Your CEO asks what your top 10 customers truly cost, all-in. Your CFO stares at her screen. The sales rep pulls out a USB stick. The project lead promises to "get back to you Friday."
It's Monday morning. The question is basic. Nobody can answer.
You know you need an audit. You just don't know where to start. And you already suspect some firm will try to sell you 40k euros of mapping for a PDF nobody will read.
A data audit doesn't start with a spreadsheet
It starts with a conversation.
Most audit engagements that end up in a drawer share the same flaw. They begin with a technical inventory. Listing tools, mapping databases, drawing an architecture diagram in 40 slides. And ignoring the only thing that matters: which decision you keep postponing because the numbers are fuzzy.
If you start with the plumbing, you end up with clean plumbing and zero decisions unblocked.
A data audit that doesn't unblock a decision is just an expensive PDF.
Step 1: list the decisions you're putting off
Take a sheet. Write down the decisions your exec committee has been dragging for 2 months because the numbers aren't clean.
Typical examples:
- Should we shut down product X, which looks unprofitable?
- Which customer segment deserves more sales focus?
- Is acquisition channel Y actually paying off, once hidden costs are included?
- What does a customer cost us over their full lifetime?
That list is your starting point. Not your CRM. Not your ERP. Not the data warehouse nobody has updated in 18 months.
If you can't list 3 blocked decisions, you don't need a data audit. You need a strategic clarification workshop. That's a different job.
Step 2: trace back to the questions
Every blocked decision hides 2 or 3 unanswered questions.
"Should we shut down product X?" becomes:
- What's the real production cost of X, including after-sales support?
- What's the net revenue after discounts and returns?
- What's its weight in the portfolio over 24 months?
That's where you start seeing where it jams. Most of the time, the questions already exist in the CEO's head. It's just that nobody can answer them in less than 3 days.
Step 3: map what exists, and only what exists
Now, and only now, you look at the tools.
For each question identified, you trace:
- Where the source data lives (CRM, ERP, Google Sheets, billing, payroll, custom business tool)
- Who updates it, how, at what frequency
- Who needs to read it to make the decision
You will find three things, every single time:
- At least one critical data point isn't captured anywhere reliably
- At least one tool acts as "source of truth" without anyone having decided it should
- At least one person knows part of the system by heart, and nobody else can replace them
Those are your real priorities. Not the choice between BigQuery and Snowflake.
Step 4: prioritize with one simple rule
For each identified track, two questions:
- How many decisions does it unblock?
- How much does it cost to set up?
Tracks that are "many decisions / low cost" go first. Often that's a simple report automation or consolidating two CSV exports into a single dashboard. Not an 80k-euro project.
Tracks that are "high cost / low immediate impact" wait. Even if they look good on paper.
The shiny new data warehouse, the MLOps platform before you even have a model in production, the migration to a "modern stack because it's the standard": all of it waits.
3 traps that kill an audit
Trap 1: only talking to leadership. If you don't talk to operators, the sales rep juggling 3 Excel files, the floor lead updating stocks by hand, you miss 80% of the reality. Leadership sees the monthly report. Operators live the machine.
Trap 2: delivering a report, not a roadmap. A useful audit ends with a list of prioritized tracks, costed, with an estimated gain. Not a 60-page PDF describing the existing. You already know the existing, it's yours. What you're buying is the path.
Trap 3: wanting to redo everything. If the audit concludes with "we need to rebuild from scratch, migrate to a modern stack, hire a data team," you didn't get an audit. You paid someone to sell you a project. A good audit proposes gains in 4 to 6 weeks, using the tooling you already have.
How long it takes
For an SME of 20 to 200 employees: 2 to 4 weeks of work, split between interviews, mapping, and debrief. Not 3 months of deep diving.
If someone sells you a 6-month audit, ask yourself what there truly is to audit. Unless you have 500 people and 15 IT systems, that's too much.
What to require in the contract
- The number of operator interviews planned, not just C-suite
- The format of the final deliverable: a list of tracks with estimated gain, cost, and sequencing, not a descriptive report
- A debrief session where you can push back on conclusions, not a PDF emailed over
- A clear commitment on when the engagement ends: an audit shouldn't mechanically lead to 12 months of billed follow-up
Where to start concretely
If you want to prep the ground before calling anyone:
- Re-read the 5 signs you need a data consultant. Spot yours.
- List the 3 decisions you're postponing.
- Note who would answer them, if the data was clean.
You've already done 60% of the audit. The rest is method.
Let's talk for 30 minutes if you want to validate your read on the situation before committing. To see how I run these engagements, head to the data consulting page, or the dedicated Geneva page if you're in the area.