AI where it helps, and the boundaries that keep it honest

We use AI tools ourselves every day, and they are incredibly powerful. Give a good lawyer the right tools and they can review faster, draft faster, find things faster and get through an extraordinary amount of work. We saw that pretty quickly.

But as we used them more, we kept coming back to a different question: how does this actually help the rest of the organisation get through the routine?

Making lawyers faster is valuable, but it doesn't necessarily solve the underlying problem. The business still sends contracts and requests to legal. Lawyers still sit in the middle of the process. The queue might move faster, but there is still a queue.

That led us to experiment with putting more capability directly in the hands of the business. And AI makes that remarkably easy. You can give someone access to a model, connect it to a playbook and tell it how the organisation generally approaches a particular kind of contract or legal issue. For a lot of tasks, the results can be very good.

What we found, though, is that there is an important difference between giving AI instructions and building a governed process.

A playbook might say what an acceptable liability cap looks like. But who can approve a departure? Does finance need to be involved above a certain value? Who has authority to sign? What happens when something falls outside the playbook? What gets recorded so someone can understand six months later why a decision was made?

Those aren't really AI questions. They are governance questions.

That changed how we thought about the problem. Instead of starting with AI and trying to add controls around it, what if we set the guardrails first?

For routine self-service, we still think automation is the gold standard where the rules matter. You can define the pathway, approvals, permissions and escalation points, and know that they will be followed. The interesting part is what happens when you put AI inside that structure.

AI can dramatically reduce the time it takes us to build those automations. It can make them much easier for people to use, because someone can simply explain what they are trying to do rather than learn a process. It can read documents, extract information, identify issues and help people move through the workflow. And when something falls outside the guardrails, it can recognise that the work needs judgment and bring in a lawyer.

We use AI on the lawyer side too. In many ways, that is where its value was most obvious to us from the beginning. Lawyers can review, compare, analyse and draft faster. The difference is that by the time something reaches them, the routine process can already have happened and the relevant context can come with it.

That is why we're wary of the idea that self-service means handing the business an AI tool, adding a “skill” or a playbook and leaving them to do the rest. It is powerful, but it doesn't solve the whole problem we set out to solve.

Our experience has pushed us towards a fairly simple model: use AI where intelligence helps, automation where the rules matter, and lawyers where judgment is the point.

The interesting bit is what happens when all three work together.