How does AI timekeeping assign work to the right client and matter?
AI timekeeping software assigns work to a client and matter by matching evidence from the lawyer's day against the firm's open matters. The strongest evidence is an exact identifier, such as a matter number or client name in a document title or email subject, and who the work is with. Weaker signals, like similar past entries, should only support a match, never decide it. When the evidence is thin, a careful tool leaves the matter blank for the lawyer instead of guessing, because a wrong matter is worse than an empty one.
This page explains how LegalTime AI makes that decision, how we measure it, and what to ask any vendor before you trust its matter matching.
How LegalTime AI picks a matter
- Start from your open matters only. LegalTime AI copies the matter list from your practice system and keeps only matters that can take time: open and pending matters. A matter closed in the last 90 days is used only on an exact identifier. Older closed matters, test matters and duplicates are excluded. Firm admins can include or exclude any matter.
- Collect the evidence for each block of work. Window and document titles, the text on screen, email threads, calendar events, call participants, and the edits you made.
- Search for candidate matters. Each candidate gets a reason it matched: an exact matter number or name, a link learned from an earlier correction, a client or contact name, keywords in the matter fields, or how the same kind of work was billed before.
- Check before assigning. A separate step reviews every proposed matter. It passes when there is an exact identifier or a link learned from your own corrections. It fails when the only support is similarity or past entries, or when two matters for the same client cannot be told apart.
- Leave it for you when unsure. If the check fails, the entry stays without a matter and shows suggestions. Drafts are visible only to you until you release them, and nothing goes into your practice system without your approval.
- Learn from your corrections. When you move an entry to a different matter, that correction becomes the strongest evidence next time. Entries you billed count more than unreviewed drafts.
What counts as evidence
| Signal | Example | How much it counts |
|---|---|---|
| Exact identifier | "2024-0117 Smith v. Jones" in a document title or email subject | Enough on its own |
| Learned link | You moved "Riverside lease review" to the Riverside matter last week | Enough on its own |
| People | An email thread or call with a contact on the matter | Strong, checked against other matters for the same client |
| Matter fields | Keywords from the matter description or practice area | Supporting only |
| Past entries | Similar work billed to the matter before | Supporting only, never decides alone |
| Neighbouring work | Research right after drafting a brief for one matter | Inherits that matter only within 15 minutes, with no other matter in between, and never a closed matter |
Internal firm work goes to an internal matter and is marked non-billable.
How accurate is it?
Benchmarked on a real law firm's Clio account. We took a small firm's real Clio account, 359 matters, and recreated its lawyers' work from the time entries the firm filed itself. Then we checked every matter LegalTime AI picked against what the firm actually billed. Across 119 blocks of screen work, it never chose the wrong client. 86% landed on the exact matter. One went to the same client's other matter, and the rest were left for the lawyer to choose.
Measured on live entries. Over the last 28 days, an independent AI review checked more than 4,000 matter matches from four law firms using LegalTime AI. 82% clearly fit the work they described.
When it isn't sure, it asks. LegalTime AI only assigns a matter when the evidence is clear. Otherwise the entry stays open with suggestions, so you pick the matter instead of finding a wrong one on a bill.
Most vendors don't publish how they measure matter accuracy. Ask yours the questions below, then try it on a real week of your own matters.
Where matching is hardest
Accuracy depends on the firm. Matter numbers and client names in document titles and email subjects make matching easier. Firms with many matters for the same client, or lots of phone work without calendar entries, will see more entries left for review.
- Several matters for one client. If the evidence names the client but not the matter, LegalTime AI asks instead of picking one.
- Work with no written trail. A call with no calendar event and no notes has little to match on.
- New matters. A matter opened today is available after the next sync with your practice system.
- Generic research. Reading case law without a document or email for a specific matter only inherits a matter from neighbouring work, under the limits above.
Questions to ask any vendor about matter accuracy
Vendors measure accuracy in different ways, so ask each one the same questions:
- What exactly is counted as correct: the matter you would have chosen, or a matter that is merely plausible?
- What happens when the tool is unsure: does it guess, or leave the matter blank?
- How often does it assign a wrong matter, separately from how often it leaves one blank?
- Was the number measured on real firms' work or on test data, and how many entries?
- Does it ever assign time to a closed matter?
- How does it tell apart two matters for the same client?
- Do your corrections change future matches?
Then test it on a real week of your own matters during the trial, and count wrong matters separately from blank ones.
Related
Frequently asked questions
Does LegalTime AI ever guess a matter?
Can it put time on a closed matter?
Who sees the entry before it reaches Clio?
Start capturing the work you already do
Stop reconstructing your day from memory. Let capture run. Review in minutes. Release only what should bill.
Capture → Create → Connect. Not a monitoring tool.