Illustrative scenarioWho is accountable for this AI decision?
Accountability in practice
- Policy
- Which rules applied?
- Authority
- Who could decide?
- Evidence
- What supported it?
AI programmes · From pilot to production
Take AI beyond
the pilot.
Move towards production with a clear account of how AI decisions are made, who is responsible and what evidence supports them.
A basis for accountability
Be clear about how AI decisions are made.
Approved policy
Know which rules governed the decision.
Clear responsibility
Know who has authority and when a person must intervene.
Evidence to inspect
Give the people accountable a basis they can return to.
Accountability in practice
Make AI decisions accountable.
Evaluate each proposed decision deterministically against the policy in force at that moment. Return a verdict your application can enforce, with the evidence to explain it.
A proposed AI decision goes to MeshQu for deterministic evaluation against the policy in force. The verdict returns to your application to enforce before action when integrated into the decision path. A signed Decision Receipt records the assessment. For example: Allow, Review, Deny.
Proposed AI decisionEvaluate against policyYour applicationApplies the returned verdict
Verdict + signed Decision Receipt
For example
When your policy calls for a person
Human review or escalationEvaluate at decision time.
MeshQu applies deterministic rules against the policy version in force, returning a verdict and a signed record of the assessment.
Fit your workflow.
Use the verdict to control the next action, connect human review when needed, or run assessment and recording alongside your existing workflow.
An example in practice
AI recommends approval. Who is accountable?
AI recommends approval using declared income. In this example, the lender’s policy requires verified income and an authorised underwriter’s judgement.
AI recommends approval. Income verification is missing.
Required reviewerHuman review requiredAwaiting authorised judgement
MeshQu flags the evidence gap.
Income verification required by policy is missing
A person owns the next decision.
Authorised mortgage underwriter
Open pending authorised judgement
The recorded basis
The policy, finding and required authority stay connected. The reviewer’s judgement is recorded alongside when made.
Inspect the record and responsibilities
- Policy
- Mortgage lending policy · fictional ratified version 2.1
- Application
- £320,000 mortgage application
- AI recommendation
- Approve, based on declared income
- MeshQu policy assessment
- Income verification required by policy is missing
- Policy authority
- Policy governance body
- Operational authority
- Authorised mortgage underwriter
- Result
- Open pending authorised judgement
Scope and limits
- This is a fictional lender policy and application, not a lending recommendation, measured performance or a customer result.
- The record makes actions inspectable; it does not itself execute or re-decide them.
- Application-controlled scope depends on the organisation’s ratified policy and implementation.
Who does what?
Responsibility boundaries
Keep each role explicit.
Responsibility boundaries
Keep each role explicit.
Policy ratifier
Approves the policy in force
Ratifies the policy version and its governance boundaries before operational use.
AI or application
Performs bounded checks
Applies the policy and surfaces evidence gaps within its granted scope.
Mortgage underwriter
Owns the lending judgement
Reviews the AI recommendation and supporting evidence within their authority. Later assurance review does not inherit that authority.
What supports the record?
Evidence and limits
Inspect the basis without overstating the outcome.
Evidence and limits
Inspect the basis without overstating the outcome.
Policy
Exact basis
Link the decision to the version and requirements applied.
Evidence
Referenced inputs
Show the information used, missing or contested at the time.
Limits
Bounded claim
The examples are illustrative. Availability, integrations and measured results require separate validation.
Published explanation
Decision Receipts: what the record contains
A Decision Receipt is a signed record of a policy assessment. It connects the result to the policy version and information used. With the necessary verification material and trusted keys, a reviewer can check the signed record and replay the assessment. It does not prove that the inputs were true, that the decision was correct or that the action was carried out.
Read the source explanationThis explains the record structure; it does not verify this illustrative workflow.
Board visibility
Give your board a clearer view of AI decisions.
Show how decisions are governed, who is accountable and the evidence behind them.
Start with one workflow
What’s holding your AI programme back?
We’ll look at one decision workflow, the accountability or evidence gap, and where MeshQu could fit.
One or two sentences is enough. You don’t need a finished brief.
Prefer the technical detail? Explore the docs.
Before we talk
Frequently asked questions.
Can we start with our existing policy?
That is the starting point: one policy and one decision workflow. We’ll explore how its requirements, responsibilities and evidence fit together.
Can we build our own solution with MeshQu?
Yes. MeshQu is infrastructure your team can build on. A commercial agreement with MeshQu lets you implement it around your own workflows and user experience. We’ll scope the integration around how you intend to use it.
How would MeshQu fit our current systems?
We’ll map where the decision happens and what needs to connect. The integration approach is scoped around your workflow.
What happens when we get in touch?
We start with your AI programme, the decision workflow and what needs to be clear before production. Together, we’ll explore a useful first step.
One workflow, a challenge or a question is enough.