You cannot automate a consulting back office in one shape. There are four shapes, and each loop only closes under one of them.
Almost every guide on this topic answers the question with one tool. Pick HoneyBook. Wire a Zapier graph. Hire a virtual assistant. Buy ChatGPT Teams and prompt your way through. The framing is always the same: there is one verb, “automate,” and the only choice is which product to point it at.
That framing is wrong, and it is why most consulting back office automation projects quietly stall. There is no single mechanical shape called “automate.” There are four. They are not interchangeable; each one only closes a subset of the loops in a real practice. Pick the wrong shape for a loop and the loop runs on the wrong clock until you give up and do it by hand on Friday again.
This page names the four shapes (Replace, Glue, Delegate, Drive), maps the seven canonical consulting loops onto them, and shows the trigger types each shape can and cannot handle. The mapping is the unique part; everything you read elsewhere assumes the answer is one shape.
“Replace, Glue, Delegate, Drive. Seven canonical consulting loops, four mechanically distinct shapes of automation. The right move is one shape per loop, not one shape for all seven.”
Framework calibrated against the consulting workflow inventory in /Users/matthewdi/ai-for-consultants/consulting-business-workflow.md (Phases 4-6) and the eleven tool categories enumerated in /Users/matthewdi/ai-for-consultants/tech-tools-landscape.md
The four shapes
What “automate” mechanically means.
Each shape is a different mechanism for getting a chore off your calendar. The differences are not branding; they are mechanical. They handle different trigger types, fail in different ways, and cost different things.
1. Replace
buy a suite that owns the loopPick a suite (HoneyBook, Dubsado, Bonsai, Ignition) that claims to own proposals, contracts, invoices, and the client portal end to end. Migrate your stack into it. The suite owns the data and the cadence. The trade you make is: one cadence for everything, the suite's cadence. Wins on a single repeatable engagement; fails when your seven loops have seven different clocks.
2. Glue
wire your existing tools with webhooksKeep your stack and stitch it together with Zapier, Make, n8n, or hand-rolled webhooks. Each loop becomes a graph: trigger, filter, branch, action. Wins on clean events with stable APIs (a DocuSign signature, a Stripe payment). Fails on fuzzy triggers (a phrase in a transcript, a number in an email body) and on per-instance tuning (different dunning tone per invoice, different status template per client).
3. Delegate
hand it to a humanHire a virtual assistant, an executive assistant, or a fractional ops person. The loop runs on their calendar, in their judgment. Wins on human-judgment chores (the rate-review call, the sensitive-tone followup, the vendor phone call held by a human voice). Fails on response-time loops because the VA's clock is their working hours, not the loop's close window.
4. Drive
software reads your screen and typesSoftware that reads what is on your screen and types into your existing apps the way you do, from a plain English instruction file. No new database. No API rotation. The instruction file is the source of truth; you edit it in TextEdit and the next event uses the new behavior. Wins on fuzzy-trigger loops (where the meaning lives in content, not in an API field) and on per-instance-tuned cadences. Fails on chores that genuinely need human judgment in the moment.
The seven loops
What you are actually trying to close.
Before you can pick a shape, you have to know what loops you are picking shapes for. The canonical inventory of a solo or boutique consulting practice is seven recurring chores. Each has a distinct trigger and a distinct close window, and each trigger fits into one of four types: clean, fuzzy, calendar, or human-judgment.
| Loop | Trigger | Type | Close window |
|---|---|---|---|
| Post-call admin | zoom_call_ended + transcript content | fuzzy | 60 seconds |
| Invoice on milestone | milestone signoff phrase in inbound email OR deal stage move | fuzzy | same business day |
| Dunning | invoice age = Net+7 / Net+14 / Net+21 | calendar | the day a threshold trips |
| Client onboarding | contract_signed (DocuSign / PandaDoc webhook) | clean | same calendar day |
| Weekly client status | Friday 5pm per active client + week's content | fuzzy | end of week |
| Monthly bookkeeping | last business day of the month | calendar | first business day of next month |
| Quarterly pipeline review | quarter_end + judgment about which clients to chase | human judgment | first week of new quarter |
clean trigger
A clean event a webhook can fire reliably.
fuzzy trigger
An event whose meaning lives in the content (a phrase in an email, a number in a transcript), not in the API.
calendar trigger
A cron-style event tied to a date arithmetic.
human judgment trigger
An event only a human reading the situation can decide on.
The mapping
Seven loops by four shapes. Which shape wins where.
