Legal AI workflow answers
Eight operating questions for independent law firms deciding whether one repeated workflow is ready for process change, a product evaluation or a small supervised pilot.
Operational guidance for England and Wales. Not legal advice or a product recommendation.
Eight direct answers
These answers state the current Margo Legal method. New questions will be added only when real conversations support a distinct, evidence-backed answer.
Baseline
Record the workflow's monthly volume, qualified lawyer time, turnaround, corrections and write-offs. Define the trigger, finished output, exceptions and current review standard as well. Without that baseline, a faster demonstration cannot show whether the completed and supervised work actually improved.
Open the Baseline KitSuitability
A workflow becomes testable when it repeats often enough to measure, has a stable trigger and finish line, and has a named owner. The firm must also know who reviews the output, which data may be used and which result would make the test stop.
Run the workflow scoreStop
Do not run the pilot when the work is too rare to measure, no one owns the decision, the approval route is unknown, representative inputs are unavailable or qualified review cannot be provided. Repairing those conditions is useful progress and may be the correct immediate outcome.
Read the readiness modelTest data
Use made-up examples, properly redacted material or an approved historical set when those inputs still represent the task's real variation. Keep the expected answer, source evidence and reviewer standard beside each example. The test should expose failures, not protect the tool with easy cases.
Read the evaluation guideOwnership
Name one person accountable for the operating decision, not only the technology. The lawyers doing the work, the qualified reviewer, the sponsor and the people responsible for risk, data and IT should each have explicit roles, while one owner remains responsible for the final decision.
Review the working boundariesSupervision
Measure the qualified review needed to reach the accepted finished output. Include corrections, source checking, escalation and repeated work, not only the time to produce the first draft. Compare that completed-work total with the same measure from the current process.
See the six-step testFailure
Set conditions for unsupported material outputs, unacceptable correction time, prohibited data use, missing evidence, review-capacity failure and operating cost. Agree who can pause the test and what must be retained for review before the first pilot input is processed.
Open the decision guideRoute
Fix the process first when delays come from unclear intake, ownership, templates, handoffs or review rules. Software cannot resolve an undefined finish line or missing accountability. A cleaner process also creates the baseline and controls needed for a later product comparison.
See how the four routes are comparedGo deeper
The operating guides explain readiness, tool evaluation and workflow architecture. The browser-only score applies the decision gates to one repeated workflow.
Put the answer to work
Describe the rule and exception route before selecting software. Automation fits only when a stable rule, exception route and owner already exist. If records are moving, preserve provenance and reconcile a representative sample before cutover. Protect records and critical work while moving systems.
These optional PDF worksheets are available without an email address, booking or client data. They are operational aids, not legal advice.