The AI Timekeeping Pilot: What to Expect Week by Week
# The AI timekeeping pilot: what to expect, week by week
Software categories split into two kinds: the ones you have to believe in, and the ones you can test. AI timekeeping is testable. Two weeks with a few real attorneys produces hard numbers, which is why almost every firm that buys in this category runs a pilot first, and why we're happy to describe exactly what one looks like.
This is written from Ajax's pilot playbook, but the structure applies to any vendor worth evaluating. If a vendor resists a measured pilot, that's information too.
Before day one: the 30 minutes of setup
A pilot needs three things decided up front.
The pilot group. Two to five timekeepers works best: volunteers who feel the timekeeping pain, plus one honest skeptic. Skeptics are the best pilot participants we get; their conversion carries the partner vote later. (If you're the person assembling the case for partners, we wrote a whole playbook for that meeting.)
The success metrics, agreed in advance. Three cover it: hours captured per timekeeper versus their trailing average, minutes per day attorneys spend reviewing entries, and the billing manager's read on entry quality. Write them down before anything installs. Metrics chosen after the fact convince nobody.
The install itself is genuinely light: a desktop app per pilot user (Mac and Windows both) and a connection to your billing system; for most supported systems, rates and matters sync over automatically. Nobody migrates anything.
Days one and two: the strange first review
Entries start appearing the first day; Ajax drafts them within about 45 seconds of the work happening. The first end-of-day review is where something interesting reliably happens.
Attorneys find work they'd already forgotten. A call from mid-morning, two emails answered between meetings, twenty minutes of research that never felt like an event. First reactions run from "huh" to mild indignation at their own past timekeeping. This is the under-capture gap becoming visible, and it's the whole reason the category exists.
Expect drafts to be roughly right and personally wrong: correct matter, correct time, narrative phrased the way a stranger would phrase it. That's normal at hour zero, and it's what the next few days fix.
Days three through seven: the tool learns your people
Every edit teaches the system. Attorneys who shorten narratives get shorter narratives. The associate who writes "correspondence with" instead of "emails to" starts seeing her own phrasing come back. Matter prediction sharpens as the system learns which people, documents, and topics belong together at your firm.
The daily review settles into its steady state this week: most attorneys land at a few minutes a day, skimming 10 to 20 entries that cover the whole day. If anyone in your pilot is spending more time than that merging fragments, say so in the check-in; review burden is precisely the thing to measure while you're measuring.
Our activation team runs this stretch with you: a kickoff, a mid-pilot check-in, and same-day answers in between. Firms consistently tell us the white-glove part mattered more than they expected, so whoever you pilot with, ask what support during the pilot actually includes.
Week two: steady state, and the numbers
The second week is the measurement week. The novelty is gone, attorneys are working normally, and the data accumulates on the three metrics you set.
What good looks like, from firms that published their results: captured hours that surprise people (Amy Robinson's published figure was over 60% more billable hours than her manual baseline; treat that as a ceiling, not a promise), review time in the minutes, and a billing manager who stops spot-checking because the narratives keep passing. Hone Law watched its on-time billing jump to 95% in the first week. Your numbers are the ones that matter; more published ones are on our case studies page to calibrate against.
The decision meeting
End of week two, put the three numbers next to the seat price and let the arithmetic talk. One recovered hour per week per timekeeper typically covers the cost several times over, and most firms find the capture delta is comfortably past that.
At Ajax, well over 90% of firms that pilot go on to subscribe. We publish that number because it says the pilot structure works: firms aren't buying a pitch, they're buying two weeks of their own data.
And if your numbers are marginal? Then the pilot did its job too, and you walk away having spent very little to learn it. That capped downside is the entire argument for piloting instead of debating.
FAQ
How long does an AI timekeeping pilot take?
About two weeks in most cases: a few days of learning-curve while the system adapts to each attorney, then a steady-state week that produces the numbers you'll decide on.
How many attorneys should participate?
Two to five. Enough to span seniority and practice styles, small enough to support closely. Include at least one skeptic on purpose.
How much work is the pilot for the attorneys?
A few minutes a day reviewing drafted entries. There are no timers to run and nothing to log; the review itself is the workflow being tested.
What should we measure during the pilot?
Three things, agreed before you start: hours captured versus each participant's trailing average, minutes per day spent on review, and entry quality as judged by whoever runs your billing.
What happens to our data if we don't move forward?
Ask this of any vendor up front and get the answer in writing. At Ajax, raw captured data deletes automatically on a rolling basis as a matter of policy, and offboarding removes the rest on request.
Ready to see it with your own matters? Book a demo and ask us to scope a pilot for your firm size.
