AI audit: what to automate, at what cost, measured how
In two weeks, a map of the tasks in your business that can be automated: potential gain, feasibility, data risks, and a before-after measure to check the result. No tool sold.
Three processes reviewed
- Invoice processingKept
steady volume, stable rules, before/after measurement possible
- Client follow-upsKept
same letter template every time, validation before sending
- Litigation filesDiscarded
too many exceptions, the firm's most sensitive data
Example. The discarded row counts as much as the other two: an audit that never says no doesn't decide anything.
Two weeks, three sessions, one decision
A useful audit is one you can act on. Each session produces a decision: what to keep, what to drop, what to measure.
Session 1
Framing interview
- One hour with the people who do the tasks day to day.
- We list the processes that are frequent, repetitive and measurable, and set aside the ones with no identifiable gain.
- You choose three candidates with us.
Between sessions
Analysis and arbitration
- We describe each process in detail: input data, steps, business rules, points needing human validation and frequent exceptions.
- Each one is scored on three axes: the real, measurable gain, the risk tied to the data and the technical feasibility.
- The sensitivity of your data drives the architecture: local processing for sensitive data, cloud for the rest, hybrid when both are justified.
Session 3
Live handover
- A decision document, not a consultant's report: the choices, the reasons, the trade-offs and the rough order of the effort ahead.
- For each automation selected: what to automate, in what order, what tests to run before going live, and what before-after measure to use.
- You leave with what your team needs to decide.
Deliverables that decide, not ones that sit on a shelf
No tool sold in the audit: a documented decision on what to automate, in what order, and how to measure the result. Here is the shape of the documents.
| Process | Value | Risk | Feasibility |
|---|---|---|---|
| Invoice processing | Value : 3 sur 3 | Risk : 1 sur 3 | Feasibility : 3 sur 3 |
| Client follow-ups | Value : 3 sur 3 | Risk : 1 sur 3 | Feasibility : 2 sur 3 |
| Quote preparation | Value : 2 sur 3 | Risk : 1 sur 3 | Feasibility : 3 sur 3 |
| Litigation files | Value : 2 sur 3 | Risk : 3 sur 3 | Feasibility : 1 sur 3 |
Example. The last row is a discarded process: high value, but high data risk and low feasibility. An audit that never says no doesn't rank anything.
D0
Audit delivered
the three processes selected
D+30
Pilot
a single process, bounded scope
D+60
Measurement
comparison against the manual process
D+90
Decision
extend, adjust, or stop
Example. Each milestone carries its owner, its test, and the human-validation rule set for the step.
- Accounting recordsstays on your machines
- Named client filesstays on your machines
- Catalogue and pricingcan go out
- Letter templatescan go out
Example. This list is what decides the architecture, not the other way round: what must stay forces the local setup, the rest can be processed elsewhere.
Before reading
Time spent per file, measured over two weeks by the people who do the task.
Validation threshold
The pilot is only kept if the same measurement, redone after go-live, clears the threshold agreed on during scoping.
Example. No figure is announced here: the threshold is set with you, on your own measurement, before the project starts.
- Filled selection matrix
- Your processes scored and ranked on value, risk and feasibility, with the note behind each position.
- 90-day roadmap
- The steps, the tests, the human validation rule and the owner for each automation selected.
- List of sensitive data
- The data that must stay local, and why, within the GDPR and AI Act framework. This list guides every architecture decision.
- Proposal for what comes next
- An option for the prototype, with a rough order of the effort, with no obligation to take it.
- Handover document
- The audit summed up in a single document: the choices, the reasons, the trade-offs and the processes set aside.
- Defined before-after measure
- For each automation selected, the indicator recorded before the project and the threshold that will confirm the gain once it is live.
Limits set in advance, for a neutral opinion
An automation audit only has value if it stays neutral. Here is what we do not do, and what that guarantees for your decision.
No tool imposed
The recommendation compares the options with no tie to any vendor. You stay free to choose the tool with your team.
No hidden subscription
The audit funds no commission. No solution is recommended because it would earn us anything.
