A staged plan beats a single big purchase. Start with general AI tools on monthly seats, prove daily use with two committed people, and only then decide whether a specialist legal platform is worth the price.
The text of this article was generated using artificial intelligence from the author's own published analysis and checked against the primary sources listed at the end. Editorial responsibility for its publication rests with Philippos Aristotelous. Disclosed under Article 50(4) of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689).
Don't buy a specialist legal AI platform first. Start with two or three monthly seats of a general tool — Claude Team, ChatGPT Business or Gemini — at roughly €17–€22 a seat.
Keep the trial away from client files for the first few months, hand it to two committed people rather than the whole floor, and build real workflows around tasks the firm already does.
Only once daily use is proven should a larger firm weigh a specialist platform such as Harvey or Legora, at a materially higher price.
Since my last piece on AI strategy, one question keeps coming up in conversations with managing partners: which tool should we actually buy, and how do we stop it turning into another subscription nobody opens by October? It's a fair question, and the honest answer has less to do with the tool than with the sequence you follow to get there.
01 Why do most law firm AI purchases fail?
A well-run firm's instinct is to evaluate the market, negotiate an annual contract and switch everyone on at once. That instinct serves the firm badly here, because it means buying a tool before anyone has worked out which of the firm's actual tasks it's meant to do.
The failure point usually isn't the signature on the contract — it's somewhere around the sixth week, once the initial curiosity has faded, the one person who was genuinely excited about it has disappeared into a busy court list, and a partner or two have quietly gone back to doing things the old way. A twelve-month commitment has no way of catching that: the direct debit keeps going out long after the login stops being used.
02 Are specialist legal AI platforms like Harvey and Legora worth it?
The sales demo for a specialist legal AI platform is genuinely impressive — a whole data room summarised in minutes, a live dashboard across every open matter. What happens after the contract is signed tends to look rather different. A large share of what got demonstrated turns out to solve problems the firm never actually had at its current size, while much of the day-to-day work it does need can already be done with a frontier general-purpose model — Claude, ChatGPT or Gemini.
That's not because the general models are secretly as capable out of the box as the specialist ones. It's that they respond well to being shaped: clear written instructions, prompts built around the firm's own precedents and templates, and workflows tuned to how its lawyers actually operate. None of that involves retraining anything — it's a design exercise, not an engineering one.
That doesn't mean the specialist platforms are a poor choice across the board. A larger firm handling serious document volume, with real integration requirements, can get genuine value from one. The point is that this is a decision to make once the firm has evidence of its own needs, not one made on the strength of a demo.
There's also a test almost no demo ever runs: how the platform handles Cyprus itself. Cyprus law, regulations and case reports sit across a scattering of different sites, in a mix of formats — PDF, plain web pages, material buried behind chains of links, and the odd source that a person clicks through in a second but that automated tools regularly cannot reach. A platform tuned for the English or American market can be genuinely strong there and still fall short here. Put a real Cyprus question to it before signing anything, and see whether it actually locates the provision, opens the real text, and cites the correct section or judgment rather than something adjacent.
| General AI tools | Specialist legal platforms | |
|---|---|---|
| Examples | Claude Team, ChatGPT Business, Gemini | Harvey, Legora |
| Cost per seat | ~€17–€22/month | Materially higher — enterprise pricing, usually negotiated per firm |
| Best suited to | Any size firm, proving what AI can actually do for it | Larger firms with heavy document volumes and integration needs |
| Setup effort | Low — buy seats, write prompts, start using it | Higher — onboarding, integration, often a pilot period |
| Cyprus-source reliability | Varies — test with a real Cyprus legal question first | Varies — test with a real Cyprus legal question first |
| When to buy | First — always | Only after the firm knows exactly what a general tool can't do |
A tool bought before anyone knows what it's for is a tool bought the wrong way round.
03 The six-step path to AI adoption for a Cyprus law firm
This is roughly the sequence I walk law firm leaders through, whether in a consulting engagement or in training: start with what the firm genuinely does day to day, hold the cost down until the tool has proved itself, and build something fitted to the firm rather than paying for a long list of features nobody asked for.
