When your office considers an artificial intelligence tool, the conversation may begin with what it can do, how much it costs and how quickly it can be acquired.
There is another question worth bringing into that conversation:
What will it take to make this capability work for the people responsible for the mission?
That question matters whether you manage a program, help define requirements, support acquisition or will use the technology every day.
GSA’s “Buy AI” resources identify multiple pathways to acquire AI, including OneGov agreements, USAi and governmentwide contracting vehicles. In July 2026, GSA announced a OneGov offering with discounts of up to 80% on selected AI capabilities. These efforts are making access more practical for federal agencies.
But access is only one part of the decision.
In its April 2026 report, the Government Accountability Office examined 13 AI acquisitions across the Departments of Defense, Homeland Security and Veterans Affairs, and GSA. Its findings highlighted challenges involving technical expertise, requirements, data rights, testing, acquisition approaches and costs.
For employees helping their agencies navigate these decisions, the opportunity is not simply to identify an available tool. It is to help connect the purchase to the work that needs to improve and the conditions under which that improvement will actually happen.
The Acquisition Vehicle Is Not the Acquisition Strategy
An available contract vehicle can help answer how your agency acquires a capability. Your team still needs a clear answer to what the capability should accomplish.
Consider the difference between these two starting points:
“We need an enterprise generative AI platform.”
“We need acquisition specialists to spend less time organizing and analyzing market research while maintaining human review of source material and final recommendations.”
The first identifies a technology category. The second describes a work requirement that a team can evaluate.
With the second, the conversation becomes more useful. How much time does the task currently take? Where do errors or delays occur? Which source materials are needed? What must remain subject to employee judgment?
OMB’s AI acquisition memorandum, M-25-22, strongly encourages performance-based acquisition techniques that connect requirements to desired outcomes rather than unnecessarily restrictive specifications.
For employees contributing to requirements development, that means bringing the details of the work into the discussion.
For example, instead of asking whether a tool can summarize documents, describe what makes a summary usable: the information it must retain, the sources a reviewer must be able to check and the omissions that would make the output unacceptable.
The most useful requirement is not that the tool produces an answer. It is that the answer helps someone complete a specific task to an acceptable standard.
The Sticker Price Is Not the Cost of the Capability
An attractive introductory price deserves attention. It should also prompt a second conversation about what happens after the pilot.
Depending on the solution, costs may include licensing, usage, cloud infrastructure, integrations, implementation support and ongoing testing. GSA’s buying guidance specifically cautions that usage costs can grow quickly without monitoring and management controls.
GAO documented an Army proposal that included AI licensing fees of approximately $300,000 per vehicle per year, more than $500 million annually across the planned fleet. Army officials rejected the proposal and began exploring other pricing approaches.
Your office may be considering a very different application. The useful question is still:
What happens to the budget if this capability is successful?
A limited pilot does not, by itself, establish what a broader deployment will cost. Before expanding, ask the team to model realistic usage: more employees, additional workflows, larger volumes of information and the support needed to keep the system operating.
Include government staff time in that assessment. If employees must review every output, correct recurring errors or maintain the information the system uses, that effort belongs in the operating picture.
For program staff, a useful contribution is a realistic estimate of how the capability would be used, not just how many people could receive access. For supervisors, it is identifying the training, review and support responsibilities the team would need to absorb.
The objective is not to predict every future expense. It is to avoid treating the price of access as the full cost of delivering the capability.
Protect the Ability to Change Course
Imagine your office adopts an AI tool to search authoritative documents, organize incoming information and prepare material for review.
Over time, the team improves its instructions, organizes information around the system and connects it to other tools. Employees develop processes that depend on it.
Now imagine that the vendor changes its pricing, the underlying model changes or another solution becomes a better fit.
What would your team need to preserve?
The answer may include more than the original documents. It could include configurations, workflow instructions, evaluation materials and the knowledge developed during implementation.
OMB’s acquisition guidance addresses vendor lock-in, data and model portability, and intellectual-property rights. Those provisions have a practical purpose: preserving options when the agency’s needs or the technology change.
Employees do not need to negotiate contract terms themselves to help surface these needs. They can identify which information, processes and work products would be difficult to lose or recreate. Acquisition, legal and technical teams can then determine the appropriate contractual and technical protections.