No handoff to another agency
The mission runs in-house, start to finish. If the audit concludes that nothing should be done, we say that too.
A documented decision
The main deliverable is not a sales pitch: it is a written decision, with its reasons and its conditions.
A before-after measure
No gain is announced without an indicator. Each automation selected has its own measure and its own validation threshold.
The freedom to do nothing
If no process is worth automating right now, the conclusion says so honestly. That, too, is a useful decision.
Different profiles, the same question: what is automation worth here?
The audit is for teams that want to decide on facts, not on promises. Here are the profiles that gain the most from it.
- SME buried under administrative tasks
- Data entry, follow-ups and filing eat up hours every week. The audit estimates the time you could get back and the measurable gain on these tasks, before any investment.
- Accounting firm
- Journal entries, reconciliations and supporting documents are natural candidates. The audit checks feasibility and the validation points before any automation.
- Law firm
- Reports, letters, case tracking: processes with a heavy writing load. The audit identifies what AI can assist without touching the lawyer's judgment.
- Estate agency
- Listings, viewings, prospect follow-ups, mandate management: the same steps repeat on every file. The audit spots the ones that can be automated safely.
- Tradesperson or micro-business
- Alone or in a small team, you have no IT department. The audit only keeps automations that are simple, maintainable and free of heavy technical dependencies.
- A business that already tried automation
- A first attempt with no result, a tool adopted then dropped. The audit reviews what failed and why, before starting again on solid ground.
What the audit looks at for each of your candidates
Here are examples of processes the audit can review. For each one, we look at the data, the steps, the rules and the validation points.
- Invoice entry
- The audit looks at the input formats, the fields to extract, the accounting rules and the validation point before payment.
- Customer follow-ups
- The audit looks at the content of the reminders, their frequency, the payment terms and the rule that decides when to move to a phone follow-up.
- Quote preparation
- The audit looks at the calculation elements, the discounts allowed, the mandatory wording and the validation before it goes to the client.
- Meeting minutes
- The audit looks at the note-taking, the structure of the minutes, how actions are assigned and the validation by participants.
- Email management
- The audit looks at the sorting into folders, the template replies, the attachments and the emails that need a personal reply.
- CRM updates
- The audit looks at the fields entered, the information sources, the duplicates and the point at which a record needs enriching or correcting.
- Document filing
- The audit looks at the document types, the naming rules, the folder structure and the access allowed for each document.
- Leave tracking
- The audit looks at the requests, the balances, the manager's approvals and the history passed on to payroll.
Frequent questions about the automation audit
The process, the commitments and the limits of the audit, in a few straight answers.
How long does the audit take?
Two weeks, across three sessions: the framing interview, the analysis of the three processes and the live handover. Between sessions, we work on your documents and your data.
Are we required to automate afterwards?
No. The proposal for what comes next carries no obligation. You can use the roadmap with your own team, with another provider, or launch nothing at all.
Which processes do we analyze?
Three processes chosen with you during the framing, from among those that are frequent, repetitive and measurable. The matrix then ranks the other candidates for later.
How is the gain measured?
The roadmap defines, for each automation, an indicator recorded before the project and a validation threshold once it is live. The gain announced is the one this measure confirms.
Does our data leave the business?
The recommendation favors local processing for sensitive data. The list of sensitive data produced by the audit sets out what can be processed in the cloud and what must stay on site.
Does the audit recommend tools?
Yes, a neutral, documented recommendation that compares the options with no tie to any vendor. We do not sell a tool: the decision stays with your team.
Get your automation potential audited
A first, free conversation to understand your challenges and tell you plainly what AI can and cannot do for you, with a rough order of the effort involved.
To go further
Other formats complement this offer, depending on where you're starting from.
Bespoke automation
Once the audit is approved, we build the automations chosen in the roadmap, following your validation rules.
Business AI agents
For processes that call for judgment, an AI agent trained on your local data and your documents.
Training on AI use
So your teams can make the most of the automations and keep control of the validation points.
Advisory audit
A broader view of your organization and the AI uses it could support, with no implementation.