Financial services adopted AI twelve to eighteen months earlier than the legal sector. For law firms that means one concrete advantage: we can see what worked, what didn't — and start where finance is today, instead of walking the same curve through the same mistakes.
Where financial services started — and where they are now
Banks, insurers and asset managers in 2023–2024 typically deployed some generic AI tool — an internal chatbot, ChatGPT Enterprise, or a packaged offering from a large advisory firm. The premise was "put AI on the intranet and people will improve." The result was predictable: usage tapered off after three months, outputs were shallow, and no one could measure ROI.
What followed was a second wave — and it looks completely different. Instead of a generic chatbot, financial institutions today deploy custom skills built on a specific workflow: a skill for credit-risk analysis, a skill for KYC review, a skill for regulatory reporting. Instead of "open the chat," AI is integrated directly into Excel, Word, Outlook and internal systems. Instead of one helper, agentic workflows run research, drafting and review in parallel.
The twelve months between "experiment in chat" and "AI in production workflow" cost the financial sector tens of millions of euros — in consultants, abandoned pilots and internal frustration. The legal profession does not need to pay that price.
Same start, one extra advantage
Law firms today are exactly where banks were eighteen months ago. Some have Claude Pro for a few lawyers, some ran a ChatGPT Enterprise pilot, some bought a deck from a large IT firm. The frustration is identical: "Pretty, but what do we actually do with it?"
The difference is that the legal sector now knows the dead ends. Generic AI thrown onto the intranet without integration into a specific workflow does not work — firms don't need to test that themselves; calling a colleague in bank risk management is enough. The only viable path is custom implementation around one recurring task (NDA review, court-file summary, draft client email) built on Enterprise API with a DPA, EU residency and a no-training clause.
This isn't theory — it is a lesson someone has already paid for. For firms starting today, six months of blind walking can be skipped entirely.
Conservatism is, in this context, an asset
The legal profession is often called conservative. In AI adoption that is not a weakness — it is a safeguard. Lawyers don't ask "is it cool?" or "what are they doing in Silicon Valley?". They ask: does it work, is it secure, who is accountable? Those are exactly the right questions for a technology that handles client data under legal privilege.
Banks that "experimented" in the first wave had to walk back after several incidents (prompt leaks, hallucinated regulations, fabricated analytics) and impose strict governance — RBAC, audit trail, mandatory human review, data segmentation. Lawyers ask for that governance from minute one. The implementation is therefore built correctly from the start.
Where to begin: not with the biggest change, but with one task
The most common mistake in the first wave at large banks was an over-ambitious initial rollout. "Deploy AI across all branches." "Our goal is a 30 % cost reduction in twelve months." That approach reliably ends badly.
For a law firm the inverse makes sense: pick one recurring task the team performs every week, and build the first AI skill around it. NDA review against your checklist. Court-file summarization. A draft client email from notes. Once one skill works and the team uses it, you build the second. Six months later, five skills cover 70 % of routine matters.
AI implementation that works looks boring. It begins with one task, one firm, one documented saving — and only then grows. Whoever begins the other way around ends with a deck and no application.
The legal profession has an advantage here it deserves to use. It isn't a bonus for arriving late — it is a bonus for the fact that someone else has already paid the price for the mistakes you can now avoid.