Judgment calls AI cannot make for you
By the ChamberOne team 4 minute read
A tool can summarise a bundle in minutes, list the issues in a contract and draft a letter. It cannot sit across from a nervous client and decide how much to say. It cannot hear a witness hesitate. It does not bear the consequences of a settlement that turns out too low. The areas where experience still decides share one feature: the right answer depends on people, timing and risk that no document fully captures. This article names several, explains why a tool does not settle them and suggests how to use AI for preparation while keeping the decision yours.
Key points
- Judgement calls have incomplete facts, reasonable alternatives and a person who owns the consequences.
- Use AI to organise options and test arguments, not to choose when to settle or whom to call.
- Witness credibility, risk appetite and difficult client conversations depend on what is not written down.
- Write your own view first and record the decision and your reasons in your own words.
- Give juniors real decisions to make and explain your reasoning so the skill is passed on.
What makes a call a judgement call
A judgement call has three features. The facts are incomplete. Reasonable people could decide differently. And someone must own the consequences. Drafting a standard clause usually lacks these features. Deciding whether to accept an offer has all three.
Language models are good at pattern and summary. They are weak where the decision depends on information that was never written down, such as the client's real tolerance for risk or the mood in the room. They can list considerations. They cannot weigh them for a particular person in a particular moment.
- Pattern and summary
- Listing considerations
- Information that was never written down
- The client's real tolerance for risk
- The mood in the room
- Weighing considerations for a particular person at a particular moment
Settlement timing
Whether to settle, when and at what figure rests on the strength of the case, the cost of going on, the client's finances and temperament, the other side's likely stance and the stage of proceedings. A tool can organise the issues and test the arithmetic. It cannot sense whether the other side is close to moving, and it does not know that your client will lose sleep over a trial.
Use AI to prepare: a summary of strengths and weaknesses, a draft comparison of outcomes. Then do the part that matters yourself, which is the conversation with the client about what they can live with.
Witness credibility
Judging whether a witness will be believed involves demeanour, consistency under pressure, how they react to a difficult question and how they will look to a particular judge or tribunal. A tool working from a written statement sees none of this.
It can help in preparation, for example by listing inconsistencies between statements or suggesting questions an opponent might ask. But whether to call a witness, and how to present them, depends on what you have seen and heard.
Risk appetite
Two sensible clients can take opposite decisions on the same facts, because one can afford to lose and one cannot, or one values certainty more than money. Risk appetite belongs to the client, and the lawyer's job is to draw it out and explain the choices.
A tool can lay out options and likely consequences. It cannot know that the client is closing a sale next quarter and needs the dispute gone. Ask. Record the answer. Revisit it when circumstances change.
Difficult conversations with clients
Telling a client that their case is weaker than they hoped, that costs have grown or that a mistake was made is a human task. The words matter, but so do tone, timing and the willingness to stay in the room. A drafted script may help you prepare. It will not replace the conversation, and a client who senses a script can lose trust.
Suppose a litigator must tell a client that the evidence has turned against them. She uses a tool to produce a short summary of the changes and a list of options. Then she phones the client, explains in her own words, listens and follows up in writing. The tool helped her prepare. It did not carry the message.
Ethical and professional judgement
Spotting a conflict, deciding whether a client's instruction crosses a line, deciding when to withdraw and when to report something are decisions for lawyers to take in light of their professional duties. Check the guidance from your regulator or bar. A tool cannot carry the duty, and a lawyer who says 'the tool did not flag it' has not answered the question.
How to use AI without handing over the decision
- Ask it to set out options and the considerations on each side, not to choose.
- Ask it to find weaknesses in your argument, then judge them yourself.
- Use it to prepare questions and scenarios, not to decide what to do.
- Write your own view before reading the tool's, so that you are not anchored by it.
- Record the decision and your reasons in your own words.
- Check any factual claim it makes against the documents.
- Ask it for options and the considerations on each side
- Ask it to find weaknesses in your argument, then judge them yourself
- Write your own view before reading the tool's
- Record the decision and your reasons in your own words
- Ask it to choose for you
- Accept a factual claim without checking it against the documents
Protecting the skill
Judgement develops by making decisions and living with the results. If juniors never face real choices because a tool pre-digests everything, they will not build the skill. Seniors can help by giving juniors decisions to make, with support, and by explaining their own reasoning aloud. 'Here is why I would settle now' teaches more than a polished memo.
Firms can also build review into the process: after a matter closes, a short conversation about which decisions mattered and how they were made. That habit passes on experience that no tool holds.
Judgement under time pressure
The temptation to defer to a tool is strongest when time is short. A confident paragraph on screen at six in the evening looks like permission to stop thinking. Build a small habit against it: before accepting any recommendation, say aloud what you would do without the tool and why. If the two differ, find out why before you decide.
A realistic view
None of this makes AI useless. It is a good assistant for the preparation that lets a lawyer decide better and sooner. The point is narrower: the decision, and the responsibility for it, stays with a person who knows the client and the matter. Lawyers who understand where that line sits will use the tools better than those who defer to them or those who refuse to look.
This article is general information about the law, legal practice and technology, current when it was published. It is not legal advice. Laws and guidance change: check the sources linked above and take advice on your own circumstances from a lawyer qualified where you are.
Read next
Training junior lawyers when AI does the first draft
If routine drafting shrinks, juniors need other ways to build judgment. Practical exercises that keep the learning in the work.
5 minute readThe skills lawyers need to work well with AI
Verification, precise instruction, source reading and client judgment: what to practise and how.
5 minute readLaw students and AI: learning to think before you prompt
AI can be a sparring partner for a law student, but only if you do the thinking first.
4 minute read