Services
AI for law firms
Reading bundles, sorting post and writing first drafts all take up fee-earner hours that need a lawyer's judgement only at the end. That's where AI for law firms can save money. Your firm may also be paying for AI already, as a feature nobody has switched on. We make sense of it by analysing what your firm does now, task by task, then say where AI would pay and where it wouldn't. Where a step is worth building, the AI drafts or summarises from the matter in front of it, and a named person at the firm checks that before it goes anywhere. Client data stays out of free and public AI tools throughout.
When this comes up
You may have tried a general AI tool and then worried about where the client data went. Perhaps the partners want a clear answer to “what happens to our files”, or a supplier is pressing an AI add-on at renewal and you can’t say what it would earn.
Start with what you already pay for
Office software, practice management systems and research tools may now include AI features, or offer them as a paid add-on. We’d check what you’re entitled to and what terms it runs under, because a feature can be included in the price and still be unsuitable for client data. If an included feature does the job, you turn it on and write a short rule for using it, with nothing for us to build.
Where AI helps, by practice area
Conveyancing
Reading a lease and pulling the term, ground rent, service charge provisions and restrictions on alterations or subletting into a first-draft lease report. Comparing the seller's replies against the enquiries raised and listing the ones left unanswered. The fee earner reviews both before anything goes to the client or the other side.
Personal injury
Summarising a bundle of medical records into a dated chronology, each entry carrying its page reference, so the fee earner can go straight to the records that matter for causation. Sorting incoming post and email onto the correct claim.
Private client and probate
Drafting a first schedule of assets and liabilities from the bank, pension and share registrar letters received on an estate, with each item linked to the letter it came from.
Employment and family
Turning a client's long narrative email into a dated timeline the fee earner can use for an ET1 or a witness statement. Pulling figures from bank statements and payslips into a working Form E schedule, with the source of every number shown beside it.
Commercial
A first pass over a data room against your due diligence checklist, marking the documents a lawyer should read first: change of control clauses, unusual termination rights, unsigned or missing agreements.
Intake
Reading a web enquiry, working out which team it belongs to and writing a two-line summary for whoever picks it up. The AI sorts enquiries for your staff, and it never answers a member of the public's legal question.
Tasks that shouldn't use AI yet
Anything that would reach a client or the public without a person reading it stays out. So do decisions that need a lawyer: setting a limitation date, accepting a client’s risk, or advising on the position. We’d also hold back where checking the output takes as long as doing the task, or where the source material is too poor for the model to read, such as old scans and handwritten notes. Client data shouldn’t go near a tool until you know whose terms it runs under and who can see what. AI tools change, so we’d look at this list again later.
Does it save more than it costs?
Take one task and count four things:
- Time on the task now set against the time left after AI, with the checking still in.
- What you’d pay for the tool: per-user licences, an add-on for only the people who’d use it, or usage fees per document.
- The cost of building it into your workflow, if an included feature won’t do.
- Running it every year: watching output quality, and supporting it when a model or provider changes.
A task done a few times a month rarely clears that bar. If you’re not sure which of your tasks would, an AI and technology review goes through them one by one. You may not need anything built.
A person approves anything AI drafts
Every AI step we build ends with a person. The draft lease report sits in the fee earner's queue marked as a draft until they approve it. The chronology shows its page references so the reviewer can test any line against the record. Nothing the AI produces goes to a client, a court or another party until someone at the firm has read it and taken responsibility for it.
AI makes mistakes, and it can state a wrong thing with confidence. Steps where an error would be expensive get a closer review.
How a project runs
Findings come first, unless your firm has already decided which task it wants AI on. We then write an outline for one workflow, agreeing with the fee earners who do the task what a good output looks like. A working prototype then runs on sample or anonymised files, with your team marking every place it got something wrong. Once your supervising partner is content, we build the step into the systems your people already use. Testing on live matters happens only after release, in your own system and with your agreement.
We’ll need a fee earner to define a good result, anonymised past examples, and a partner to decide which steps need a second reviewer. To read what a project asks of your team, see how we work.
Ownership
The workflows, prompts, templates and review steps are built for your firm. We also build so the AI model underneath can be changed if a cheaper one comes along. Ownership of the code and your data is set out in your agreement.
Data and confidentiality
Client data never goes into free or public AI tools. We use AI services under business terms that exclude your data from model training, and tell you which ones and where they process it.Data is shared only with the hosting and AI providers named in your agreement, and we intend to host in the UK by default, or the EU if your firm prefers. Nothing we build is trained on your clients’ data, and each request and its result is logged against the matter.
A step only reads the matters the user in front of it is allowed to see. It’s designed to respect your information barriers, and you’d test that. Your firm stays responsible for its compliance and for supervising the work its people sign off. The security and data page has the detail.
The integrations page lists the systems we can connect to.
Questions
Questions firms ask
Does the AI give legal advice?
No. It drafts, summarises and sorts work for your staff, and a person at your firm checks every output.
Will our client data be used to train an AI model?
No. Our AI services run under business terms that exclude your data from training.
We already use Microsoft 365. Do we need anything else?
Possibly not. Check what your plan includes and the terms it runs under. If that falls short, Microsoft 365 offers API access, so we can build AI steps under the permissions your staff already have.
How much time would AI save us?
We don’t quote a figure, and Basys is new, so we have no results from other firms. Time one task on anonymised files, checking included, and compare.
Pick one task to try
Is there a reading or drafting task your fee earners would gladly hand to a checked first draft? Talk to us about that one task.