Batch consulting admin work, but only the half of it that does not decay.

The split, side by side

One block on the calendar vs two clocks on the events.

The left tab is the calendar block most batching guides recommend. Six tasks crammed into 150 minutes on Friday afternoon, with 18 minutes left at the end to approve and send. The right tab is the same week, split: drafts fire on the event the loop closes on, approvals queue, you flush the queue when convenient.

# the calendar block most guides recommend
# Friday 2:00pm to 4:30pm

2:00 - 2:18  draft post-call summaries from this week's 6 Zoom recordings
2:18 - 2:34  log activities to HubSpot for this week's calls
2:34 - 2:51  draft followup emails to the 6 clients
2:51 - 3:08  draft the 4 milestone invoices in QuickBooks
3:08 - 3:21  chase the 3 invoices that hit Net+7 on Tuesday
3:21 - 3:39  write 5 weekly status emails, one per active client
3:39 - 3:52  approve and send everything

# what actually happens

# Tuesday's call summary is now 3 days cold; the action item
# you remember as "they hesitated on price" is reduced to
# "discussed pricing." the Tuesday Net+7 invoice is now Net+10.
# 18 minutes is what's left for the most expensive thing on the
# list (the per-client status email) and you ship it copy-pasted.

The left tab is what most consultants land on after a year of trying to keep admin off their daily schedule. The right tab is what the calendar looks like once the draft clock moves off Friday and onto the events that fire it. The right tab is what the keyword is actually asking for, even when the question is phrased as “how do I batch my admin.”

The queue, named

~/.clone/queue/approvals.json: where the held drafts wait.

The approve clock has a file. Each row is one held draft. The shape is seven fields. You can read the file in TextEdit, grep it from the terminal, or open it from Clone's menu bar. The queue is plain JSON because the queue is something you will look at, edit, and occasionally clear by hand; a binary database would be wrong here.

{
  "id": "apr_2026-04-28_0007",
  "drafted_at": "2026-04-28T14:03:11-04:00",
  "loop": "post-call.md",
  "target_app": "hubspot.com",
  "action": "create_activity",
  "hold_reason": "deal_value > $25k threshold",
  "preview": "Sarah · Acme Renewal · 2026-04-28 1:30pm\nDecision: extend pilot through Q3, revisit pricing in July.\nNext step: send proposal v3 by Friday."
}

The queue is the entire approve clock. There is no separate approval app, no inbox plugin, no Slack channel. You open the file (the menu bar icon does this), scan, tap, close. Most consultants on Clone trial accounts settle into 4 to 12 rows per day for a solo practice. A 10-minute end-of-day window covers it.

Per loop, by clock

How drafts and approvals decay differently, loop by loop.

The reason the two-clock split is load-bearing is that for every loop in the back office, the draft half decays sharply and the approve half barely decays at all. The four loops below are the ones that fire most often in a solo or boutique practice. Compare the two halves, line by line.

The asymmetry is the entire reason this works. If approvals decayed at the same rate as drafts, batching would not help and the only honest answer would be a calendar block. They do not. Approvals are pure judgment on artifacts that already exist; the cost of approving an hour later is the same as approving immediately, until you cross the day boundary on time-sensitive items, which is exactly what hold_reason flags for you on the first scan.

Picking your approve window

The batch shape is yours; the draft clock is not.

Once the draft clock is on the event, the approve clock can fit your calendar however you want. Three patterns that work, on Clone trial accounts:

1. End-of-day flush, 10 minutes. The most common. You scan the queue at 5:50pm before logging off, approve the day's drafts, log the rejections. By 6:00pm the queue is empty for the night. The post-call summaries from the day's calls have already been written from warm context; the invoices fired on milestone signoff; the only thing you spent time on was the judgment.
2. Twice a day, 5 minutes each. 10:30am and 4:30pm. Suits consultants whose followup commitments are tight (you said “I'll send it this morning” on a 9am call and you want to honor the promise without checking email between calls). The morning flush catches yesterday-evening drafts and any post-call drafts from the 8am or 9am call; the afternoon flush catches the rest.
3. Friday 25 minutes plus a Tuesday spot-check. For consultants who genuinely prefer a single weekly admin window. The trick that makes this work is the hold_reason flag: time-sensitive items (a Net+0 invoice, a same-day followup, a sensitive_client message) flag at the top so they get approved Tuesday instead of slipping to Friday. Everything else waits comfortably for the Friday block, because the drafts are already written.

