Frame the decision
Define the policy or practice, who it affects, and the evidence needed to make a useful decision.
Our approach
A clear scope. A usable result.
Criminal-justice research, public-policy evaluation, and applied AI/ML engineering, connected by a clear question and a useful result.
Research & evaluation
Start with the question.
Follow the evidence.
For a policy, program, or technology, we agree a study plan: the questions, methods, measures, and deliverables. The design determines what the findings can establish.
Investigate with care
Review studies, analyze records, or gather stakeholder perspectives using methods suited to the question.
Explain the findings
Receive a findings report and decision briefing, with source references, limitations, and practical implications.
AI & data workflows
Know what you are buying.
Know when to move forward.
For AI and data workflows, each stage has a defined output and a decision point. Deployment and ongoing support are agreed separately.
Start with the fit
Choose one useful problem.
Discuss the workflow, the people using the result, and the decision it should support. Identify the current process and constraints before choosing a technical approach.
What you receive
A shared starting point. Where needed, we define a paid assessment; otherwise, we agree the project scope directly.
The decision
Do we understand the task and have a practical reason to investigate it?
Paid assessment, when needed
Make the project testable.
Inspect representative inputs, establish a baseline, and identify data gaps, access requirements, and failure cases. Agree on what is included and how the result will be judged.
What you receive
An assessment, a test plan, and a scoped proposal with deliverables, fees, and decision points.
The decision
Is the data suitable, is the expected value credible, and is a pilot justified?
Bounded pilot
Build it. Test it. Review it.
Develop the agreed workflow or analytical deliverable for a defined set of inputs and users. Compare results with the baseline and examine errors, review effort, and operating costs.
What you receive
A working deliverable, test results, documented limitations, and a handover.
The decision
Does it meet the agreed acceptance criteria, or should we revise or stop?
Separately scoped
Plan for everyday use.
If the pilot supports moving forward, define production integration, access, hosting, monitoring, and support. Agree on ownership and responsibilities before extending the system.
What you receive
A deployment and support scope, with named responsibilities and a process for changes.
The decision
Who will operate the system, review its outputs, and maintain it as inputs change?
How we evaluate AI
A pilot needs
a fair test.
We agree on measures, thresholds, and excluded cases before testing. These checks apply to a bounded pilot or a separately scoped evaluation of an existing AI workflow.
You receive the results, documented limitations, and a recommendation tied to your decision.
Test against a baseline
Compare accuracy and completeness with the current process using representative, human-reviewed examples held out from development. Check source support where outputs rely on records.
Examine consequential errors
Test missed cases and incorrect outputs, prioritizing the errors that matter most. Where the data supports it, examine performance across relevant groups and report coverage limits.
Define human oversight
Agree which outputs need approval, how uncertain or unsupported results reach a reviewer, and when processing must stop. Test the review workflow as well as the model.
Judge practical value
Measure processing time, human review effort, and operating cost at the expected volume. Use the agreed thresholds and remaining limitations to recommend wider use, revision, or stopping.
Practical responsibilities
The handover is part of the work.
A useful deliverable needs clear ownership. We agree on responsibilities before the relevant work begins. Data access and handling are agreed before records are shared.
Your team
Context, access, and decisions.
Name a project owner, provide authorized access to representative data, and identify the people who review outputs and accept the result. Your team remains responsible for the decisions the work informs.
AREAS
Delivery, testing, and handover.
Build the agreed deliverables, report test results and limitations, and provide the documentation and handover included in scope. Acceptance testing of our own work is part of delivery.
Agreed together
Ownership, hosting, and changes.
Define rights to code and outputs, third-party dependencies, data handling, the hosting owner, and access arrangements. Specify monitoring, maintenance, support, and how new requirements are priced and approved.
An evaluation of an existing system can be a separate engagement. We make our role and any implementation relationship explicit in the evaluation scope.
The method behind delivery
Research. Analyze. Inform.
Data-driven research
Research
Understand the question, examine the evidence, and define what a useful result would establish.
AI & ML Engineering
Analyze
Connect the data, build the appropriate workflow, and test performance, uncertainty, and failure cases.
Decision support
Inform
Explain what the results support, document the limits, and equip your team to decide what comes next.