A free newsletter from Randy Cloud, a law firm administrator, on putting AI to work at small and mid-size firms: what to try, how to roll it out, what went wrong, and how to tell if it’s working.
Free when you subscribe: Exhibit “Ai”, a kit for your first 90 days of AI: a one-page brief for your managing partner, staff AI rules, a vendor demo scorecard, before-and-after worksheets and a calculator.
| Hours the tool saved | 40.0 |
|---|---|
| Checking its work | (14.5) |
| Handling exceptions | (6.0) |
| Fixing mistakes | (4.0) |
| Upkeep | (3.0) |
| Time the team got back | 12.5 |
A green check is not a read.
Which tools and workflows are worth trying, how to judge a vendor, and how to roll it out to the people who do the work.
Failures from real rollouts, told plainly. The part a vendor demo leaves out.
The risk, ethics or pricing point your managing partner needs, short enough to forward.
Plus training and policy, measuring what changed, and whatever comes next.
Ten real items from your own work: touches and minutes, before anything changes.
The same items through the new workflow. The machine’s part is the dashed line.
Someone named checks the output end to end. This is the solid line, and it has a cost in minutes.
The number, who was counted, the period, and its grade from E0 to E5. No receipt, no number.
AI sorts each document and matches it to a matter. The legal assistant reviews every exception before anything is saved.
A deadline tool that is right 19 times in 20 still needs a person on every date. Here is the math that gets one switched off in month four.
Failure FileExample · pattern, not a case
Every Receipt Check reads a published AI study and asks what was really measured: who answered, the exact question, and what was left out.
Grade E1Self-reportedThe extractor reads every court notice and proposes a deadline. It is right most of the time, which is exactly the problem: a deadline that is usually right still has to be checked every time.1
This is the line to show your partners.
So the administrator moved the person, not the work. The legal assistant no longer types dates. They confirm them against the docket, and that check is the only solid line in the diagram.2
What was measured, the sample, the period, and who counted it. No vendor pays for an outcome, and any sponsorship is labeled. Small samples say so. When we don’t know yet, we write “unknown.” Each number carries a mark from E0 to E5 so you can see at a glance how far to trust it.
Six working papers for your first 90 days of AI: the Partner Brief, Staff AI Rules, Pick Your First Workflow, the Before Sheet, the Vendor Demo Scorecard and the After Sheet, with worked examples, fill-in fields and a calculator that does the math.
Subscribe to get itPutting AI to work at small and mid-size law firms: choosing tools and judging vendors, rolling AI out to a team, building workflows across the systems you already have, risk and ethics, pricing, and measuring what changed. Topics follow what firms need.
When there is something worth sharing. There is no fixed schedule, so nothing gets padded to fill a slot.
Nothing. The newsletter and Exhibit “Ai” are free, and no vendor pays for an outcome. If a sponsor ever appears, it will be labeled.
Randy Cloud, a law firm administrator who builds, rolls out and tests AI tools as part of the job. Proveup is his own project, separate from his firm.
No. It is operations and practice: how to choose, roll out and measure AI at a firm, and what to take to your partners. When an issue touches ethics or billing rules it cites the source, and you should read that source before setting policy.
Proveup is written by Randy Cloud, a law firm administrator who builds, rolls out and tests AI tools as part of the job. It is his own project, separate from his firm. Cases are composites and firms are never named. No vendor pays for an outcome, any sponsorship is labeled, and nothing here is legal advice.
