Roma, Italy +39 389 972 2602

All services

Services · Monte del Gallo

AI & automation

Document processing, CRM integrations and knowledge assistants, with clear review steps and tests against your business workflow.

Automate the work around your documents and systems

We help businesses automate repetitive document, CRM and back-office work. If your team copies information from emails, searches through procedures or retypes document fields into another application, we can assess the workflow and build the software around it.

You can bring an existing application or a process that still runs across inboxes and spreadsheets. We start with one defined task, the people responsible for it and the result they need. That gives us a concrete way to decide whether AI belongs in the solution.

What we can build

  • Document processing: extract fields from PDFs, forms or scanned documents, then apply validation rules and route exceptions to a reviewer.
  • CRM workflows: classify incoming messages, identify intent and connect information to customer or deal records, within the integrations your systems support.
  • Knowledge assistants: search an agreed collection of documents and draft answers with references to the passages retrieved. This approach is often called retrieval-augmented generation, or RAG.
  • Back-office integrations: connect data entry, review and updates across existing tools. Use a language model for unstructured text and explicit rules for predictable steps.

We define what the system may read, what it may change and where a person must approve an action. Drafting a reply and sending it to a customer are separate permissions.

When AI is useful, and when it is not

AI is worth evaluating when people handle varied language or document formats and can judge a proposed result. You also need representative examples and someone who can resolve exceptions.

If a fixed rule or an API integration solves the problem, we can build that instead. Cleaning up records, standardising a form or improving search may be a better starting point than adding an assistant. We do not propose unattended decisions for a process whose errors your team cannot detect or accept.

Relevant work from our studio

These studio products and projects illustrate implementation patterns, not independently measured customer outcomes:

  • Chat RAG combines document upload, semantic search and answers with source references. Citations help people check an answer; they do not establish that it is true or complete.
  • Soshiki includes LLM-based processing of incoming emails, extracting intent and routing information to contacts or deals.
  • Alloggiati.pro uses AI document scanning to extract guest data within a hospitality workflow. This is an example of document handling, not a promise of regulatory compliance for your business.

Start with data, permissions and a test

Together, we map inputs, existing systems, hand-offs and exceptions. Before connecting real records, we agree which data the workflow needs, who may access it and which providers receive it. We discuss retention and operational responsibilities rather than assuming privacy follows from using AI.

We then agree a bounded test with representative examples and expected results. Based on accessible data and the project scope, we agree which errors to measure, such as incorrect fields, misrouted messages, unsupported answers and cases the system should send to a person. We also agree whether and how to measure review effort, handling time and operating costs.

We compare those observations with the current process and agree acceptance criteria. If the test falls short, we narrow the scope, change the approach or keep the step manual. Any rollout includes agreed review points and a way for staff to handle failures.

Discuss one workflow with Giovanni

Book a call with Giovanni. Bring a description of the task, the tools involved and an anonymised example. We can identify what needs investigation and whether to start with AI, an integration or a simpler process change.

Prefer to write? Contact the studio with a brief description of your workflow.

Frequently asked questions

Do we need AI to automate a process?

Not necessarily. If the inputs and rules are stable, an integration or a conventional workflow may be enough. We assess that option before proposing a language model.

Can you connect AI to our existing CRM?

We assess the CRM's APIs, permissions and data structure first. Those constraints determine which steps we can integrate and which need a different approach.

What data do you need to get started?

Representative documents or messages, examples of correct results, and the rules your team follows. Begin with anonymised samples rather than sharing personal or confidential records in an introductory call.

Do source citations guarantee correct answers?

No. Citations help a reviewer inspect the passages used. A model can still misread a source or miss relevant information, so we test answers and define when a person must review them.

Will our data stay inside our company?

That depends on the architecture and providers. Before connecting real data, we agree what each service receives, access permissions, retention requirements and the checks your organisation needs.

How do you assess whether automation is useful?

We agree which errors, human corrections, handling time and operating costs to measure, based on accessible data and the project scope. That plan defines the comparison with the current process and the acceptance criteria before any expansion.

All services

Start here

Talk to us about your project.

In 30 minutes we look at what you need and define a practical path to a working result.