Note what is not on this list: a daily 90-minute Friday admin block. That shape is what you do when drafting is on the same clock as approving. With two clocks, 90 minutes of admin per week is unnecessary; the work that needed warm context already ran on warm context, and what is left is a few minutes of judgment.

Where the one-block answer actually wins

When a single calendar block is the right answer.

The two-clock model is not always correct. There are two cases where the conventional answer (one calendar block, do all your admin inside it) is genuinely better. Both are about volume and rhythm.

The first case is a low-volume practice. If you have one or two active clients, two calls a week, an invoice a month, and no recurring weekly status email, the per-week admin volume is small enough that the recall decay across a week is bearable, and the overhead of running a queue plus a draft clock is not worth it. A 30-minute Friday block does the job. The two-clock split starts paying off around 5 or more active client touches per week, which is roughly when the post-call summaries and the milestone invoices stop being something you can hold in your head from Tuesday through Friday.

The second case is a fixed-cadence practice. Wedding photography, fixed-scope brand identity, single-modality coaching with a flat-fee package: the back office is genuinely contract-to-invoice on a single repeatable cadence, and HoneyBook or Dubsado closes the one loop you care about with less ceremony than two clocks plus a queue. The two-clock model is for consulting practices where the back office spans the post-call window, milestone invoicing, dunning, onboarding, weekly status, monthly bookkeeping, and quarterly pipeline review on independent clocks. If your practice does not have those seven loops, the conventional advice is the right advice.

For everyone else (the solo consultant or boutique firm of 1 to 10 with a CRM, a calendar, an invoicing tool, and a meeting cadence), the two-clock split is the architecture the keyword is actually asking for. Once you see drafts and approvals as two operations, the question of when to batch becomes a question only about approvals, which is the half that admits an answer.

The smallest start

How to ship the two-clock model in three weeks.

You do not need to write the full seven handlers on day one. Pick the loop that fires most often (post-call admin) and run only that one through the queue for the first week. Add a second loop in week two (milestone invoicing). Add a third in week three (dunning). By the end of week three, three of the seven loops are off the calendar and onto the events that fire them, and your only Friday admin is a 15-minute approval scan.

1. Week 1: post-call.md. The handler fires on zoom_call_ended. Five concurrent draft jobs (call_summary, hubspot_log, followup_draft, action_items, next_meeting), all set to hold_default = true so the drafts go to the queue instead of sending. Inside the first week you will see the queue accumulate previews you can scan in 90 seconds. The credibility win lands when the Tuesday 1:30pm call summary is in the queue at 1:31pm with the right action items, and you approve it at 5:50pm without re-listening to the recording.
2. Week 2: invoice.md. Fires on milestone_signoff (an inbound email matching the SOW phrase, or a deal stage change in HubSpot). Three handlers (timesheet_pull, invoice_draft, invoice_send), with invoice_send held by default if the amount differs from the SOW expected number by more than 10%. The DSO improvement shows up in the first month; on a $50k month with a 9-day average DSO improvement, that is roughly $1,200 in working capital you were leaving on the table per month.
3. Week 3: dunning.md. Fires on invoice_age = 7 / 14 / 21. Three handlers, one per tranche, tone graduated from gentle nudge to firm. Holds anything over $5,000 for review. This loop is the one most consultants quietly fail at because the day Net+7 trips always loses to client work; with the draft clock decoupled, the reminders draft on the threshold day and you approve them on your normal flush.

The remaining four loops (onboarding, weekly status, monthly bookkeeping, quarterly pipeline review) get added the next time their event fires. You write onboarding.md the next time you sign a new client. You write weekly-report.md the next Friday afternoon. You write monthly-bookkeeping.md at the end of the month you start. You write quarterly-pipeline.md the week before the quarter closes. The whole stack is event-driven, including the order in which you build it.