04 What do Claude Team, ChatGPT Business and Gemini cost in 2026?
Current list pricing, checked 17 September 2026: Claude Team and ChatGPT Business both sit at roughly US$25 (about €22) per standard seat a month on monthly billing, dropping to around US$20 (about €17) a seat on annual billing, from a two-seat minimum. Once a champion is genuinely pushing the tool and starts hitting usage limits, a premium seat currently runs to about US$125 (about €108) a month on monthly billing. These figures move — confirm current rates directly with the vendor before committing.
05 Why champions matter more than the tool
Two lawyers who've spent a few months genuinely working the tool into the firm's own precedents and habits will outperform an entire floor handed a platform nobody has bothered to configure. There's a visibility effect too. Once an associate watches a colleague they respect turn out a solid first draft in ten minutes, and watches a partner review and sign it off properly, using the tool stops feeling like an experiment and starts feeling like how the work gets done.
In practice, the champions rarely get there alone. Someone who already implements these tools inside law firms usually needs to help turn a single recurring task — a first-draft engagement letter, a long correspondence summary — into an actual workflow: a written prompt, reusable instructions, and one named person responsible for checking the output. A licence on its own rarely gets a firm there.
06 What does "ring-fencing" the trial actually mean?
Ring-fencing a trial doesn't mean keeping the firm in a permanent holding pattern. It means the firm gets to decide, deliberately and on its own timeline, exactly what the tool is eventually allowed to touch — which categories of data, whose client accounts, anonymised to what level, and reviewed by whom. Those are decisions for a written AI policy, not an afterthought.
They also have a legal backdrop. Article 4 of the EU AI Act has required AI literacy training since 2 February 2025, so whatever the policy says needs to be backed by training that actually happened, not a sentence that reads well in a document nobody has opened.
- 1Which AI tool should a Cyprus law firm buy first?A general business tool — Claude Team, ChatGPT Business or Gemini — on two or three monthly seats. Not a specialist legal platform, and not a firm-wide rollout.
- 2How much do Claude Team, ChatGPT Business and Gemini cost per seat?As of 17 September 2026, roughly €22 a month on monthly billing or €17 on annual billing, from a minimum of two seats. A premium seat runs to about €108 a month.
- 3When is a specialist legal AI platform like Harvey or Legora worth the price?Once a firm — usually a larger one with heavy document volumes and integration needs — has proven what a general tool can and can't do, and can test each specialist feature against work it already does well.
- 4Does general AI handle Cyprus law reliably?It varies by platform and source. Test it on a real Cyprus question before relying on it — check whether it finds the law, opens the actual text, and cites the right section or judgment.
- 5What does the EU AI Act require before staff start using AI tools?Article 4 of Regulation (EU) 2024/1689 has required AI literacy training since 2 February 2025. It needs to be real training, recorded and understood, not a clause in a policy document.
The firms that go furthest with AI in Cyprus are unlikely to be the ones who bought the most expensive platform on day one. They're more likely to be the ones still opening theirs in week sixty, not week six.
Sources & notes: Adapted from the author's own analysis, first published as "Choosing AI for a Cyprus law firm: Start small, then scale", CBN, 28 September 2026. Pricing: Claude Team and ChatGPT Business list pricing as published 17 September 2026. EU AI Act: Regulation (EU) 2024/1689, Article 4 (AI literacy obligation, applicable since 2 February 2025).
This article was produced with AI assistance; see the disclosure above.
This article is published for general information and education. It is necessarily general, is not legal, tax, accounting, financial or other professional advice, and should not be relied on in place of advice tailored to your circumstances. Reading it creates no client or advisory relationship. Reasonable care is taken in preparing what is published here, but laws, regulations and guidance change, and no warranty is given as to continuing accuracy or completeness. Obtain appropriate professional advice before acting, or deciding not to act, on anything set out here. To the maximum extent permitted by law, Philippos Aristotelous and Philarist accept no liability for loss arising from reliance on this article. Nothing here excludes or limits any liability that cannot lawfully be excluded or limited.
Videos
Watch: choosing and staging AI adoption in a law firm
Swipe for more videos →