Trust Audit / 3–4 weeks / Fixed scope
One workflow.
All five layers.
Know where your AI fails, why it fails, and what to fix first, assessed across data, model, action, runtime, and accountability. Leaders leave with a score they can defend. Engineers leave with a plan they can execute.
What you take away
Five deliverables.
One defensible answer.
Findings are reviewed with your team and connected to the decisions you need to make about your system. The last one is the artifact the room in security review has been waiting for.
AI system map
The models, tools, data sources, and decision points that shape your workflow, including where every prompt actually lands.
Layer-by-layer findings
Where you stand on each of the five layers: data, model, action, runtime, accountability. Evidence you hold, and evidence you are missing.
Failure analysis
Failed decisions traced to their causes in retrieval, prompts, tool use, and policy logic.
Risk & impact assessment
Technical failures connected to operational consequences, so remediation starts where it actually matters.
The Trust Score
One score across the whole stack, measured end to end, dated and signed by a named human. The artifact you take to the board.
An illustrative four-week engagement
A clear cadence.
Concrete outputs.
We define the boundaries upfront, then work through discovery, evaluation, analysis, and recommendations.
Map the system
Trace the models, tools, data sources, policy checks, and integration boundaries with your team, including where every prompt actually lands.
Define the evaluation
Agree acceptance criteria for the workflow and the evidence standard for each layer. Build evaluation scenarios from available workflow evidence and synthetic edge cases.
Evaluate & investigate
Run the scenarios, analyze decision chains, and connect failures to technical causes and business consequences.
Score & prioritize
Walk the findings through with engineers and leaders. Issue the Trust Score and a prioritized roadmap for remediation.
Before we begin
Know what
to expect.
Scope, access, and responsibilities should be clear before an engagement starts.
Is my system ready for a Trust Audit?
The audit is designed for AI systems in production or advanced pilot: copilots, agents, and decision-automation workflows. Start with a workflow where inconsistent outputs, opaque decisions, or uncertainty about scaling are creating a concrete concern.
What exactly is the Trust Score?
A single assessment of one workflow across all five layers (data, model, action, runtime, and accountability) measured end to end, scored, dated, and signed by a named engineer. It is designed to be handed to somebody who does not work for you: a board, a regulator, or the security review that has your program sitting in a queue.
Do you assess all five layers, or just the model?
All five. A model-only assessment is the most common way an audit misses the failure that actually matters: outdated context, an unbounded tool call, or an action nobody logged. Depth varies by layer according to what your workflow exposes; we agree that during scoping.
What do you need from our team?
During scoping, we identify a workflow owner and technical counterpart, review the available architecture and workflow evidence, and agree the access and sample data needed for evaluation. The exact level of participation depends on your system.
How is sensitive data handled?
Data access, evaluation environment, handling requirements, and permitted use are agreed during scoping. Use the initial conversation to describe the workflow and its constraints; you do not need to share production data to discuss fit.
Does the audit include implementing every fix?
The audit delivers diagnosis, a score, and a prioritized remediation roadmap. Broader engineering implementation and ongoing evaluation can be scoped as a follow-on engagement. Your team can also use the roadmap to guide its own work.
How long does it take, and what does it cost?
The typical audit runs for three to four weeks. The sequence below illustrates a four-week engagement. Timing and fees are confirmed against the workflow, access requirements, and evaluation scope.
Start with one workflow
How many of the five layers
can you evidence today?
Thirty minutes to discuss your system, the layer you are least confident about, and whether a Trust Audit is the right next step.