A Cyprus law firm should adopt AI in stages rather than through one big purchase. Buy a few monthly seats of a general business tool such as Claude Team, ChatGPT Business or Gemini. Keep the trial fenced off from client files and firm servers. Give it to one or two lawyers who want it to work, and design real workflows around them. Only once daily use is proven should the firm consider a specialist legal platform such as Harvey or Legora.
Since my last piece, the question I get most from managing partners is a practical one. Which tool do we buy, and how do we stop it becoming another subscription nobody opens by October?
Why the big purchase usually fails
The instinct in a well-run firm is to compare platforms, sign an annual contract and roll it out to everyone. With AI, that order is backwards. The firm is choosing a tool before it knows which of its own tasks the tool will do.
Adoption doesn't fail at the purchase. It fails around week six, when the novelty has gone, a court week has taken the one enthusiastic user away, and a partner has quietly said he will carry on as before.
Don't be dazzled by the demo
A specialist legal AI platform gives a beautiful demonstration: a data room reviewed in minutes, a dashboard for every matter. Then the contract is signed and the firm lands in reality. Most of those features answer problems the firm doesn't have at its size. Much of what it does need turns out to be achievable with the frontier general models it could have bought by the seat: Claude, ChatGPT or Gemini.
Not because they are magic out of the box, but because they can be shaped. With well-written instructions, reusable prompts and skills built around the firm's own templates, and workflows calibrated to how its lawyers actually work, a general model can do a surprising share of what the specialist platform promised. No retraining of the model is involved. What's involved is careful design.
None of this makes the specialist platforms a bad buy. For a large firm with heavy document volumes and complex integration needs, they can earn their price. But that is a decision to make once the firm knows what it needs, not in the glow of a demo.
There is a local test too, and it is one demos rarely run. Cyprus legislation, regulations and judgments are published across several websites, in mixed forms: PDFs, web pages, documents reached through chains of hyperlinks, and some sources behind click-through screens that a person passes in a second but software often cannot. A platform built for the English or American market may be excellent on its home sources and struggle to reach ours. So before paying, ask it a real Cyprus question and check whether it finds the law, opens the actual text, and cites the right section or judgment.
This is how I work with law firms today: start from the ground up, with the tasks the firm actually does, keep the cost low while use is proven, and build something genuinely custom-made rather than rent features nobody asked for.
A phased path to AI adoption for law firms in Cyprus
1 Buy small and monthly. Start with a general business plan that has admin controls and a commitment not to train on your data. As of 17 September 2026, Claude Team and ChatGPT Business both list roughly US$25 (about €22) per standard seat per month on monthly billing, or about US$20 (about €17) on annual billing, from a minimum of two seats.
2 Fence off the trial. For the first months the tool gets no access to the firm's servers, document management system or client files. Lawyers work on templates, public law and anonymised facts, which keeps confidentiality and GDPR questions simple.
3 Choose champions, not volunteers. Pick one or two people who are curious, driven and willing to push the tool hard every day. If they hit usage limits, give them a premium seat, which currently lists at roughly US$125 (about €108) a month on monthly billing.
4 Design workflows, not experiments. Working with someone who implements these tools for law firms, the champions should turn specific tasks into repeatable workflows, such as first drafts of engagement letters or summaries of long correspondence. Each workflow gets written prompts, reusable instructions and a named reviewer.
5 Widen access on evidence. Add seats only when a workflow is in regular use and a colleague has asked for it. Count use, not impressions.
6 Graduate when the foundations hold. Larger firms with high document volumes may then justify a specialist legal platform such as Harvey or Legora, at a materially higher price. By then the firm can test each feature against work it already does well, and buy only what the general tools cannot match.
Why the champions matter more than the tool
Two capable lawyers who have spent three months shaping the tool around the firm's real templates will get more from it than a whole floor will from a platform nobody has shaped. Champions also make adoption visible. When an associate sees a respected colleague finish a first draft in ten minutes, and sees a partner sign it off after proper review, the tool stops being optional.
What ring-fencing does not mean
Fencing off the trial is not avoiding risk forever. It means the firm decides deliberately, and later, what the tool may touch: which data, on whose accounts, anonymised to what standard, and checked by whom. Those decisions belong in a written AI policy. Under the EU AI Act, the AI literacy obligation has applied since February 2025, so the training has to be real, not a line in a policy.
This is the sequence I take law firm leaders through in my consulting engagements and trainings. The firms that get furthest with AI in Cyprus won't be the ones that bought the most expensive tool first. They'll be the ones still using theirs in week sixty.
*Philippos Aristotelous is a former Partner at Cyprus's largest law firm and now advises law firms on legaltech and AI implementation, practice management and leadership, working across Cyprus, Greece and the Baltics. His website is philarist.com.
Also read: Would your legal advice survive an AI check?
