Why do AI tools not improve productivity on their own?

What is the difference between a tool and a process?

A tool is something a person can use. A process is the way a piece of work gets done: the inputs it needs and where they come from, the steps in order, the template the output follows, who hands what to whom, what gets checked and by whom, and who is responsible for the result. A process exists whether or not anyone has written it down. Every firm has one for its monthly reports, its proposals and its file reviews.

An AI licence changes the tools. It does not touch the process. The fee earner who produced last month's report by hand still opens the same folder, still collects figures from the same places, still copies the same template. Now there is also a chat window, which may help with a paragraph or two. The job is the same job.

Why does a licence leave the work unchanged?

Look at what happens when a fee earner tries to use a general AI assistant on a real piece of recurring work.

• The inputs are scattered. The assistant only knows what it is given. Someone still has to find the figures, the notes and the previous version, and paste them in.
• Every person asks differently. Ten fee earners write ten different instructions and get ten different styles of output. The firm's standard becomes whatever each person remembers to ask for.
• The output lands outside the work. The draft appears in a chat window, not in the firm's template. It has to be copied across and reformatted, which eats much of the time it saved.
• Review is improvised. Nobody has defined what to check in an AI draft, so careful people either check everything, which saves nothing, or avoid using it on client work.
• It depends on remembering. Using the tool is optional and relies on habit. Under deadline pressure, people do what they know.

None of these is a fault of the tool. They are the gaps between a capable tool and a job that was never redesigned to use it.

What does the same job look like when it is rebuilt?

Take a fee proposal as a general illustration. With a licence alone, a partner opens the last similar proposal, asks the assistant to rewrite the approach section for the new client, pastes the result back, updates the team biographies by hand, checks the terms and formats the whole document. Some time is saved on one section. The rest is unchanged.

Rebuilt, the job starts differently. The partner records the essentials of the opportunity in a short, fixed form. The process draws the right sections from the firm's own library of approaches, experience and team profiles, drafts the tailored parts in the firm's language, and assembles everything into the house template. The draft arrives with the new material highlighted. The partner's job becomes reading, adjusting and approving, and a named person signs off before anything is sent. The same assistant may sit underneath, but it now does a defined part of a designed process, the same way every time, for everyone.

That is the difference. The tool did not get better. The job was rebuilt around it. More examples are on the pages for each process that can be rebuilt.

Why does more training not close the gap?

Training makes individuals better at using a tool. It cannot remove scattered inputs, missing templates or undefined review, because those belong to the process, not the person. Firms that respond to low usage with another round of training usually get a short rise in use and then the same decline. Where training fits alongside rebuilding the work is covered in training or implementation.

Where do general AI tools genuinely help?

They are good at one-off, individual tasks: summarising a long document, drafting a first reply to an unusual email, rephrasing a paragraph, thinking through a problem. That is real value, and it is worth keeping. It is just not the same as changing how the firm's recurring work gets done. What Copilot is good for looks at this in more detail.

What does rebuilding a process actually involve?

Mapping the job as it really runs, with the people who do it, including the steps nobody wrote down. Deciding what the output should be and what a reviewer checks. Building the new way on the firm's own documents, templates and language, in the tools the team already uses. Testing it on live work until it holds up. Training the team so it becomes the normal way the job is done, and supporting it while it beds in. Whether that is built bespoke or bought as a product depends on the job, a choice weighed in custom or off-the-shelf.

If your firm has licences and little to show for them, the audit shows which recurring job is costing your fee earners the most, which is the right one to rebuild first.