01
Documents that become work
Read requests, attachments or forms; extract what matters; prepare a draft; ask for confirmation when a fact is missing.
Quotes, intake, applications, reports and contracts under review.
AI automation in Italy
AI automation built around the work people actually do.
I do not start with the model. I start with a queue of work, the exceptions slowing it down and the person accountable for the outcome. Then I build the part of the system that can genuinely return time without giving up control.
For founders, operations leaders and teams working in Italy who already have a process, not for anyone looking for a demo to show around.
Automation is not another chatbot.
The value appears when a system takes responsibility for a well-bounded operating step, uses the right data and knows when to stop.
02
Queues that require judgement
Classify, enrich and route cases without pretending that every exception is the same as the one before it.
Shared inboxes, tickets, leads, supplier requests and escalations.
03
Systems that do not talk
Connect the tools already in use and preserve data provenance instead of creating one more dashboard to maintain.
CRM, ERP, email, spreadsheets, databases and internal portals.
Five questions before you spend a euro.
A serious AI automation company should also tell you when not to build. This is the filter I use at the start.
Does the work repeat often enough to observe?
Good signal: Real cases, examples and a recognisable frequency exist.
Caution: The process changes every time and nobody describes it the same way.
Is the required data accessible and trustworthy?
Good signal: Sources and permissions are known; errors have an owner.
Caution: The data lives only in people's memory or files with no history.
Can the exceptions be named?
Good signal: We know when the system must stop and who needs to decide.
Caution: The implicit goal is to automate everything, including ambiguous cases.
Can the outcome be measured before the build?
Good signal: Time, errors, waiting or capacity have an observable baseline.
Caution: Success is described only as innovation or using AI.
Will someone own the process after launch?
Good signal: A person owns the rules, exceptions and improvements.
Caution: The system is expected to run alone without maintenance or decisions.
Three or more caution signals do not necessarily mean no. They mean the first project is making the process observable.
Production experience, not AI theatre.
I have spent fifteen years building software with real users, incidents, deadlines and consequences: from London startups and Amsterdam fintech to AI systems and applied-AI teaching at SDA Bocconi. I bring the same discipline to business automation.
- 15+: years in software engineering
- Seed → IPO: product leadership through a public listing
- 3.2★ → 4.7★: App Store rating during the team's work at bunq
- SDA Bocconi: two sessions on applied and agentic AI
These are facts about my experience. The result of your automation will be defined and measured against your process, not borrowed from someone else's case study.
Process first. System second. Proof last.
Four stages, each with a verifiable exit. If one does not hold, the problem does not get hidden in code.
01
Observe
We reconstruct the work as it happens: inputs, decisions, waiting, manual steps, exceptions and current cost.
Output: flow map and baseline.
02
Bound
We choose what to automate, what remains human, which data is allowed and where the system must stop.
Output: scope, ownership and acceptance criteria.
03
Build
We connect systems, make decisions visible and treat errors and rollback as part of the product.
Output: verifiable automation in a controlled environment.
04
Prove
We compare the new flow with the baseline. Only what genuinely reduces work, waiting or risk gets extended.
Output: pre/post result and rollout decision.
A useful yes is worth as much as a timely no.
It is worth a conversation when
- a process consumes hours every week or slows revenue and service;
- the team already uses digital tools but spends its time copying and checking;
- errors or waiting have a recognisable cost;
- you are willing to start with a small scope and test it properly.
It is probably too early when
- you want an AI feature without a precise operating problem;
- nobody can show real examples of the work;
- the process has no accountable owner;
- the project requires removing every human check on day one.
Automating quotes without automating mistakes.
I published the full method against a workflow many Italian businesses know: customer data, pricing rules, exceptions, human review and the choice between SaaS and a custom system.
What is worth clarifying early.
What does an AI automation company actually do?
It should study a process, choose a useful boundary, connect data and tools, handle exceptions and human control, then prove the result against a baseline. Selling a chatbot or a demo covers a much smaller part of that job.
How much does an AI automation project cost?
It depends on data access, the number of systems involved and the risk of the actions. When the use case is not yet defined, I start with a 5,000 euro AI Assessment. Complete implementations normally start at 40,000 euro. Scope comes before the quote.
How long does it take?
A focused assessment can close in one week. Complete specifications usually take two to three weeks. An implemented solution often takes six to twelve weeks, but a timeline is meaningful only after process, integrations and exceptions are bounded.
Do you work with Italian companies from Amsterdam?
Yes. I am Italian, work in Italian and keep Italian working hours; Italy is the primary market. The Dutch base is the legal and operating home of DL Solutions, not a barrier to working with teams in Italy.
Do we need to replace the software we already use?
Usually not. The first goal is to use what exists more effectively by connecting systems and data with clear ownership. A tool is replaced only when the constraint is demonstrated, not to make the project look more impressive.
Bring the process, not a list of technologies.
Tell me what work is piling up, who handles it and what happens when it goes wrong. My first reply will be a fit assessment, not an automated demo.