You cannot automate a consulting back office in one shape. There are four shapes, and each loop only closes under one of them.

It means picking the right mechanical shape for each recurring loop in your back office. Most articles treat it as a single decision (which suite to buy, which Zapier graph to wire). It is actually a per-loop decision; some loops want to be Replaced into a suite, some want to be Glued by Zapier, some want to be Delegated to a human, and some only close honestly under Drive (software that reads your screen and types like you do, on plain English instructions).

Replace: pick a suite that owns the loop end to end (HoneyBook, Dubsado, Bonsai, Ignition). The trade is you migrate your stack into the suite. Glue: keep your tools and wire them with Zapier or Make using webhooks and triggers. The trade is fragility on fuzzy triggers. Delegate: a human (a virtual assistant, an EA, an offshore bookkeeper) runs the loop on their calendar. The trade is response time and cost. Drive: software reads your screen and types into your existing apps from a plain English instruction file. The trade is you write the instruction once and edit it as your practice changes.

Because consulting back office loops have four different trigger types and each shape only handles a subset. Clean triggers (a contract is signed, a webhook fires) are fine for Glue. Fuzzy triggers (the milestone signoff phrase appeared in an inbound email; the proposal number was mentioned in a transcript) only close under Drive or Delegate. Calendar triggers (Net+7, end of month, end of quarter) work under any shape but pay back fastest under Drive because the cadence can be tuned per invoice or per client. Human-judgment triggers (which segment to raise rates on, which dormant client to chase) only close under Delegate or you. One shape applied to all seven is wrong on at least three of them.

It is more specific than that. Drive means: the software reads what is on your screen, types like you do, and does not call APIs. The instructions live in plain English in a file you edit (in the case of Clone, files in ~/.clone/memory/). When HubSpot redesigns its UI, a Drive handler keeps working in most cases because it sees the new screen the way you do. When you switch from HubSpot to Pipedrive, you change one line in the file. APIs are not the surface; the screen is. That is what makes it different from a Zapier graph (which depends on stable APIs) and from a generic chatbot (which has no persistent instructions and no integration with your stack).

A CRM is one slice. Replace means buying a suite that claims to own the entire back office (proposals, contracts, invoices, client portal, tasks, sometimes booking). HoneyBook, Dubsado, and Bonsai are the canonical examples for service businesses. The fit is honest for one specific shape of practice: a single repeatable engagement on a single cadence. If your work is contract-to-invoice on a fixed pattern (wedding photographers, fixed-scope brand work), Replace closes the loop with less ceremony than the other three shapes. If you have seven independent loops on seven calendars, Replace covers two well and the rest poorly.

Glue wins on clean-trigger loops. Contract signed in DocuSign? Webhook fires, Zapier creates a Drive folder, sends a welcome email, posts to Slack. That is one Zap and it works for years. Glue starts to fail when the trigger is content rather than an event: the moment a Zap has to read a transcript and decide whether the milestone was actually signed off, the if-this-then-that branching tree gets unmaintainable. The honest line is: clean triggers go to Glue; fuzzy triggers do not.

Delegate wins on human-judgment loops. The quarterly pipeline review where you flag dormant clients. The sensitive-tone followup after a hard call. The vendor call where the line is held by a human voice. Delegate fails on response-time loops because a VA's clock is not Net+7, it is the VA's working hours; if Net+7 trips on a Friday afternoon and the VA is part-time Mon to Wed, the chase email goes out at Net+11 in a tone that has already escalated. The trade is hourly cost vs. response-time mismatch.

Drive wins on fuzzy-trigger loops and on calendar-trigger loops where the cadence has to be tuned per invoice or per client. Post-call admin (the trigger is content in a transcript). Milestone invoicing (the trigger is a phrase in an email). Weekly client status (the trigger is the week's content). Tone-graduated dunning (the cadence is per-invoice, with a hold rule over a dollar threshold). Of the seven canonical loops, Drive closes four cleanly, two passably (with a hold for human judgment), and one (quarterly pipeline review) sits better with Delegate plus a Drive-prepared brief.

You should mix. Most healthy consulting practices end up with: a clean-trigger Glue path (DocuSign signed -> Drive folder + welcome email), a Drive path for the fuzzy loops (post-call, milestone invoice, weekly status, dunning), and a human (you or a part-time VA) for the human-judgment chores. The mistake is picking one and forcing all seven loops through it. Picking one usually happens because of marketing pressure; HoneyBook tells you the back office is one workflow, Zapier tells you it is one graph, the VA tells you it is one set of hours. The honest answer is that the back office is seven loops with four trigger types, and the right move is shape-per-loop.

Clone is the Drive shape, productized. It reads your screen, types into your existing apps, and runs from plain English instructions in ~/.clone/memory/. It is not a Replace (it does not own a database; your tools are still your tools), not a Glue (it does not depend on stable APIs), and not a Delegate (it has no human shift hours). For the four-to-five loops in a typical consulting practice that have fuzzy or per-instance-tuned triggers, Drive is the only shape that closes them on the right clock. For the loops where Replace, Glue, or Delegate are honestly better, Clone gets out of the way; it does not try to own everything.

Pick one fuzzy-trigger loop and Drive it. The cheapest one to start with is post-call admin: the trigger fires several times a day, the win is visible inside 24 hours, and a single instruction file (~/.clone/memory/post-call.md) closes the loop end to end. Once that runs for a week, add the milestone invoicing loop, then dunning, then weekly status. By week four, four of the seven loops are running on the events they actually close on, and the remaining three (onboarding, monthly bookkeeping, quarterly pipeline) sit on Glue or with you, on the right shape.

It is fair on the loops where they overlap; it is not fair on the loops only one shape can do. A VA at $25-40/hour for 20-40 hours a month is $500 to $1,600 entry cost; a senior VA running a full back office is $3K to $6K. Drive (Clone solo) is $49/month and runs 24/7, so the per-loop cost on the four fuzzy-trigger loops is roughly 50x lower. But the VA can hold a sensitive client call and Drive cannot, so the right comparison is per-loop, not per-month. Most healthy stacks end up at $49 to $129 of Drive plus a part-time VA or fractional ops person for the human-judgment chores, not one or the other.