Pilot
One workflow is live in your environment by week 10, with measured results.
What this engagement addresses
- The policy is adopted and the use case is obvious, and there the program has stalled, because nobody on staff builds software.
- A vendor demonstration cannot answer the question that matters: whether this works on your caseload, your records, and your staff's working week.
- A public failure would cost the program more than a year of quiet progress earns it, so whatever ships has to be reviewable and explainable.
- Pilots have a reputation for running forever. Yours needs a defined end and a decision.
Scope, price, and deliverables
We build one workflow under your adopted AI policy and run it with your staff. Acceptance criteria are agreed in writing before work begins, and at week 10 your agency decides whether to continue.
A written scope fixes the price before your agency commits to anything. After that, the price changes only through the published change process.
- Starting price
- From $35,000
- Timeline
- 10 weeks
- What moves it
- Whether the content it works from is already clean, and how much evaluation you need before it goes live.
What your team uses
- The working system, deployed in your environment, with human review at every point that touches a case
- Source code, prompts, and evaluation data, delivered to your repository
- An operations runbook for your team
What documents the result
- Measured results against the agreed acceptance criteria
What we need from you
- A process owner and one subject expert, weekly
- A test environment, or ours until yours is ready
- A decision at week 10 on whether to continue. Stopping is a normal outcome.
- Third-party services
- Most pilots need one or two third-party services, for example email, messaging, or identity. We start with the services your agency already runs.
Your agency owns everything we deliver: the deliverables, the source code, and the data. What happens after a build and how to buy this work are published alongside the full schedule.
What changes for your agency
Measured results by week 10
You get results against acceptance criteria your agency wrote, which settles the internal argument about whether the approach works here.
Adoption starts on day one
Your staff run the pilot with us during the supervised weeks, so go-live is never the first time they see it.
Stopping is a designed outcome
The engagement ends at a decision gate. Continue, expand, or stop, and stopping costs nothing further.
Nothing is locked in
Source, prompts, evaluation data, and the runbook sit in your repository. Your own team, or any firm, can take it from week 10.
The standards the work is held to
These are written into the engagement agreement, and each one can be checked during the work rather than taken on trust. The full posture is published under security and compliance.
- A person reviews every case decision
- Human review points are designed in wherever the work touches a case, documented one by one, and testable.
- Built inside your adopted policy
- The pilot follows the rules your agency adopted, including disclosure to the public where the tool is public-facing.
- Audit logging
- The system records what it did and what a person decided, in a form your IT team can retain and review.
- Accessibility
- Public-facing interfaces meet WCAG 2.2 AA and are tested with assistive technology.
- Criminal justice information
- Where the work touches criminal justice information (CJI), the relevant security addendum and personnel screening are completed before access is granted.
Common questions
What can a pilot cover?
One workflow with a clear owner and a measurable output: self-help answers, document processing, a transcript workflow, guided forms. The first conversation is about choosing it, and the written scope fixes it.
Do we need an AI policy first?
Yes, in some form, because the pilot is built to your adopted rules. If you do not have one yet, the Readiness and Modernization Roadmap produces it, and the two engagements are often bought in sequence.
What happens at week 10 if the results are poor?
You stop, and you keep everything: the evaluation data, the source, and a written account of why it missed. That is a far cheaper way to learn it than a procurement.
Which AI model does it use?
The one that fits your policy, your data handling requirements, and the task. The choice is documented and changeable without a rebuild, and agency data does not enter a third-party model without a written agreement covering it.
Who hosts the pilot?
Your environment by default, or ours until yours is ready. After week 10 the hosting choice is yours, and self-hosting is always available.
Get a written scope and a price
Send us your name and agency. We reply within one business day with a short set of clarifying questions, followed by a written scope and a fixed price for your agency. There is no obligation, and the scope document is yours to keep whether or not you proceed.
Prefer to talk first? Start a conversation