditions determine whether an agency has a practical switching option. Price matters, but a low subscription rate is incomplete if migration requires rebuilding the service.

Our recommended negotiation checklist is:

• Data and artifacts: Define ownership, permitted use, export formats, retention, and deletion for agency data, prompts, evaluation sets, configuration, and integration code. Do not assume a state can acquire ownership of a vendor's underlying model or third-party intellectual property.
• Model changes: Require notice of material changes and deprecations, the ability to test before migration where feasible, and clear remedies when performance changes.
• Audit evidence: Specify access to records and performance evidence sufficient for the agency's oversight needs, with appropriate protections for sensitive information.
• Cost controls: Define usage units, monitoring, budget alerts, and authorized limits rather than relying on a headline unit price.
• Exit assistance: Establish transition deliverables, support periods, charges, and responsibilities before the relationship ends.

A portability clause should be exercised. If no one has demonstrated an export and a usable handoff, the contract describes an intention rather than an operational capability.

A practical sequence for the next purchase

1. Choose a bounded service problem. State what a successful outcome looks like for staff and residents, including acceptable failure handling.
2. Check the buying route. Have procurement and counsel confirm authority, competition scope, and any local conditions for a cooperative purchase.
3. Test on representative work. Ask technical and program staff to evaluate accuracy, workflow impact, accessibility, security, and total operating cost.
4. Buy a measurable increment. Set acceptance criteria and decide in advance what evidence permits expansion, correction, or termination.
5. Rehearse change. Demonstrate a version update, a service interruption, and an export before committing to a larger footprint.

Procurement officers, technologists, and program owners each hold part of the decision. Vendors can make responsible purchasing easier by offering clear contract terms, usable evidence, and realistic transition support. The durable advantage is a service the agency can understand, govern, and change as its needs evolve.

Sources and further reading

• NASPO: supplier and cooperative-contracting overview.
• NASPO: Modular Procurement—A Primer, updated September 2023.

This is where Spartan X’s combination of AI architecture and technical program coordination can be useful: translating a buyer’s need for flexibility into interfaces, acceptance evidence, and a delivery plan that can survive a change of vendor.