AI in the workplace: why do companies set up separate entities to innovate without slowing down their business?

The idea is simple: innovation requires a different timeframe to production. Those working on existing products must ensure reliability. Those working on the future must be able to proceed by trial and error.

This philosophy is summed up in a phrase spoken by Linda Gilli, president of Inaz and recipient of the ‘Cavaliere del Lavoro’ honour: ‘being allowed to make mistakes in order to learn’. It is a concept that often struggles to gain acceptance in traditional companies, but which is almost a prerequisite in the world of artificial intelligence.

Initia was thus founded as an in-house start-up, backed by a group with almost eighty years’ experience in software and services for the world of work.

The project is led by Massimo Pegori, the group’s Director of Innovation and Software and Infrastructure Development. His role is to bridge the gap between research and business, ensuring that experimentation does not remain confined to the laboratory and that the solutions developed are put to practical use in the company’s products and processes.

“The first solution we’ve been working on is called LexMate,” explains Massimo Pegori. It operates on several fronts: on the one hand, it answers regulatory questions relating to the specific situation of individual employees, such as the notice periods to be applied in a given context, or how to verify a non-competition agreement entered into years earlier that may have changed in the meantime due to subsequent regulatory changes. On the other hand, the automated component independently analyses all documents relating to the employee, flagging those that may present issues in relation to the updated regulations.”

The key difference compared to a generic tool such as ChatGPT is that here we work with certified sources that we manage directly. Our team, with the support of employment lawyers and the INAZ research centre (which has been interpreting legislation as it is enacted for over 40 years), has selected and validated the sources on which the system is based. It is not simply a matter of copying and pasting the legislation: it is about understanding and contextualisation.

The other solution is called PayrolLens. It is a tool designed to help companies carry out payroll checks. The payslip production cycle always involves thorough checks to ensure there are no errors; these are almost never attributable to the software, but rather to the complexity of remuneration structures: every company has its own specific characteristics.

With PayrolLens, rather than following the old approach (manually extracting a thousand entries, creating an Excel spreadsheet, and cross-checking), the operator simply describes in natural language within the system which checks they wish to carry out.

The most interesting aspect is the design of the agent-based assembly line that Inaz is working on.

‘We envisaged it in truly collaborative terms,’ explains the director of innovation, ‘in the sense that, by choice, we have not implemented any fully automated processes based on artificial intelligence at any stage. So it is not the AI that decides to make a change to the document, but rather the AI is capable of performing its task autonomously, which is to analyse vast amounts of data very quickly and identify the areas where the user can intervene, whilst providing the user with the certified sources on which its assessment is based. So, when the user sees that the AI is suggesting a change, they also see, right there on the screen, all the sources that have been retrieved; they can analyse these point by point and, at that stage, validate the suggestion made by the artificial intelligence.’

The team consists of six people. It is deliberately small. Data engineers, data scientists, software engineers and AI specialists work within a lean structure, without complex hierarchies. Each week, time is also set aside for personal research projects – a sort of permanent laboratory where they can explore emerging technologies and new applications.

Initia will not develop software intended directly for the market. Its role will be to design AI solutions to be integrated into the group’s products, or tools to improve internal processes. This is a pragmatic approach that focuses less on making a splash and more on creating measurable value.

There is also a second factor that sets the project apart. INAZ regards its wealth of knowledge about the world of work and human resources as a strategic asset. At a time when many organisations are entrusting their data and expertise to external platforms, the company has chosen to establish an in-house unit to oversee the adoption of artificial intelligence.

The key word is control. Control over data, intellectual property and the ways in which algorithms are trained and used.

This is why Initia is presented not only as a technology laboratory but also as a guardian of ethical standards. Its stated aim is to develop AI systems that respect clients’ privacy and build on the wealth of expertise accumulated over almost a century of operation.

Rather than a Californian start-up, perhaps, Initia is more like a Renaissance workshop. A small team of specialists, diverse skills, ongoing research and the freedom to experiment. With one key difference: instead of painting frescoes, their raw materials are data, algorithms and artificial intelligence models.