Morgan McPherson
The Morgan McPherson Brief

Edition 001 · Legal technology & AI

AI adoption should remove work, not create another review queue.

A practical operating framework for choosing, testing and supervising AI-assisted workflows in a growing law firm.

Morgan McPherson Editorial · · 2 min read

The practical takeaway

Choose one bounded workflow, define who reviews it and measure the whole process — not just the time spent generating a draft.

The management question behind the technology

The useful question is not whether a law firm should “use AI”. It is which recurring task is consuming time, what a better outcome would look like and whether a supervised tool can help achieve it. A product demonstration can show a fast draft. It does not show how that draft fits the firm’s information systems, review responsibilities or client commitments.

The Law Society of Ontario’s guidance encourages informed experimentation while emphasising the need to understand risks and professional obligations. Its practical tips specifically address confidentiality, privacy and verification of generated outputs. Those points belong in the implementation plan, rather than in a policy document that no one consults during the working day.

Define a small, observable workflow

Start with a bounded process: for example, preparing an internal checklist from approved, non-client-specific material. Identify its inputs, the expected output and the person responsible for checking that output. Avoid making the first experiment an open-ended assistant expected to handle every matter type.

Write down what the tool must not do. It should not silently invent missing facts, present an unverified citation as reliable or move sensitive material into an unapproved environment. When a necessary input is missing, the workflow should flag the gap and route it to a person. A clear exception path is part of a useful design, not evidence that automation has failed.

Measure review effort as well as generation time

A faster first draft can still produce a slower process if someone must correct inconsistent formatting, trace unsupported statements or rebuild the context. Compare the complete workflow before and after the pilot: preparation, generation, review, corrections and final approval.

Keep a short record of recurring errors. If the same problem appears repeatedly, decide whether to change the task, the approved source material or the tool. Prompt adjustments alone will not repair a process that has no reliable source of truth. Evaluate quality and total time together; neither output volume nor a persuasive demo establishes an operational return.

Make ownership visible

Assign one workflow owner and one accountable reviewer. Agree when a failed test requires the work to return to the existing process. Staff should know which system contains the approved version and which material is only an intermediate draft.

A sensible pilot ends with a management decision: adopt, revise or stop. If the workflow produces consistent, reviewed outputs and lowers total effort, document it and train the people who will use it. If not, retain the learning without expanding the tool’s remit. The objective is a more dependable firm, not a larger technology stack.

A useful first discussion

Choose a recurring process that currently creates repeated questions or unnecessary rework. Map its inputs and hand-offs on one page. Then ask whether automation would remove a step, support a step or merely make an unresolved step faster.

Morgan McPherson treats AI as part of growth infrastructure: a way to support trained people and clearer systems. The commercial value depends on what changes in the operation, not on how prominently the firm can advertise that it has adopted a tool.

Sources & further reading

Keep a sharper perspective.

Published by Morgan McPherson / SwiftScale, Morgan Business Group Limited. Contact or withdraw consent at contact@morganmcpherson.ca · Privacy. Email delivery is being prepared.