AI audit & architecture scoping

Who the AI audit engagement is for

• You've identified two to five AI use cases but don't know where to start
• Your executive committee wants a budgeted AI vision before investing
• You want to avoid the prototype-that-sleeps-on-the-shelf trap

When AI audit is not the right engagement

• You're looking for a marketing pitch on generative AI in general
• You want an immediate POC without scoping — see the RAG POC offer

How the AI audit engagement runs

1. Day 1 — Discovery — Interviews with executives, business owners and technical teams. Collection of data, security, budget and timeline constraints.
2. Days 2 and 3 — Technical audit — Mapping of your infrastructure, data governance and security. Identification of blocking points.
3. Day 4 — Prioritisation — ROI and feasibility scoring workshop with your team. Selection of one to three use cases to industrialise.
4. Day 5 — Roadmap and readout — 90-day roadmap, budgeting, architecture recommendations, readout to the executive committee. This programme describes the full five-day format; a narrower scope fits into three days.

What AI audit delivers

• Audit of your existing infrastructure, data and security (written report)
• Use-case prioritisation matrix (ROI and feasibility)
• 90-day technical roadmap with milestones and budgets
• Architecture recommendations (LLM, vector database, orchestration)
• Initial AI Act compliance assessment
• Strategic readout with your executive committee

AI audit: what the engagement expects from you

• A sponsor who can open doors: access to business teams, technical teams and whoever holds the budget
• Read access to the existing infrastructure and documentation, however incomplete — the audit starts from what exists, not from a questionnaire
• Two hours of executive committee time for the readout: a roadmap that is not arbitrated on the day it is handed over never will be

What you gain

• Fast decision — Under a week instead of three months of uncertainty. You know what to do, in what order, at what budget.
• Zero complacency — If a use case is not profitable, SeedVision says so. You avoid the useless prototype.
• Compliance built in — AI Act, GDPR, sovereignty: SeedVision flags risks during scoping, not after.

Questions about AI audit

What's the concrete deliverable of the audit?

A written report of 40 to 60 pages covering the infrastructure, data and security audit, the prioritisation matrix, a budgeted 90-day roadmap, architecture recommendations and an AI Act assessment. Plus a 90-minute oral readout for your executive committee.

Can the audit lead into a POC or industrialisation?

Yes — it's the recommended path. The audit roadmap sets the foundation; the POC validates the priority use case; industrialisation takes it to production. Three successive packages, controlled timeline.

Is the audit billed even if SeedVision recommends not continuing?

Yes. SeedVision is paid for technical truth. If no use case reaches the expected ROI, you leave with a report that justifies stopping — an avoided cost, not wasted spend.

Do we need clean data before starting the audit?

No. The actual state of your data is part of what the audit measures: sources, quality, access rights, duplicates. Finding out that a use case first requires a clean-up is a useful conclusion, not a failed audit.

Who should take part in the interviews?

The people who will live with the outcome: the business teams concerned, operations, security, and the executives who arbitrate the budget. An audit run with the IT department alone produces a roadmap the business will not follow.

Related reading on the blog

• LiteLLM on Bedrock: The Gateway That Budgets Your AI Agents
• AI Spend Quotas in Real Time: Jamf's IAM Pattern for Bedrock
• AgentCore Evaluations: Grading AI Agents Across Any Framework
• deepsec: An Open-Source AI Agent for SAST, by Vercel
• Mistral Medium 3.5 + Remote Agents: A New Open-Weight MLOps Stack
• Sleeper Agents: A $1,000 Backdoor That Benchmarks Miss
• EU AI Act on August 2: The MLOps Stack for Production Compliance

Talk about your project

Considering the AI audit engagement (3 to 5 days) — write to contact@seedvision.fr — reply within 24 hours.