AI adoption
Adopting AI and building it into processes
Your people already use AI, each in their own way. I help them use it deliberately, with rules we write together, and I build AI into the processes where the numbers say it pays off.
Before the engagement, we agree which systems to assess, the documents to deliver and a fixed price. Development and ongoing support are agreed separately.
From isolated experiments to a process you can measure
Between pilots that never reach production and tools used without rules, the risk is spending without results and leaking confidential data. I start from one of your processes that costs time or money and check whether and how AI improves it, by measuring it.
Adopting: from ad hoc use to use with rules
01
Map
Which AI tools you actually use, who uses them and with what data. There are usually more than IT knows about.
02
Assessment
The opportunities worth pursuing, ranked by impact and feasibility, the AI Act points to check, and an action plan for the first three months. Fixed price.
03
Rules and training
Guidelines developed with the teams: which data can be used, where and with which tools. Content and availability of role-specific training to be agreed.
04
Measurement and ongoing support
I check the results against the process figures and, if you wish, support you over time.
Where we look first in a technical company
Typical examples, to be checked against your own case.
- Technical department · search across drawings, standards and past jobs.
- Bids · first draft of the technical bid based on the client's requirements.
- Quality · analysis of non-conformities and draft reports.
- Support · answers from machine manuals, in several languages.
We assess on-premises models where specific restrictions on data and external connections are needed.
What I build
AI agents and assistants connected to your data
Documents, procedures, knowledge bases: answers with sources, clear limits and human validation where mistakes are costly.
Automating repetitive workflows
Document management, email, reports: the tedious parts that a person does today.
Integrations with your business systems
ERP, CRM and internal systems connected to AI models, with control over costs and access.
AI that stays in-house
Models installed on premises, with per-person access and logs, designed so that data stays in your infrastructure and the only calls to external services are the ones you decide on.
How I build an agent
Five controls I put into every system before it goes into production.
-
1
Automated tests
Representative test cases rerun after each change, with acceptance thresholds agreed before release.
-
2
Least privilege
The agent uses only the tools and data it needs, nothing more.
-
3
Human approval
Before any action that cannot be undone, a person decides.
-
4
Audit trail
I define which events to log and how long to retain them, limiting personal and confidential data in the logs.
-
5
Cost limits
Budgets per user and per project, with alerts before they are exceeded.
Privacy, costs and control
I design starting from where the data ends up, how much each request to the models costs and who controls the sources. On compliance matters I work with the company's DPO and legal counsel.
Usage logs also concern employees
Usage logs may contain workers’ personal data and enable remote monitoring (Article 4 of Italy’s Workers’ Statute, Law 300/1970). I design them with the company’s privacy and legal teams: defined purposes, limited retention, restricted access and information for workers. Where required, the procedures for a union agreement or, failing agreement, Labour Inspectorate authorisation must be followed.
Someone who builds, as well as advises
- I advise on an AI adoption programme at a high-tech industrial company: teams, rules, an on-premises model.
- I built and run a platform of 6 applications in production for a non-profit organisation.
- I work with AI agents every day, using my own method of rules, automated tests and reviews.
- I can explain architectures and methods; I share project materials only where permitted.
Let's start from one of your processes
Tell me which activity takes up most of your time: I will reply with feasibility, risks and a first step.