An AI audit is a structured diagnosis of a company's operations that identifies where artificial intelligence delivers a measurable return on investment. The topic is no longer theoretical: according to MIT's study "The GenAI Divide: State of AI in Business" (2025), 95% of enterprise generative AI pilots produce no measurable return. The main cause is not the technology, but the choice of use cases.
This guide describes the method Tandem applies on its AI audit engagements with SMEs, mid-caps and investment funds: the steps, the real cost, the deliverables to demand, the mistakes to avoid, and the French public scheme that funds 40% of the process.
What exactly is an AI audit?
An AI audit maps a company's processes, identifies tasks that AI can automate or augment, quantifies the expected gain of each use case, then ranks them into a roadmap. The final deliverable answers three questions: where to start, how much it costs, how much it returns.
An AI audit differs from classic transformation consulting through its short horizon and its obligation of concrete results: a few weeks of work and a prioritised list of directly actionable use cases, not an 80-page report that ends up in a drawer.
Why audit before investing in AI
The cost of a wrong use-case choice far exceeds the cost of an audit. An AI agent deployed on a low-volume or poorly framed process mobilises teams for months with zero gain. Conversely, the right use cases pay back fast: on Tandem's deployments within fund portfolio companies, payback is under 6 months, sometimes under 2.
This matches what we detailed in our analysis of why 80% of companies see no ROI on their AI: the companies that succeed are not the ones testing the most tools, but the ones that pick their battles before spending.
The 3 steps of an AI audit method
Across our 40 engagements, Tandem's AI audit method follows three steps over a few weeks.
Step 1: map processes and pain points
The audit starts with an immersion in the teams: business interviews, observation of recurring tasks, inventory of tools and available data. The goal is to spot the time deposits: manual entries, double entries between tools, document processing, repetitive replies.
Step 2: qualify and quantify use cases
Each identified use case is scored on two axes: value created (hours saved, revenue generated, risk reduced) and implementation difficulty (data, integrations, change management). This quantification turns a list of ideas into comparable investment decisions.
Step 3: prioritise into an actionable roadmap
Use cases are positioned on a value/difficulty matrix that surfaces the quick wins (high value, low difficulty) and the long-term bets. The roadmap sequences the rollouts, starting with a visible first deliverable in 3 weeks to bring teams on board.
Typical use cases identified, by function
Every company has its specifics, but after 40 engagements, the same use-case families come up in most AI audits:
- Sales: lead enrichment and scoring, meeting preparation, automatic call notes, personalised follow-ups.
- Operations: inbound document processing, cross-tool data entry, quality checks, replies to recurring requests.
- Finance: reconciliations, supplier invoice processing, payment reminders, monthly reporting preparation.
- Customer support: ticket triage and qualification, reply drafts, a knowledge base searchable in natural language.
- Leadership: structured market watch, meeting summaries, consolidated dashboards, committee preparation.
The automation side of these use cases often runs on tooled workflows: that is the role of our n8n agency once prioritisation is done.
How much does an AI audit cost?
A company AI audit costs thousands of euros, not tens of thousands: the French market range for an SME or mid-cap is 5,000 to 15,000 euros excl. VAT depending on scope (teams covered, depth of quantification). Beware of both extremes: the free audit, which is really a disguised pre-sale, and the 50,000-euro audit, which mostly funds slides.
| Format | Budget excl. VAT | Duration | For whom |
|---|---|---|---|
| Focused audit (1-2 departments) | 5,000 to 8,000 € | 3-4 weeks | SMEs, first AI project |
| Diag Data IA (Bpifrance) | 6,000 € after subsidy | 8 days over 3 months max | SMEs and mid-caps, 10 to 2,000 employees |
| Full multi-department audit | 10,000 to 15,000 € | 6-8 weeks | Mid-caps, groups, funds and portfolio companies |
Funded AI audit: Bpifrance's Diag Data IA subsidy
Since 2026, Bpifrance's Diag Data IA co-funds the AI audit of French SMEs and mid-caps: 8 days of support by an approved expert, subsidised at 40%. The remaining cost is 6,000 euros excl. VAT for a 10,000-euro service, and 2,000 engagements are co-funded over 2026-2027 (source: francenum.gouv.fr, June 2026).
Eligibility conditions, the 8-day programme and the application process are detailed in our dedicated guide: Diag Data IA: the Bpifrance-subsidised AI audit.
The 5 mistakes that ruin an AI audit
- Auditing tools instead of processes. Comparing ChatGPT and Copilot before mapping where time is lost means choosing a hammer before knowing whether there are nails.
- Trying to cover everything at once. An audit across 8 departments dilutes the analysis. Two departments in depth beat eight skimmed, extend later.
- Quantifying without the operators. Gains estimated in a meeting room are off by a factor of 2 to 3. Real volumes are observed in the field, not on an org chart.
- Delivering a list without prioritisation. 30 unranked use cases produce the same effect as zero: paralysis. The value/difficulty matrix is the heart of the deliverable.
- Stopping at the report. An audit without a first quick win deployed right after fades within three weeks. Moving to action is part of the exercise.
How to choose your AI audit provider
The French AI agency market is young and uneven. Five questions are enough to sort it:
- Have you already deployed in production the use cases you recommend, and for whom?
- What is the delay between the end of the audit and the first operational deliverable?
- Who runs the interviews: a senior consultant who then deploys, or a junior who reports?
- Does the quantification commit the provider on what follows (deployment quote based on the estimated gains)?
- Are you approved for Bpifrance's Diag Data IA, or can you articulate your audit with that funding?
Our verdict after 40 engagements
The AI audit is not a bureaucratic step, it is the best investment of the project: it costs a few thousand euros and prevents six-figure deployments on the wrong topics. Louis Graffeuil, co-founder of Tandem: "The payback of an AI deployment in a fund portfolio company: under 6 months. Sometimes under 2." That performance does not come from the models, it comes from selecting use cases upstream.
If you are starting from scratch, start small: a focused audit on one or two departments (ops, sales, finance), a quick win delivered in 3 weeks, then progressive extension. That is the trajectory that maximises the odds of measurable ROI in year one.
