A concept for discussion / September 2026
Inbox
in Law.
Software that turns attorney email into draft time entries for review and approval.
Patrick Iwanicki / Working title / Working prototype on fictional data
01The problem
Attorneys reconstruct time entries.
The financial impact is uncertain.
reconstructed their time
A 2010 survey of 155 respondents reported roughly 85 minutes of lost time a week.
Vendor-sponsored. Skewed toward large firms. Self-reported.
The widely repeated claim is unsubstantiated
“10% lost the same day, 50–70% after a week” traces to an uncited 2019 consultant blog compilation. No sample or method. Often misattributed to the ABA.
Delayed entry may also inflate bills
One vendor’s platform data suggests contemporaneous entry reduced client bills by up to 23% compared with delayed entry.
There is evidence of reconstruction. There is no reliable dollar figure to build this pitch around.
02Why the problem persists
Time tracking still requires manual work
The standing advice
Record as you go
Bar associations, consultants and vendors have repeated the advice for thirty years.
Yet the traceable survey found six in ten attorneys reconstructing time.
Timers
Timers require a habit
A timer asks for a new habit at the moment an attorney is absorbed in legal work.
Passive capture
Automatic tracking requires cleanup
Desktop tracking removes the timer habit, but brings surveillance concerns and a cleanup burden.
At month’s end, the fallback is still the calendar, the sent folder and memory.
03The proposed product
Draft time entries
from attorney email
Read Gmail or Outlook messages, identify client work, and suggest the matter, description and time spent. The attorney corrects and approves each entry before export to the billing system.
No timer to start. No email to tag. Nothing billed without review.
04The workflow
How the software would work
01
Connect
The firm authorizes Gmail or Outlook read access.
02
Classify
Identify messages about client work and the relevant matter.
03
Propose
Draft a work description and suggest time, calibrated to this attorney’s past approvals, for them to verify.
04
Review
The attorney checks, corrects and approves the proposed entries.
05
Export
Send approved entries to the existing billing system.
The goal is to review a month in one sitting. Mailbox access and billing integrations remain practical risks.
05Privacy & privilege / Pilot commitments
Built for privileged email
Bodies never stored
Email text stays in memory during analysis or review. Never saved to our database.
Narrow reads, all logged
Only sent mail for the chosen month, plus limited earlier messages in the same thread. Every read is audited.
Content-free logs
Logs and error reports carry IDs and counts only. Never subjects, bodies or addresses.
No cross-client training
Estimates learn per attorney from approved entries only. They store minutes, never text.
Delete on disconnect
Disconnecting deletes the attorney’s mail data, drafts and review history in one step.
Written terms up front
A plain-language agreement names what is kept, for how long, and every outside service that sees the text.
Not there yet. No SOC 2 or ISO 27001 certification. Zero data retention from every AI provider is being arranged, not confirmed. Subjects and recipient addresses are stored for review, encrypted at rest.
Describes the pilot build. The public demo holds fictional data only.
06Built in the prototype
Suggested time learns from
the attorney’s own approvals
Every approval teaches
Approved minutes, minus time stated in the email, update that activity and matter.
Confirmations count too.
Large corrections ask why
One tap: more complex, quicker, work outside the email, one-off, or wrong activity.
One-offs are ignored; wrong activity is reassigned.
Untouched drafts update
Unedited drafts are re-estimated, with each change logged in history.
Stated durations and edited entries never change.
Example: three 30-minute approvals on one matter take its legal-advice suggestion from 12 to 24 minutes.
Estimates shrink toward defaults and round to 6-minute units. Per attorney only; stores minutes, activity and matter, never email text.
Not yet measured: whether this reduces correction time for real attorneys.
Why now
AI can interpret email.
Existing products already use it.
AI can identify client work
Models can infer the matter, activity and effort from language, beyond matching senders and subject lines.
Gmail and Outlook support integration
Gmail and Microsoft Graph are mature and documented. Restricted-scope review remains a gate.
One competitor says it cleared review. No validation letter is published.
Competing products are available
At least four products already turn attorney email into draft time entries.
The newest launch was roughly three weeks before this September 2026 discussion.
The question is whether we can offer something firms prefer to existing products.
08The competitive landscape
Five direct competitors. One is free.
| Product | Captures from | Published price | Billing system |
|---|---|---|---|
| Thyme | Email only | Free to 4 users, then $85/seat | Clio |
| Lawgbook | Email, calendar, meetings | $29.99/seat | Clio |
| Tempello | Email only | Per matched email, $59–79/mo | Clio, MyCase |
| Time Miner | Email, calls, texts | $39.99–49.99/seat | Clio, MyCase |
| Billables AI | Email, docs, browser | $39–99/seat | Clio, MyCase, LeanLaw |
Smokeball and Actionstep offer native capture. Clio lists four of these apps and offers one-click time entry from Outlook and Gmail.
