SCAIO Learn · Primer 03

AI procurement for South Carolina government.

A primer for legislators and agency staff on the most powerful policy lever the state already has.

SCAIO · scaio.org

The thesis

Procurement is policy.

South Carolina's state agencies, technical-college system, and public-university system collectively spend significant amounts on enterprise software each year. That spending is the most powerful AI policy lever the state already controls — and it does not require new legislation to use.

What South Carolina chooses to buy, from whom, on what terms, will shape the AI vendor ecosystem more than any individual bill that passes this session.

The two layers of AI procurement

What's in the contract is half the work.

Layer 1 · Contract

What's in writing

Data flows, training-data use, vendor disclosures, audit rights, exit terms, evaluation requirements, security commitments. Set at acquisition.

Layer 2 · Operations

What happens in practice

Day-to-day deployment, evaluation, human oversight, model-output review, security operations, retraining cycles. Set after acquisition.

Five questions every SC contracting officer should ask

Five contract terms that change everything.

Q1

What data does the vendor's tool collect, and where does it go?

The single most important question. Government data flowing to vendors is a different category from commercial data.

Q2

Does the vendor use our data to train models? On what terms can we opt out?

The default for most consumer AI services is yes. The default for government AI services should be no — and the contract should say so.

Q3

What evidence is there that this tool works for our use case?

Vendor claims are not evidence. Independent evaluation, comparable deployments, or a piloted result in our environment is.

Five questions (continued)

Two more contract terms that matter.

Q4

What are the documented failure modes, and what happens when the tool is wrong?

Every AI system has failure modes. The question is whether the vendor will document them honestly and what the operational response is when they occur.

Q5

What are our exit terms?

If we stop using this tool, what happens to our data? Can we get it back? In what format? On what timeline? Lock-in is the silent cost of AI procurement.

If South Carolina's agencies all asked the same five questions of every AI vendor, the SC vendor market would change in 18 months. That is the lever."

The coordination dividend

A standard SC could publish

What a shared procurement guidance document would do.

Name the disclosures. What vendors must say about training data, model evaluation, deployment limits, and known failure modes.
Define the preferences. Vendors that meet specific standards (transparent data practices, SC-based deployment partners, workforce commitments) get evaluation preference.
Provide the contracting language. A model set of clauses any SC agency can borrow rather than draft from scratch.
Set the evaluation criteria. What "working" means for the tool, how it gets measured, and who reviews the result.

Federal templates SC can borrow

SC does not need to write this from scratch.

NIST AI RMF

National risk-management framework

The default reference for evaluating AI risk. SC can adopt or adapt; most peer states already have.

OMB M-24-10

Federal-agency AI standards

The federal model for AI-use inventories, evaluation, and oversight. Provides consistency anchors SC agencies can borrow.

GSA · GenAI guide

Federal acquisition guidance

The General Services Administration's AI acquisition guidance offers usable contract templates that SC can adapt locally.

Peer-state precedents

States doing this already.

California's GenAI procurement guidance (under EO N-12-23) — the most developed state-level template.
New York City's Algorithmic Tools registry — the longest-running municipal AI-use registry.
Pennsylvania's Generative AI Governing Board — executive-branch cross-sector coordination.
Texas Workforce Commission · High Demand Job Training — workforce-aligned procurement preferences.
Maryland AI Subcabinet — agency-level AI guidance with a procurement-preferences component.

The SC opportunity

Two institutions already positioned to lead this.

Materials Management Office

The SC procurement function

Within the Department of Administration. Already manages statewide procurement standards, contracts, and vendor coordination across agencies. The institutional home for AI procurement guidance.

AI Center of Excellence

The SC AI policy function

Established under the 2024 AI Strategy. Convenes agency staff, universities, and private companies. The natural counterpart to MMO on the AI-specific content of guidance.

Five things a legislator can do

Concrete moves that do not require new law.

Request a public AI use-case inventory from your committee oversight. The data exists; consolidation is the work.
Ask agency leadership about AI procurement standards at budget hearings. The question itself moves the conversation.
Coordinate with the Center of Excellence and Materials Management Office on a shared procurement guidance document. Convening is the lever.
Champion the workforce angle. Procurement preferences for SC-based deployment partners and workforce commitments turn AI spending into economic-development policy.
Support sustained legislative-staff training on AI procurement specifically. The capacity to evaluate vendor claims is itself a public good.

SCAIO Learn

Public-interest AI research for South Carolina.

SCAIO is an independent observatory tracking AI's impact on South Carolina. Browse more primers, the flagship report, and the SCAIO Journal at scaio.org.

scaio.org · jimmy@scaio.org