AI production-readiness audit
A fixed-price, fixed-scope audit of an LLM feature or agent — for teams whose AI works in the demo and now has to work on live data at live volume.
From £5,500
Last updated:An AI production-readiness audit is a fixed-scope assessment of an LLM feature or agent against production reality: evaluation coverage, guardrails, prompt and tool design, model choice, cost and latency, and the failure modes that only appear under production traffic. Revenant Systems delivers a severity-ranked findings report with a hardening roadmap — at a fixed price, from £5,500, agreed up front.
Who is the audit for?
The audit fits teams that have built an LLM feature or agent — in-house or with a contractor — and need it to survive contact with production: a launch approaching, output quality wobbling, costs drifting, or no clear way to tell whether a model change makes things better or worse.
- A working LLM feature or agent, pre- or post-launch
- No dedicated AI-infrastructure experience in-house
- Quality, cost, or safety questions without clear answers
Who is the audit not for?
Teams that have not built anything yet — when the feature is still an idea, the AI feature feasibility sprint is the entry point, and auditing a system that does not exist would be theatre. Nor is it a general AI-readiness review: the audit examines one real feature or agent in depth, not an organisation's appetite for AI.
What does the audit assess?
The audit assesses the six areas where LLM features fail in production: evaluation coverage (whether quality is measured at all), guardrails and failure handling, prompt and tool design, model selection with its cost and latency profile, prompt-injection and data-exposure risk, and production monitoring. Each is judged against the feature's actual traffic and stakes.
Revenant Systems builds and runs LLM features across Anthropic, OpenAI, Google, and OpenRouter models — and local models where data residency or cost demands — so the assessment reflects how these systems behave in production, not in a benchmark.
- Evaluation coverage and regression protection
- Guardrails and failure modes
- Prompt, tool, and agent design
- Model choice, cost, and latency
- Prompt-injection and data-exposure risk
- Production monitoring
What does the AI audit cost?
The audit starts from £5,500 for one LLM feature or agent. Scope and price are fixed before work starts; the fee buys the assessment and hardening roadmap, with implementation quoted separately. Ongoing and unusual engagements are scoped separately, and there is no obligation to continue into engineering.
Ian Compton leads the work, with specialist help where required. He reads your enquiry personally; the usual next step is a free 20–30 minute technical fit call to check scope, access and likely budget.
Discuss your prototypeUK company and contracting entity · UK GDPR-aware delivery · Vendor-neutral model selection · Private and self-hosted deployment options
What you receive
- A written findings report — severity-ranked, each issue with its production impact and a concrete recommendation
- A hardening roadmap — quick wins vs structural changes
- A “further investigation recommended” section for anything material outside the fixed scope
- A findings walkthrough call
How the engagement runs
- Typical turnaround: two weeks from start to report
- Fixed scope and fixed price — from £5,500, agreed before work starts
What we need from you
- Read access to the feature's code and prompts — or a guided walkthrough where access is restricted
- Sample inputs and outputs, including known failures
- Model and provider configuration, with usage or cost data where available
- A short walkthrough with the engineer who owns the feature
Access and client data are handled under our information security statement.
Frequently asked questions
Is the price fixed here too?
Yes — one scope, one price, agreed before the work begins: audits start at £5,500, and the figure is fixed once the scope is. Nothing about the engagement is metered.
What if the audit finds issues outside its scope?
Material findings beyond the agreed scope go in the report under further investigation — described, weighted, and ready to be scoped separately if you choose to pursue them.
Can you do the hardening work as well?
Yes, as separately scoped follow-on work — evals, guardrails, or a redesign where the audit calls for one. The audit stands alone regardless: the roadmap is written for your own engineers to execute.
Each package is an audit or assessment offered on its own — the process behind it is covered in how we work.
AI feature stuck at demo quality? Let's talk.
Get in touchIan Compton reads your message personally. The usual next step is a free 20–30 minute technical fit call to check the problem, scope and likely budget.