OutcomeA reproducible prototype on public or synthetic data, an evaluation record against agreed cases, and a decision memo on whether to proceed.

What this covers

Document search and summarization for case workers, classification and routing of incoming requests, first-draft generation for routine correspondence, and review support that flags what a person should look at. Each function is scoped to a named user and a named decision.

Guardrails that are part of the scope

Public or synthetic data by default. Production or sensitive data requires a buyer-approved data, hosting, security, and privacy path before use.
A human decision stays visible. The function recommends; a person decides, and the record shows who did.
Evaluation before expansion. The prototype is measured against agreed cases and the results are delivered, including the misses.

Where BCA fits

A bounded assessment and prototype is a natural micro-purchase or simplified acquisition. Inside a larger program, it is a discrete work package that lets a prime or agency test an AI function without committing the platform.

Outcomes

What changes once the work is accepted.

The agency learns whether an AI function is worth pursuing before committing a platform or budget.
The human decision stays visible in the record, with the function recommending and a person deciding.
Evaluation results, including misses, are documented against agreed cases.
Deployment gaps, risks, and data requirements are written down for the go/no-go decision.

Typical deliverables

Named the way they appear in a scope.

Use-case framing and risk boundary: The named user, the decision supported, the data boundary, and the failure-handling rules.
Evaluation criteria and test cases: Agreed before the build, so results are measured rather than demonstrated.
Reproducible prototype: Built on public or synthetic data and runnable again by the agency.
Evaluation record: Observed results against every agreed case, including failures.
Deployment-gap analysis and go/no-go memo: What production would require in data, hosting, security, privacy, and oversight.

How engagements work

Three common ways this capability is bought.

Each starts with a written scope, deliverables, and acceptance evidence. Thresholds and authorities are on the federal agencies page.

Micro-purchase or simplified acquisition
A bounded assessment and prototype for one named function on synthetic data, with an evaluation record and a go/no-go memo, scoped to fit under the threshold.
How micro-purchase works with BCA
Subcontract work package
A discrete AI prototype or evaluation package inside a prime's program, letting the prime and agency test a function without committing the platform.
The six-part work-package model
Sources sought and RFI response
BCA responds with a framed use case, the evaluation approach, the data it would and would not use, and the human-review design.
What BCA sends back

Capability limits

What BCA will not claim or perform here.

Stated before award so the file, the compliance matrix, and the work all match.

Prototypes use public or synthetic data by default. Production or sensitive data requires a buyer-approved data, hosting, security, and privacy path.
BCA does not operate production AI systems or hold an authorization to do so.
AI functions support a human decision. BCA does not build functions that make final determinations about members of the public.

Company-wide boundaries

No federal contract past performance yet. Commercial delivery and owned-system evidence are labeled on the past performance page.
No personnel or facility security clearance. Unclassified work only.
No FedRAMP authorization, agency ATO, CMMC certification, ISO certification, or SOC 2 report. BCA works inside the boundary the agency or prime authorizes.
Not a GSA Multiple Award Schedule holder.

Source and verification dates for every boundary statement are kept with the evidence records and the capability statement.

Other capabilities

For Arizona firms: the commercial version

Next step

Send the requirement. We reply with a bounded scope and evidence, not a pitch.

A sources-sought notice, a section of a statement of work, or a one-paragraph problem description is enough to start. Expect a written response within two business days.

Contact Teton Wilson Capability statement (PDF)

teton@bechosenagency.com · 602-533-0553

AI-enabled functions & bounded prototypes