The question is not simply whether the government owns every component.
It is whether the agency has the rights, access and documentation needed to continue the work when something changes.
Decide How You Will Evaluate AI Before You Depend on It
A successful demonstration should be the beginning of an evaluation conversation, not the end.
OMB’s guidance calls for demonstrations that reflect the intended operating environment where practicable. It also requires applicable contract terms to provide for ongoing monitoring and evaluation of AI performance, risks and effectiveness.
For the team using the capability, the practical questions are straightforward: What does acceptable performance look like? Who will review it? How will problems be reported? What would cause the team to pause or change its use?
Consider a tool intended to help prepare market-research summaries. A useful evaluation could examine whether it preserves important qualifications, accurately represents source material and reduces the total time required to produce a reviewed work product.
That last point matters. A faster first draft is not necessarily a faster completed task.
If a draft takes two minutes to generate but requires extensive checking and correction, the evaluation should capture that effort. Otherwise, the team risks measuring output speed rather than improvement in the work.
Employees familiar with the task can help identify realistic test cases and failure conditions. Those examples should include difficult or ambiguous situations, not only the cleanest documents or simplest requests.
Before deployment, agree on how that feedback will reach the people responsible for performance oversight.
Success should mean the capability continues to produce an acceptable mission outcome at an acceptable cost and level of risk…not simply that it was deployed.
Make the Next Acquisition Easier by Capturing What You Learn
GAO also found that the four agencies in its in-depth review were not systematically collecting lessons from AI acquisitions because their departmental policies did not require it. All four agreed with recommendations to establish processes for capturing and sharing those lessons.
For an individual team, the practical takeaway is to document the decisions and discoveries that another team would benefit from knowing.
Which requirements proved essential? Which assumptions about usage were wrong? What did the team underestimate? Which tests revealed meaningful differences between products? What did employees need that the initial demonstration did not show?
The useful record is not just whether the pilot succeeded. It is what the agency learned about making the capability work — and under what conditions it did not.
A team that decides not to expand a pilot should still be able to explain why. That decision can be valuable evidence for the next program considering a similar purchase.
Capture those lessons through the agency’s established processes while the people involved still remember the details. The goal is to make useful knowledge available beyond the individuals who happened to participate.
Five Questions to Bring to the Next AI Discussion
You do not need to own the entire acquisition process to help improve the decision. Before your office moves forward, bring these questions to the appropriate program, acquisition and technical teams:
- What specific mission outcome are we trying to improve? Define the task and the standard the capability must meet.
- What will it cost if adoption grows? Include usage, support, infrastructure and the employee effort required to operate it.
- What will the government retain? Clarify the rights and access needed for data, configurations, documentation and implementation work products.
- How will we know it is still performing well enough to use? Establish measures, review responsibilities and a process for addressing problems.
- What happens if we need to change models, vendors or approaches? Understand what a transition would require before the capability becomes difficult to replace.
These are not questions designed to stop a purchase. They are questions that help make a purchase worth making.
Easier Buying Should Lead to More Useful Capabilities
The goal is not to turn every federal employee into an AI acquisition specialist.
It is to make sure the people who understand the mission, the information and the day-to-day work help shape the decision.
A program manager can clarify the outcome. An employee who performs the task can identify what a demonstration overlooks. A technical specialist can surface integration dependencies. An acquisition professional can help translate those needs into an appropriate buying approach.
An award can establish access to technology. The work that surrounds it determines whether that access becomes a useful, sustainable capability.
Before your agency’s next AI purchase, ask not only what the technology can do today, but what your team will need to operate it, evaluate it and change course when necessary.
Raitchele Arnell is the CMO of ArtForm Business Solutions, a women-owned digital agency supporting government contractors, enterprise organizations, and public-sector initiatives. She specializes in AI adoption, market intelligence, strategic communications, and modernization strategies for highly regulated and mission-driven industries. Her work spans cybersecurity, healthcare modernization, emergency communications, critical infrastructure, and federal technology initiatives. Known for helping organizations rethink how marketing supports mission outcomes, Raitchele focuses on the intersection of AI, operational efficiency, audience intelligence, and trust-driven communications.
Photo Credit: Joshua Woroniecki, Pexels
Leave a Reply
You must be logged in to post a comment.