AI in the law firm: what works and what's hype

Few phrases have worn out faster in recent years than “artificial intelligence”. In the legal sector the noise is especially loud: every week brings a tool that promises to draft pleadings on its own, predict the outcome of a case, or replace hours of study with a single click. It is worth separating what the technology genuinely does well from what is, plainly, hype.

The distinction is not academic. A firm that overestimates what AI can do ends up reviewing defective work or, worse, not reviewing it at all. A firm that underestimates it keeps losing hours on mechanical tasks a machine settles in seconds. The right balance demands honesty about real capabilities.

Where AI genuinely saves time

There are four tasks where generative AI delivers tangible, verifiable value in a firm’s daily work.

• Summarising long documents. A fifty-page deed, a dense contract, a bulky administrative file — all condensed in minutes. The lawyer still reads the original when it matters, but arrives at it oriented.
• Extracting data from documents. Names, dates, tax numbers, cadastral references, specific clauses: AI locates and structures them from deeds, passports or land-registry extracts the firm already holds. This is assisted transcription, not guesswork.
• Preparing first drafts. A routine filing, a client letter, a standard demand. AI produces a starting point the professional corrects and makes their own. The draft is not the finished product; it is the scaffolding.
• Flagging regulatory impact. When the official gazettes publish something matching what the firm follows, AI flags which matters may be affected and why. It is a radar, not an opinion.

In Mandato, these functions exist because they solve real problems, not because they look good in a demo. The official gazettes watch is a good example: the tool flags the publications matching your list and which matters may be affected, but the reading and the decision to act stay with the lawyer.

Where the hype begins

The same enthusiasm that drives the useful functions feeds promises that do not hold up. They deserve scepticism.

AI does not replace legal judgement. It can order information and suggest wording, but it does not weigh interests, does not assess the strategy of a case, and does not take responsibility. Any tool implying otherwise is selling an illusion.

AI without context is confidently wrong. A model that answers without knowing the matter can invent case law, cite repealed articles, or state something false with full assurance. The phenomenon has a technical name, but for the firm what matters is the consequence: convincing, erroneous text. The only defence is human review and a system that works on the real data of the matter, not on a vacuum.

Not everything labelled AI needs it. Many functions presented as artificial intelligence are, in truth, simple automations. There is nothing wrong with that, but there is no reason to pay an innovation premium for a rules-based template.

The principle that ties it together: the human in the loop

The difference between a useful tool and a dangerous one lies not in the model but in the design. Responsible AI presents its work as a reviewable proposal, shows where its information comes from, and always leaves the last word to the professional. The lawyer signs; the machine assists.

This principle has a practical consequence in software. AI that lives inside the matter — that knows the client, the documents, the deadlines and the history — produces results anchored in concrete facts. A disconnected chatbot, however powerful the model behind it, answers into thin air. The first reduces error; the second invites it.

That is why in Mandato artificial intelligence is not a separate tab. It works on the open matter: it summarises the document sitting in the case, extracts the data from the deed the client already uploaded, drafts from what is on record. Context is not an ornament; it is what separates a reliable answer from a merely plausible one.

How to evaluate a legal AI tool

Before adding any AI function to your practice, three questions serve as a filter.

• What data does it work on? If the answer is “the internet in general”, be wary. If it is “the matter and the documents you control”, it is on the right track.
• Does it show its reasoning and its sources? A tool that does not let you verify where a statement comes from passes to the professional a risk they cannot take blindly.
• Who signs? The correct answer is always the lawyer. AI that promises to decide for you does not save work: it adds a responsibility that is hard to control.

Properly understood, artificial intelligence is a powerful lever for the small and mid-sized firm: it cuts the time spent on the mechanical and frees hours for what truly needs a lawyer. But it is a lever, not a substitute. The day a tool promises you the latter, you will know it has crossed the line into hype.

To see how this approach translates into daily work, start with Mandato’s features and judge it by what it does, not by what it promises.