Consulting back office automation is not one workflow. It is seven loops with seven different close windows.

It is the practice of having software, not you, do the seven recurring chores that surround client work: post-call admin, invoicing on milestone, chasing unpaid invoices, onboarding new clients, the weekly status email per active client, the monthly bookkeeping close, and the quarterly pipeline review. Most software treats this as one workflow on one schedule. Honest automation runs all seven on their own clocks.

Each loop has a different trigger event and a different close window. Post-call fires on zoom_call_ended and the close window is 60 seconds. Invoicing fires on milestone signoff and the window is the same business day. Dunning fires on invoice age and the window is the day a Net+N threshold trips. Onboarding fires on contract signed. Weekly status fires on a calendar event (Friday 5pm). Monthly bookkeeping fires on the last business day of the month. Quarterly pipeline fires on quarter end. There is no single cadence that catches all seven; one cadence will be wrong for at least five of them.

HoneyBook, Dubsado, and Bonsai replace your stack. They are most useful for the contract-to-invoice loop on milestone signoff, which is one of the seven. The other six loops they handle either generically (a default Net+30 dunning schedule, a copy-pasted weekly status template) or not at all (post-call admin, monthly bookkeeping reconciliation, quarterly pipeline review). The trade you make is: replace six of your existing tools with a single tool that closes one loop well and five loops poorly. Clone takes the opposite trade: keep the tools, run the loops on their own clocks.

A Zapier graph for the back office tends to look like: every morning at 8am, scan Gmail for invoices, scan Stripe for payments, scan HubSpot for stage changes, and fire some downstream actions. The graph runs on one cadence (daily, hourly, or whatever you set), so each loop closes either too early or too late. Post-call admin runs 18 hours after the call instead of 60 seconds. Dunning runs on the next morning after the Net+N threshold trips, not the day it trips. Quarterly pipeline review tends to never run because it does not fit a daily cron. Clone runs each loop on the event the loop actually fires on, not the next tick of a shared cron.

Each file is a plain-text declaration of: the trigger event, the target apps, the action steps, and the hold rules. post-call.md has five handlers (call_summary, hubspot_log, followup_draft, action_items, next_meeting) firing in parallel on zoom_call_ended. invoice.md has three handlers firing on milestone_signoff. dunning.md has three handlers, one per Net+N tranche. onboarding.md has six handlers firing on contract_signed. weekly-report.md has one handler per active client firing Friday 5pm. monthly-bookkeeping.md has four handlers on eom_close. quarterly-pipeline.md has three on quarter_end. The whole thing is something you can read in twenty minutes and edit in TextEdit.

The handler keeps working in most cases because Clone reads the screen and types like you do, instead of calling an API. A HubSpot UI redesign costs you a few minutes of reteaching, not a broken integration. A new field on a QuickBooks invoice screen is a one-line edit. The architectural principle from src/components/architecture.tsx is tool agnostic by design: switch from HubSpot to Pipedrive, change one line in post-call.md and weekly-report.md. No OAuth re-wiring, no graph rebuild.

Yes. Each handler has explicit hold rules. The followup_draft handler holds when the proposal amount exceeds your declared threshold or the client is tagged sensitive. The dunning handler holds anything over $5,000 for review before sending. The invoice handler holds if the milestone signoff email is ambiguous. When a handler holds, it queues the action in your review surface with the reason, and you approve in one tap. The drafting and the sending are two different decisions; this approach separates them so the drafting can run on the loop's own clock while the sending stays under your control.

Every handler writes one entry to ~/.clone/log/<date>.json. Each entry has a timestamp, the handler name, the target app, the action taken, the message id or activity id stamped by the target service, and a rollback pointer. You can roll back any single action with one click. The log lives on your machine; nothing is sent to a remote dashboard. The architectural commitment is in src/components/architecture.tsx: every action Clone takes is logged and reversible. Roll back an entire morning of work with one click if you need to.

One file. Start with post-call.md because the credibility win is immediate and the loop fires multiple times a day, so you see the agent driving your stack inside 24 hours. Add invoice.md in week two so milestone signoff stops sliding into the Friday batch. Add dunning.md in week three because most consultants quietly carry $3,000 to $10,000 of unchased invoices at any given time. Add weekly-report.md in week four because it is the loop the rest of the seven all feed into. By the end of week four, four of the seven loops have moved off your calendar and onto the events they belong on.

A virtual assistant at $25-40 an hour for 20-40 hours a month runs $500 to $1,600. A senior VA running a full back office runs $3,000 to $6,000 a month. The hourly cost is not the only cost; the bigger one is that a VA fits the same mistake as a single tool: one cadence, applied to seven loops with seven different windows. Most VAs end up doing post-call admin in a Friday batch because that is when their hours fit. The result is the same delayed close on five of the seven loops. Clone runs at $49/month per seat and fires each loop on its own event, 24/7. The two coexist well: keep the VA for the judgment calls Clone holds for review and for the things a VA is genuinely better at (vendor calls, scheduling for difficult clients, paper mail).

Yes. Each consultant gets their own ~/.clone/memory/ directory and their own handlers. Shared loops (monthly bookkeeping, quarterly pipeline) point at one shared file in the firm's Drive. The boutique tier at $129/seat/month is built around this shape: per-consultant handler files plus shared loops at the firm level. The hold queue is per-consultant, so partners do not approve each other's drafts.

You can declare a synthetic trigger. Add closeout.md with `trigger: deal_stage = closed-won` (a HubSpot or Pipedrive event). Or `trigger: drive_folder_archived` if you mark closeouts by moving the client folder. Or `trigger: manual_invoke` and call it from the menu bar. The point is each loop has its own event, even if you have to declare the event yourself. The cost of declaring a synthetic trigger is one line.