AI applied to management software

Source: NT+ Fisco – Il Sole 24 Ore
Edited by Fabio Giordano, Comitato tecnico AssoSoftware.

An increasing number of professionals are incorporating AI into their daily work—often using it in very different, and sometimes highly original, ways. On social media, within online spaces dedicated to accountants and tax intermediaries, there are now numerous discussions where professionals share their experiences with AI and the benefits it offers in their line of work.

Admittedly, many of the comments are banal and full of clichés. But there are some interesting ones, too.

The use of AI

Then there are the “super-technicians,” who speak with ease about LLMs and model training, and who cite the most advanced and specialized versions of Claude, ChatGPT, Gemini, Copilot, and the like.

Then there are the “hyper-operational” types who try feeding financial statements into AI, seeking to delegate the preparation of client reports on business performance to the technology. Some even go as far as training models and creating agents capable of operating with complete autonomy. And then there are those who use it for “psychotherapeutic” purposes. A few weeks ago, one professional mentioned in a post that they use AI to draft emails: they write down exactly what they are thinking, then ask the AI ​​to rewrite it all in a more “diplomatic” tone.

All of these are fulfilling activities that are also useful in terms of professional growth. But what about the results?
Well, you have to be very careful about what you do, and above all about how you do it.

The risks of using AI

The first issue concerns the confidentiality of the data fed into AI systems—systems which, unless they are “private” (such as those integrated into management software or publishing solutions), effectively operate almost exclusively online in the public sphere, with data frequently ending up on servers located outside the EU.

Even in the United States, the phenomenon of employees using external AI tools—other than the one authorized by their company—persists, seemingly unchecked, despite increasingly thorough training and the imposition of severe disciplinary measures, up to and including termination. This is particularly concerning given that only the company-authorized tool guarantees the confidentiality of corporate data.

In this way, contract data, confidential documents, trade secrets, lists of clients and suppliers, and so on, are effectively “gifted” to AI developers—who could then potentially resell them in some form. And that’s not to mention employees who, in some cases, even delegate the management of their corporate email to AI.

Why do they do it? Out of a lack of awareness, for the sake of convenience, or because they find their personal AI tools more comfortable to use than the ones provided by their company—thereby saving time and effort. The trouble is that these AI systems store everything, and they do so “forever.” Consequently, that data becomes the property of major overseas tech companies. What is happening in Italy? Much the same thing—though, fortunately, our digital literacy regarding AI still lags somewhat behind, which is why the damage reported so far has remained, all things considered, quite limited.

The difference between AI and “old” algorithms

The second issue concerns the very functionality of AI, which—as is well known—operates based on probabilistic criteria; unlike traditional algorithms, it can provide vastly different responses to the exact same query. Consequently, using a general-purpose AI to analyze financial statements or draft the explanatory notes for an EU-compliant balance sheet offers no guarantees, and the results it produces must therefore be carefully verified. Unfortunately, the transition from the tried-and-true algorithm to more “powerful” AI comes at the cost of the certainty that traditional computing has provided us with until now.

Let us now try to draw some conclusions.

For a professional, given the current state of technology, delegating the processing of clients’ accounting and tax data—extracted from their management software—to an AI via external solutions not endorsed by their own software provider is not an advisable approach, except in very specific cases and provided that the utmost precautions are taken.

Moreover, this is a highly individualistic approach—one that is eerily reminiscent of those 1990s scenarios where a professional, acting as an amateur systems analyst, would commission the development of their own management software from a trusted programmer, simply because they wanted it exactly as they had envisioned it.

Only to find themselves in difficulty a few years later, when the complexity of tax declaration forms increased and their trusted programmer became increasingly hard to reach.

AI in management software

On the other hand, although the software companies associated with AssoSoftware are working on it with great commitment, the application of AI to management software is still in its early stages; while its features are being progressively enhanced month by month, it is still far from full maturity. This does not represent a lag in the development of management software, but rather the need to manage this technological evolution in a proper and structured manner, taking into account:

• of constantly evolving AI technologies, which make choosing a reliable technology partner a complex task;
• the need to find solutions that reconcile the requirements for data confidentiality—which, in cases where data must be “transferred” to non-national entities for processing, can no longer be guaranteed solely through contractual means as they once were, due to increasingly intrusive interference from the institutions of even allied nations;
• the need to ensure maximum reliability for AI analyses/outputs, minimizing errors—so-called “hallucinations”—as much as possible;
• the need to ensure high performance, which requires dedicated and very expensive hardware if one chooses to operate with proprietary installations rather than utilizing the data centers of multinational AI companies;

In any case, it is important for professionals to explore the potential of new technologies right away—not least so they can provide their software providers with valuable input regarding their specific operational needs. At the same time, it is advisable to make the most of innovative yet well-established tools that have long been available, such as management control systems or business intelligence; these can often deliver excellent results, even without the use of AI.

Stakeholders, including technology providers, standards organizations and end-users, must work together to develop solutions that maximize the value of RFID and IoT systems.

Such partnerships will be instrumental in realizing the full potential of IoT-enabled supply chains.

Read the original article >