Enterprise, quote-gated: Laurel ($100M Series C), PointOne, Intapp and Aderant offer full-workday capture.
Prices and market snapshot as supplied in the September 2026 source deck.
Billing requirements
Attorneys must verify time
before billing it
The billing constraint
ABA Opinion 93-379 bars billing more time than actually spent, subject to the rounding described in the opinion. Opinion 512 (2024) reaffirms the principle for AI-assisted work.
Rounding discussed: 0.1 or 0.25 hours.
The proposed design
An inferred duration is a suggestion. The attorney must verify it before representing it as time worked on an invoice.
Our interpretation: the rules do not require a stopwatch or contemporaneous entry. Review and ownership remain essential.
No bar ruling on passive time capture is identified in the source. This posture is an inference from Opinions 93-379 and 512 and Rule 8.4(c), not a direct holding.
10Commercial risks
Four risks to the business
01 / Free alternatives
Thyme is free for up to four users and came from a law firm. Lawgbook costs $29.99. Small-firm pricing is already low.
02 / Acquisitions by larger vendors
Bellefield and ZERO went to Aderant. WiseTime went to Anaqua, Smart Time to BigHand, and Traced to Actionstep in January 2026.
03 / Review takes too long
Reported complaints about the best-funded incumbent include review that takes longer than manual entry and double counting of concurrent activity.
04 / IT approval slows adoption
Working assumption: an IT administrator step at every firm, plus annual paid restricted-scope review. Access is a distribution constraint.
No independent adoption measurement is identified. The circulating 10–15% of timekeepers estimate came from a vendor founder.
11The proposed cost advantage
Use low-cost AI to classify email.
Use a language model to draft entries.
Classify every message
A classification model identifies potentially billable messages and returns a confidence score. It does not generate text.
Draft entries for relevant messages
Use the more expensive language model only to write descriptions for proposed time entries.
to classify a month’s inbox
About 70× less than using a language model on every message in this example.
A competitor charges $0.49 per matched email. Whether lower cost creates customer value remains unproven.
Source rate: $0.042 / million input tokens, output free. Assumption: 2,000 messages × 750 tokens.
The source states no training on customer requests and that zero retention is available. Classification cost only.
Illustrative unit economics
Illustrative processing costs
for one attorney’s monthly email
Cost of a full-inbox pass
Same 1.5 million input tokens.
About 71× lower input processing cost.
Messages in the proposed workflow
A one-in-six billable share.
Approximately 333, shown as ~300.
The language model drafts entries only for messages classified as potentially billable.
Illustrative, not measured. Assumptions: 2,000 messages/month, 750 tokens/message, $3/million generative input tokens, one-in-six billable share. Classifier: $0.042/million, output free. These are not total service costs.
13Potential differentiation
We would need lower costs
and faster review
Lower processing costs
Typed classification could lower marginal cost per message. Competitors would need to change their processing architecture to match it.
But a lower cost base does not establish a business in a market with a free product.
Less time checking and correcting entries
The complaints concern cleanup and double counting. Fast, trustworthy approval of a month’s work is the proposed opening.
The prototype’s suggested minutes learn from each approval. Competitors that publish a method use fixed increments or timestamps. Whether corrections shrink is not yet measured.
The alternative conclusion: email capture is already a commodity, and there is no opening.
All five direct competitors started with Clio. Other billing systems may offer room, or insufficient demand.
Passing on the concept remains a legitimate outcome.
Business assumptions
What we need to validate
01
Value
Lower processing costs produce a benefit firms will pay for.
02
Review
Review beats writing entries, and learned suggestions cut corrections.
03
Willingness to pay
A buyer chooses us over a free, firm-built alternative.
04
Distribution
We reach firms and get through the admin-consent step.
05
Independence
Acquisition is a viable outcome for a sustainable business.
Which assumption should we test first?
15Feedback requested
Is this worth
developing further?
We are asking for feedback on the idea, with no request for funding or a career commitment.
What are the five existing products missing?
Would your firm pay for this when a free option exists?
Could this support an independent business, or would it need a larger software vendor?
A clear “no” is useful. If the answer is “yes”: three attorney conversations focused on review burden, with the prototype in hand.
16