Success Story: How a Multi-Company Services Group Caught Up on AI, from Discovery to a Trained Office Team

At a glance

• 8 departments visited on site, each one in its own session
• 75 processes mapped for AI and automation
• 1,600+ hours a year freed up across the group
• 0 ports opened to the internet to connect AI to their systems
• The whole office team trained, first together, then department by department
• One outsourced AI team working alongside their IT department from start to finish

The client

A group of companies in the Balearic Islands working in waste management, urban furniture, landscaping and public services, with a technical office that bids for public tenders and shared finance, IT and HR teams. Strong in operations, and held together by spreadsheets, paper and a very busy IT department.

The starting point: “we’re missing out”

Management saw it clearly: competitors and partners were starting to work with AI, and they weren’t. They didn’t have a list of projects, and they didn’t need one yet. They knew they were losing time every day in ways that AI could fix, and they wanted someone to tell them where, how and in what order.

Their IT team was strong but stretched: two people covering data, systems, security and support for the whole group, with projects waiting months behind the day’s urgencies. Hiring an in-house AI team wasn’t realistic. So they brought in Aitomatik as their outsourced AI and automation team: the specialists who bring the know-how and the method, and who work hand in hand with their own tech people instead of replacing them.

What we found when we started:

• Excel and paper running critical processes. Payroll checks, quotes, council reports and certificates were all assembled by hand.
• Systems that didn’t talk to each other. An in-house ERP, Business Central, HR software and several internal apps, with people copying data between them.
• Inboxes and phones swamping the office. In high season, around 100 quote requests a day arrived by email at a single desk.
• Years of knowledge locked on a file server. Seventeen years of documents that nobody could search.
• One non-negotiable. Company data and system credentials could not leave the building.

Step 1: Preparation

Before setting foot in their offices, we did our homework. We worked through the group’s own internal report on problems and needs, mapped the companies and departments, and prepared a set of questions for every area. We also agreed the visit agenda with their IT lead, so each session went to the right people with the right questions.

Step 2: Two days on site, one department at a time

We sat down with every department: technical office, urban furniture, landscaping, quality and health & safety, waste management, finance, IT and HR. Each session started open, with one question: “If you had a sharp intern, what would you hand over?” Then we went through the questions from the preparation phase.

Some ideas were validated live in the room. By the end of the IT session, the team had automated a fleet-data download and was reading quarry weighbridge tickets with vision AI, and IT started using Claude Code that same day.

Step 3: Conclusions, 75 processes

From the visit came the full map: 75 processes where AI or automation could help, each one scored on impact and effort, with an architecture recommendation and a starting plan. Transversal quick wins went first, because they deliver value in weeks and fund the rest.

Step 4: The turn: faster tools, and making them compliant

While the plan was being put into motion, the technology moved faster than anyone expected. Much of what would have required custom development a year earlier now came built in. A close look at the 75 processes showed that most of them could be done with Claude Cowork, an AI agent that works on the employee’s own computer with their files, email and tools. That made the route to value simpler and much cheaper.

It also raised the question management had asked from day one: what exactly leaves our network? We took them through every possible architecture: everything inside their own servers, the agent on the employee’s computer, or a server published on the internet. For each one, an interactive explainer walked step by step through what is sent, what stays inside, who approves it and what gets logged. Leadership decided with the full picture.

Then we built it together with their tech team. We brought the architecture, the security model and the reviews; their IT people brought the knowledge of their own systems and permissions, and own the result. The chosen option was a read-only gateway inside their own network:

• Nobody connects from outside. The connection always starts from the employee’s computer, inside the network. No server is published and no port is opened.
• Read-only by design. The AI can consult data but has no way to change it.
• Everyone sees only what they already could. Permissions are resolved per person against Active Directory.
• Credentials never move. They live in the gateway, so neither the employee nor the AI model ever sees them.
• Everything is logged on the company’s own infrastructure.

Now an employee asks, “Which invoices from this client are overdue?”, and only the answer travels to the model, never access to the system.

Step 5: Training the whole office team

Tools only pay off when people know how to use them, so the last step was AI literacy for the entire office team, in two rounds.

A general workshop for everyone: From chatbot to agent.

• Delegating results, not asking questions. The file ends up finished in the right folder.
• Turning repetitive work into reusable skills. Commands like /quote, /follow-up and /weekly-report.
• Scheduled tasks, so recurring work runs on its own.
• Email triage and drafts, always reviewed before sending.
• A clear usage policy. Corporate accounts only, read-only access to systems, and a person signs off everything that leaves the company.

Then one session per department, on their own use cases. We sat with each team and ran their real processes from the 75: payroll checks with HR, quotes with sales, council reports with the technical office, invoice reconciliation with finance. Each department left with its own skills working, not a list of ideas.

The programme also covers the AI literacy obligation that the EU AI Act (Article 4) places on companies whose staff use AI.

Their outsourced AI team, for the long run

The engagement doesn’t end with the training. We stay on as the group’s outsourced AI and automation team:

• Advising and backing their IT team. Architecture decisions, code and security reviews, and a second opinion whenever a new tool or idea comes up.
• Development reviews, to get more out of Claude Code across the roadmap.
• Follow-up sessions with each department head after a few weeks of daily use, to review how they use AI and raise the bar on their own cases.
• New automations as the business needs them, picked from the 75 and beyond.
• A pool of advisory hours for questions, new ideas and priority changes as the business evolves.

Their IT team stays in charge of their systems. We make sure it never has to figure out AI alone.

The results

Process	Before	With the new process
Payroll verification against the payroll provider’s PDFs	16–25 h/month	under 2 h/month
Monthly reports for councils	3–4 h per report	30–45 min of review
Registering a new employee across 6 systems	15+ h/month in peak season	one single entry
Quote requests in high season	~100 a day, typed by hand	drafted automatically, reviewed and sent
Analysing tenders for large jobs	up to two weeks to quote	a few days

Across the group, that adds up to more than 1,600 hours a year handed from repetitive admin work back to the business. Just as important:

• Less dependence on single people. Reports and tenders that only one person knew how to prepare are now reproducible.
• An IT team that ships faster. The same team, with an AI accelerator, now moves projects forward that had been waiting for months.
• AI adopted safely. Data stays inside, credentials never move, and every query is logged.

Key takeaways

1. You don’t need a list of projects to start. A structured discovery turns “we’re missing out” into a prioritised map.
2. Re-check the plan as the tools move. What needed custom development last year may come built in today.
3. Keep your data where it is. The right architecture lets AI read your ERP without exposing it.
4. You don’t need to hire an AI department. An outsourced AI team alongside your own IT gives you the expertise without the headcount.
5. Train everyone, then each team on its own work. General AI literacy gets people started; department sessions on real use cases make it stick. And it’s now a legal requirement in the EU.

Could your company do the same?

If you feel you’re missing out on AI but don’t know where to start, that’s exactly where this group was. We can be your AI and automation team too, working alongside the people you already have. Book a free discovery call and we’ll ask you the intern question too.