Legal AI spending is easier to argue about than to measure, so a set of audited half-year figures is worth more than a season of predictions. RELX, the owner of LexisNexis, published its results for the six months to 30 June on 23 July, and the legal division grew faster than the rest of the group. For a small firm the interest lies less in one publisher's fortunes than in what the numbers reveal about how other firms behaved when they came to spend.
What the figures show
Legal revenue reached £959 million against £900 million in the same period last year, a rise of 7 percent in sterling and 10 percent on the underlying measure the company uses to strip out currency movements and portfolio changes. Adjusted operating profit rose from £188 million to £208 million, up 13 percent underlying, and the division's margin improved from 20.9 percent to 21.7 percent. The wider group reported underlying revenue growth of 7 percent, so legal ran ahead of it. The company puts the improvement down to adoption of AI-enabled research and analytics tools, and records renewals and new sales holding firm across its main segments.
The money stayed close to the work
Set that against the mood earlier in the year. When the large AI providers began building for legal work directly, a common reading was that firms would drop their research subscriptions and run their research through a general assistant instead. Six months of trading points the other way. Firms bought the AI that arrived inside the product they were already using, and they renewed. The reasons are practical rather than sentimental. The content is licensed and the citations lead somewhere you can check. The tool opens where the work is already sitting. One supplier to supervise is a lighter burden than three, and the partner responsible for compliance has one contract to read rather than several.
One publisher's accounts settle nothing about which tool suits a particular firm. Thomson Reuters, the specialist legal platforms and the general assistants report separately and compete for the same budget, and a firm doing high volumes of routine drafting will reach a different answer from one running complex litigation. What the figures do show is that the incumbent products were not abandoned, which removes the fear of being stranded on a dying platform from your list of reasons to switch.
Your renewal is the AI decision
For a firm of two to twenty fee earners this changes where the decision happens. The AI is being folded into subscriptions you hold already, and its cost arrives inside the renewal quote rather than as a separate purchase to be approved. Sign the renewal without reading it and you have made an AI buying decision by default, at whatever price the uplift carries, with no test of whether anyone in the office uses the thing.
Ask the supplier for usage data before the renewal date, broken down by person and by month. Most will provide it, and the answer is often uncomfortable. Ask what the increase is against last year and what has been added to earn it. Check whether the AI answers cite authorities you can open and verify, because an answer you have to research again saves nobody any time. Establish what happens to your searches and your matter data, and whether either trains anything. Then weigh the cost of moving, which is rarely the licence fee and almost always the retraining, the new supervision routine and the months of slower work while people learn.
What a firm should do now
Put your renewal dates in one list with the annual cost against each, then get the usage figures for the three months before each date falls due. Where a tool is used daily and the output holds up, renew and stop worrying about the market. Where the AI features sit untouched, take them out of the package or move to a smaller one, and put the difference towards training the people who would use them. A subscription renewed out of habit is the most expensive form of legal AI spending, because nothing about it has been measured and nobody has had to defend it.
The divisional figures are set out in RELX's first half results presentation, published on 23 July 2026.
If a renewal is coming and you cannot say whether the AI in it is being used, a short review of your subscriptions and their real usage will tell you before you sign rather than a year afterwards: start with a conversation.