AI in business is not a tool question. It is a question of workflows, people and margin.
Practical guides to understand where to intervene, how to build useful AI systems and how to verify their value over time.
All the guides, by pillar
What integrating AI in a company really costs
What AI costs a company is not one number: it is the price of the pilot plus the price of the system that stays switched on in production, and almost nobody budgets for the second one at the start. Worldwide AI spending will grow 44-47% in 2026 to 2.5 trillion dollars (Gartner), and in Italy the market is worth 1.8 billion euros with 71% of large companies already running a project against 8% of smaller ones (Osservatorio AI, Politecnico di Milano). The real cost depends on the scope you choose to take on, not on the tool you buy.
Read →How to measure AI ROI (beyond the POC)
AI ROI is not measured by counting hours saved or active licences: that is a usage criterion, not a value criterion. It is measured by defining, before you start, a single number in euros to verify every month. McKinsey (State of AI 2025, 1,993 companies) confirms the pattern from the outside: 88% of companies use AI regularly, but only 5.5% report real financial impact. The gap is almost always the same one: nobody decided up front what to count.
Read →The AI Act and smaller companies: the 'AI literacy' you are obliged to have (since 2025)
Article 4 of the European AI Act has been in force since 2 February 2025: anyone who develops or uses AI systems must ensure an adequate level of 'AI literacy' among the people working with them. It applies to smaller companies too, even if you only use third-party tools like ChatGPT or Copilot. There are no minimum hours or mandatory certifications, but you have to train staff proportionately and be able to document it. This article is a practical guide, not legal advice.
Read →AI agents in business: what they really automate (and where they need boundaries)
The difference between an AI agent that works and one that creates a bigger problem than it solves is not the power of the model: it is how clear the scope is and how tight the supervision is. On a repetitive, rule-based process with sharp boundaries, an agent multiplies results: in a real case, second contact on hot leads went from 50% to over 90%. With no scope and no control over what it answers, the same agent at scale becomes a bigger risk than the one it was meant to solve.
Read →AI for small business: where to actually start (without wasting budget)
In 2025 only 14.2% of small Italian companies use AI, against 53% of large ones (Istat). The gap does not come from budget: it comes from the fact that almost no smaller company starts from the right process. The ones that made it did not buy an AI project, they picked one small, specific thing that was costing them time or customers every day, and they fixed it. The starting point is not which tool to buy, it is which process to measure first.
Read →The AI skills a company actually needs (and the AI Champion role)
Companies name skills as the number one brake on AI, ahead of cost and technology (Istat 2025). But a one-off course for the whole team has the same flaw as customer onboarding without tracking: nobody really knows who learned something they use every day and who just attended. What works better is an internal reference point per department, the AI Champion, who practises on real processes and spreads what they learn. The skills that count are not technical but practical: delegating the right tasks, writing clear instructions, recognising when not to trust the output.
Read →Automating quotes and documents with AI: the real problem is not speed
The problem with quotes and documents is almost never writing speed: it is the total darkness about what happens after you hit send. AI automates data gathering, the first draft and consistency checks well, but the real jump shows up when you know who opened the document and when: the follow-up latches onto the moment the client is deciding, not onto the calendar of whoever has to remember to call. Review and pricing stay human decisions, always.
Read →How to choose an AI consultancy (or agency): the honest guide
'Which is the best AI agency' is the wrong question: anyone giving you an answer is almost always selling themselves or a sponsor. The question that counts is a different one: who takes responsibility for the number this project is supposed to improve? From there you pick the right type of supplier (freelance, agency, system integrator, software house, operating partner) and the real criteria. At the end, the 7 questions to ask before signing, useful with any supplier, not just with us.
Read →How to integrate AI into business processes: the operating guide (in 4 phases)
Integrating AI does not mean picking the right tool: it means finding the process that is actually losing value, often a problem that looks technological and is not. The sequence that works has four phases: diagnosis (map the process and quantify the loss in euros), system (build and ship to production on a bounded scope), value (measure every month what changed) and autonomy (train somebody inside the company so the system holds on its own). Skipping the diagnosis to rush to the tool is why most AI projects leave no trace.
Read →SaaS or a custom AI system: what to choose (and when each one wins)
A SaaS dashboard can show green numbers while a real problem is burning cash, because it reads one channel and does not talk to the company's other systems. A custom system, embedded in your data, costs more up front and needs a partner, but it sees what SaaS cannot see by construction: where the data intersects. Practical rule: SaaS for standard processes, a custom system for the process where you need to read several systems together or where your competitive difference is at stake.
Read →Why 95% of AI projects fail (and how to be in the 5% that works)
MIT NANDA measured that 95% of enterprise generative AI projects produce no measurable impact on the P&L (The GenAI Divide, 2025). This is not a model problem: it is a problem of integration into processes, of daily monitoring, and of who answers for the results. Projects built with specialised partners succeed roughly 67% of the time, against a third for purely internal builds. The 5% that works starts from a bounded scope, a problem measured in euros, and one person who follows it every day.
Read →Does AI take your job? The right question is a different one
The question 'does AI take your job' is badly framed: an occupation is a bundle of different tasks, and AI absorbs some of them, not the whole package. At an energy reseller we worked with, the task that disappeared was keeping in mind who to call back: the salesperson stayed, and second contact on hot leads went from 50% to over 90%. The 2026 data (Anthropic Economic Index) confirms that the unit being automated is the task, not the occupation, with different effects role by role. The useful question for a company is not whether to cut, it is which task inside the job should be removed because it was never the job.
Read →Before you choose AI, let's get the words right.
Five terms that come back on every page. If they are not clear, every discussion about AI becomes a discussion about what we mean.
You know where to start. Now let's find where to intervene.
If you have recognised a problem in your workflows, the next step is not another article. It is finding out whether that problem is already leaving margin on the table.