We have run about 40 AI Audits in the last year and a half, in small and medium Italian companies. The structure below is the one that settled after the first 10-15. It works for us and for clients — I am releasing it as is, take the pieces you need.
The Audit is a day on site. One day only. The longer you stretch it the more it turns into consulting, and at that point you put it under contract. The Audit is the step before.
The output of a well-run Audit
You walk out with three things:
1. A map of processes that are candidates for AI automation. Typically 5-8 processes, each described in half a page: what a human does today, why it is repetitive, how much time it costs per month, which model or agent could replace it or assist it. 2. A ROI estimate for each one. Round numbers, ranges, explicit assumptions. No precise promises like "−43.7%". Conservative ranges like "10-20 hours/week saved, €15-30k/year equivalent value, payback 3-5 months". 3. A 3-month roadmap. Which process to start first, why, and with what setup. Quick wins first (things that produce value in 2-3 weeks), then the structural wins (3-6 months).
No 80-slide PowerPoint. A 12-15 page document, readable in 40 minutes, with numbers that can be challenged line by line.
The structure of the day
08:30 — Kick-off (45 min). Executive sponsor + CFO + CIO/CTO only. Briefing on objectives, constraints (budget, time-to-market, compliance), minimum acceptable ROI. Align expectations up front or you are running the Audit for nothing.
09:30 — Process mapping (3 hours). Workshop with 4-6 people from the operating teams. A big whiteboard, post-its, two questions:
- ◆"What do you do every day that you could explain to a new colleague in 10 minutes?" — these are the primary candidates.
- ◆"What do you do every week that bores you?" — these are the secondary candidates.
Stay agnostic. Do not talk about AI. Just map the work. The AI part comes later.
12:30 — Lunch with the team. Do not skip it. It is where half of the real information comes out.
14:00 — Deep-dive on the top 3 (2 hours). Take the 3 most expensive processes from the morning and go deep. How many tools are involved, what data flows, where the humans-in-the-loop sit, what happens when things go wrong, how success is measured today.
16:00 — Technical constraints (1 hour). With IT only. What we already have (CRM, ERP, line-of-business systems, cloud), what is on-prem, where we already have GDPR consent on the data, where we do not, what the client fears exposing.
17:00 — Wrap-up with executive sponsor (30 min). First takeaways, next steps. No definitive conclusions on the day itself.
The questions worth their weight in gold
Six questions to ask at least once:
1. What does an hour of the team cost on this task? Not the gross hourly rate. The fully loaded company cost: salary + contributions + space + tools + recruiting amortisation. In Italy, for a mid-level employee it is between €30 and €60/h, for a professional between €60 and €150/h.
2. What happens if the AI gets this task wrong? If the answer is "nothing, at most the client complains once" → full automation. If it is "compliance gaps, money out the door, reputational damage" → human-in-the-loop is mandatory.
3. Is there already a manual or a written procedure? If yes, the AI has raw material for the prompt. If not, before automating you have to document. The cost of documentation is a mortgage on automation.
4. Who owns the training data? It is the client's emails, the contracts they cite, signed documents. If the legal answers are fuzzy, stop earlier.
5. How many exceptions do you handle per week? If the answer is "two or three out of a hundred", good candidate. If it is "twenty out of a hundred", the pattern is too branched to start with full automation — better to assist the human work.
6. What do you hate about how you do it today? Often a pain emerges that has nothing to do with AI but that, once fixed, cuts the perceived problem by 60%. Fix it before laying AI on top.
ROI estimation without selling smoke
The trick is to declare all assumptions and stay conservative. Real example (sanitised):
> Accounting firm, 12-person team. Client email triage. > > - Volume today: ~300 emails/week (observed over a 4-week sample). > - Average time per email: 4-6 min (declared by the team, validated by sampling). > - Total weekly time: 20-30 h. > - % automatable: 60% (FAQ-like). Conservative: 50%. > - Savings: 10-15 h/week × €45/h company cost × 46 weeks = €20,700-31,000/year. > - Setup: €4-6k one-off + €150/month operating. > - Payback: 3-4 months.
Three visible numbers, assumptions in the margins. The client can challenge every number — that is the point, it is a conversation. No fake decimals.
What we do after the Audit
We send the document within 5 days. A week later we run a 60-minute call to discuss it. Often the client has reached slightly different conclusions from ours — almost always more conservative — and the first thing to do is understand why. The document is meant to be challenged.
Only after that call do we talk about a contract. Never before.
The cost of an Audit
Ours: €2,000 one-off, scaled if it leads to rollout. Half a day of prep, one day on site, two days of analysis and writing. Break-even on cost: zero — it is part of the commercial funnel, not a profit line. It works because anyone who buys a serious Audit rarely buys only that: 7 out of 10 then enter into a month of rollout with us.
If you want to run it internally, the template is this. If you want us to run it for you, you know where to find us.