AI consulting for businesses and SMEs across Italy. We examine how your company works, select the process with the greatest value and put it into production with the right model, measuring the difference. The people who analyse the process also write the code.
AI consultancy identifies where artificial intelligence creates value. We analyse processes, choose a starting point and model, build a prototype using real data and deploy it within existing software, measuring time and errors before and after.
The context
The tools are there. What is missing is someone who knows where to use them.
15.7%
of Italian SMEs with at least 10 employees used at least one AI technology in 2025. A year earlier, the figure was 7.7%.
58.6%
of businesses that considered AI without adopting it cite a lack of skills. Next are clarity of regulations (47.3%) and data quality (45.2%).
Models are available through subscriptions. The missing piece is knowing where to use them, with which data and rules. That is the work of effective AI consultancy.
Source: ISTAT, Enterprises and ICT, 2025, published on 15 December 2025.
The process
Six steps, always in this order. Their duration can vary, but the sequence stays the same. An audit provides the process map; measurement shows whether the work has succeeded.
We listen to the people who do the work every day, as well as those who manage it. We examine emails, files, business management software and verbal handovers. Where possible, we measure requests, time and errors.
What comes out of here: a list of processes, with actual volumes and timings.
We map responsibilities, data and bottlenecks. We start with a frequent process that has clear rules and accessible data, while keeping other candidates on the list with reasons for their priority.
What comes out of here: the process to start with, and why it is the right choice.
We evaluate Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1, GPT-6.1 Sol and GPT-6 Astra, or an open model on a European server. Sometimes a rule is enough. Tests on your data, volumes, data-processing terms and cost per task guide the choice.
What comes out of here: the architecture and an estimate of running costs.
On your real data, in days. We write it with programming agents like Claude Code and one person reviews every change. You try it on your work, not on a prepared demo.
What comes out of here: a working workflow to put to the test.
The prototype becomes a working system: connections to business management software, CRM and email, permissions, a log of every action and human handover when a case is uncertain.
What comes out of here: the workflow running within your tools.
After implementation, we measure the same audit metrics: hours, errors and response times. If results do not improve, we adjust or stop the workflow. Then we move to the next process.
What comes out of here: the before and after, measured using the same figures.
For model selection, see our guide to how to choose the AI model; for the financial calculation, see the guide to how to calculate the ROI of an AI project.
Who it is for
Quotes, orders, supplier invoices and enquiry emails: repetitive work that keeps talented people busy.
Individual use works. The next step is a business workflow, with controlled data and verifiable results.
Built with an AI builder or by a collaborator: it needs to be secured, connected to your systems and maintained.
Off-the-shelf software, automation or custom software: consultancy also helps you avoid building something you can buy.
That is fine: this is what an audit is for. And sometimes the honest answer is “not yet”.
Anywhere
Much of the work is remote: calls, screen sharing and trial access to your tools. Businesses in Bari or Bolzano follow the same process with the same care.
Milan
Our studio is at Via Lazzaro Spallanzani 10 in Milan, close to the M1 metro and the Passante rail service. We can meet there or spend a day at your premises: some processes are best understood by observing them.
AI consulting in Milan for SMEsThe rules
For most SMEs the obligations are few and clear. They must be put into the project, not added at the end. Situation updated to October 2026.
2 February 2025
2 August 2026
2 December 2027
2 August 2028
ContractsEvery provider processing data on your behalf signs a data processor agreement (Art. 28), including the model provider.
Impact assessmentIf processing may create a high risk to people, an impact assessment is carried out first (Art. 35).
Data minimisationModels only get what they need. We choose vendors that don't train on your data and, where needed, servers in Europe.
Further readingThe full explanation is in the guides AI Act, obligations for SMEs and AI and GDPR.
What it is not
If a process with a clear return does not come out of the audit, we tell you and stop there. Better one more call than a useless project.
| AI consulting | AI training course | An AI assistant licence | |
|---|---|---|---|
| What you get | A running process, with measured results | Better-informed people | A tool for each person |
| Which data it uses | Yours, connected to your systems | Generic examples | Whatever each person pastes into it |
| Who runs it | The workflow, overseen by a person | Who took the course | Everyone, in their own way |
| When it makes sense | Repetitive work with clear rules | To start with the basics | Writing, summarising, searching |
Case studies
Two projects that explain our method better than any slide.
AI-powered business software · luxury rentals
Fleet OS, the custom business management system, uses AI to read vehicle registration documents, insurance policies and road tax documents, then fills in the vehicle record. When a traffic fine arrives, it matches the date and time against rental bookings, finds the correct customer and drafts an email or WhatsApp message. The operator checks it and sends it.
The Rial Rent case studySaaS · iOS and Android apps
The personal trainer platform combines athlete management, training, nutrition, invoicing and chat. One web codebase is delivered through App Store and Google Play and an installable PWA, in six languages.
The Athleex case studyAfter the audit
The questions we hear on calls, with the answers we give.
An AI consultant studies how the business works, identifies processes where AI saves time or reduces errors, chooses the model and architecture, builds a prototype and deploys it. At Giallo Studio, the people doing the analysis also write the code, so recommendations are tested in practice.
The initial 30-minute call is free. The audit is quoted separately and deducted from the project cost. The first project has an agreed fixed price, so you know the cost before starting. Pricing factors are explained on the page prices.
The call takes half an hour; audit duration depends on the processes assessed. The prototype arrives in days. Integration and data requirements determine production timelines, which we agree in writing before starting.
No. We need a person who knows the process and access to the tools where the data is. If the data is messy we see it in the audit and we put it in the plan, instead of finding out on the go.
Yes, almost everything is done remotely. If you prefer to see us, the studio is in Milan: the details are on the page AI consulting in Milan.
Training builds skills, licences provide tools and consultancy integrates AI into a business process with your data, systems and measured results. These often work together.
From the start of the project: the assistant identifies itself as AI (Article 50 of the AI Act), providers sign data processor agreements, models receive only necessary data and, where risk is high, an impact assessment is carried out first (Art. 35 of the GDPR).
That happens, and we tell you. Sometimes a rule, an integration or better use of existing software is enough. Consultancy that recommends AI everywhere does not help you.
Free initial call · 30 minutes
Tell us briefly which task takes up the most time. Within one working day, we will contact you to arrange a half-hour call and assess whether AI makes sense for your situation. We will tell you if it does not.
Continue with
Related Guides
