Does it do the job?
Test agreed tasks against representative cases and known answers. Measure mistakes, missed cases and uncertainty, including the situations where the system should stop or ask for help.
Technical review · Christchurch, New Zealand
AI system audits grounded in evidence.
Understand what your AI system does, where it fails and what needs to change. We review agreed workflows, test their behaviour and document findings you can act on.
A second technical opinion
Before you rely on it. When it stops earning your trust.
You have an AI implementation, but people still check every answer. An agent updates business systems without a clear record. A supplier has delivered a demo and you need to understand what remains before deployment.
A scoped review turns those concerns into questions that can be tested. Human engineering judgement defines the assessment and interprets the evidence; automated checks make repeatable behaviours easier to verify.
Are actions controlled?
Examine what the system can read and change, how approvals work, and what happens when inputs are misleading, incomplete or outside the intended workflow.
Do the integrations hold?
Follow actions through connected systems. Check failed requests, retries, duplicate actions, handoffs and whether the final state matches what the user was told.
Can you operate and verify it?
Review logs, change controls and recovery paths. Measure latency and cost within the agreed test scope, and identify the evidence an operator needs to investigate a failure.
A report you can use to make decisions.
01
Agree the review
Identify the workflow, system version, decisions affected, access and acceptance criteria. State what will and will not be tested.
02
Investigate and document
Record reproducible findings where access permits, their business impact and supporting evidence. Show coverage gaps and unresolved questions alongside results.
03
Prioritise the next steps
Receive a remediation plan and review session. Choose which fixes to commission and whether to re-test, with the scope of each next step agreed separately.
Passing a test establishes evidence for that test and its conditions. It does not establish that every requirement is complete or that every future output will be correct. The report makes that boundary explicit.
Owned demonstration · reproducible evidence
See a finding, its evidence and its correction.
Inspect three deliberately introduced failures around approval, retries and reported success. The specimen includes observed before-and-after results, scope limits and the source needed to reproduce them.
Read the control review specimenQuestions before a review
What is an AI system audit?
It is a scoped technical review of an AI system against agreed requirements and test cases. AcesForge examines behaviour, integrations and operational controls, then documents findings, evidence, limitations and recommended fixes.
Can you review a system built by another supplier?
Yes, subject to agreed access and scope. We can review an existing chatbot, agent workflow, voice integration, retrieval system or classifier. A black-box review can assess observable behaviour; source and configuration access allow a deeper investigation.
What access do you need?
We start with the intended workflow, known issues and available documentation. Depending on scope, we may need an isolated test environment, representative data, logs, configuration or source access. Access and data handling are agreed before testing; do not send sensitive data in the initial enquiry.
Does an audit certify that the system is safe or compliant?
No. This is a technical assessment with stated coverage and limitations, not certification, accreditation, a legal opinion or a guarantee of correctness. Specialist legal, regulatory or penetration-testing work is separately scoped with an appropriate provider where needed.
Can you implement the recommended fixes?
Yes. Remediation can be scoped after the findings are reviewed, followed by a re-test of the agreed changes. We disclose our involvement in the original implementation; reviewing our own work is technical verification, not an independent audit.
How much does a review cost?
The price depends on the system, access and depth of testing. We agree the scope, deliverables and price in writing before work begins. An initial conversation establishes whether a review is the right next step.
Tell us what you need confidence in.
Start with the system, its intended job and the behaviour that concerns you. We will discuss fit and define a review before requesting technical access.
Discuss your systemBuilding something new? Explore AI consulting and implementation or computer vision engineering.