Read this as: for each loop down the left column, which of the four shapes actually closes it. Teal cells (“closes the loop”) are honest fits. Amber cells are partial fits where the shape works but is sub-optimal. Red cells are wrong-shape pairings that are common but break in predictable ways.
| Loop | Replace | Glue | Delegate | Drive |
|---|---|---|---|---|
Post-call admin fuzzy trigger | wrong shape | wrong shape | partial fit | closes the loop |
Invoice on milestone fuzzy trigger | wrong shape | wrong shape | partial fit | closes the loop |
Dunning calendar trigger | partial fit | closes the loop | partial fit | closes the loop |
Client onboarding clean trigger | partial fit | closes the loop | partial fit | closes the loop |
Weekly client status fuzzy trigger | wrong shape | wrong shape | partial fit | closes the loop |
Monthly bookkeeping calendar trigger | partial fit | closes the loop | partial fit | closes the loop |
Quarterly pipeline review human judgment trigger | wrong shape | wrong shape | closes the loop | partial fit |
Two things stand out. First: no single shape closes all seven loops. Replace closes one well; Glue closes two well; Delegate closes one well; Drive closes four well and is passable on the rest. Second: every loop has at least one shape that wins on it, so the right move is shape-per-loop, not shape-for-all.
The Drive shape, executing
What happens when a fuzzy trigger fires.
The Drive shape is the unfamiliar one of the four for most consultants, so it is worth walking through one execution. Below is what fires the moment an inbound email matches the milestone signoff phrase declared in the invoice handler. The handler reads the screen, types into QuickBooks, runs an amount-vs-SOW hold check, and writes an audit entry, all without an API key for QuickBooks.
What 'Drive' looks like when an invoice loop fires on milestone signoff
The thing a Glue graph cannot do here is the second-to-last step: the hold check. The amount-vs-SOW comparison reads the SOW you uploaded to Drive, the timesheet you pulled from Toggl, and the line items the QuickBooks form expects. A Zap can fire on a webhook, but it cannot read the screen and decide. That is the line where Drive starts and Glue ends.
The cost of picking wrong
What happens when you give a loop the wrong shape.
The failure mode is not loud. The Zap does not error; it just runs on the wrong cadence. The VA does not refuse; they just close the loop on Friday at 3pm instead of Tuesday at 2pm. The suite does not crash; it just makes you migrate your stack and gives you a default Net+30 dunning schedule that you cannot tune. Five common wrong-shape pairings, each from a real consulting back office:
Five wrong-shape costs you can predict in advance
- Wrong-shape cost on post-call: a fuzzy trigger handed to Glue runs on a daily cron, the call summary lands 18 hours later, and the client is no longer in the room. The cost grows roughly 4x per delayed day.
- Wrong-shape cost on milestone invoicing: a fuzzy trigger handed to Replace fires off the deal-stage move that the consultant always forgets to log on Friday. Net invoice slip: 3-5 business days, every cycle.
- Wrong-shape cost on dunning: a calendar trigger handed to Delegate runs on the VA's hours, not the day Net+7 trips. The day Net+14 trips, the VA is on a Monday, and the chase email goes out at Net+17 in a tone that has already escalated.
- Wrong-shape cost on quarterly pipeline review: a human-judgment trigger handed to Glue produces a generic CSV export every quarter that nobody reads. The actual rate-review decision never happens.
- Wrong-shape cost on onboarding: a clean trigger handed to Replace forces you to migrate your CRM into a suite you do not want, just to get the contract_signed webhook the suite already supports natively.
The pattern: the cost of a wrong-shape pairing always shows up as a delayed close, not as a visible error. That is what makes it slow to spot. The Zap that runs on the wrong cadence looks fine in the dashboard; the chase email going out at Net+17 looks like “the VA is doing their job”; the suite's default Net+30 looks like a sensible policy. The diagnostic is always: which clock did the loop actually close on?
The apps in play
Where each loop's trigger and target actually live.
The reason Drive matters for the fuzzy-trigger loops is that the trigger lives in the content of an existing app, not in an API field of a suite you migrated to. A milestone signoff phrase lives in Gmail. A proposal amount lives in a Zoom transcript. A SOW expected number lives in a Drive document. A handler that can drive Gmail, Zoom, HubSpot, and QuickBooks without API keys can read all four; a Zap can read the API surface of each, which is a strict subset of the screen.
Gmail
fuzzy phrase trigger
Zoom
transcript content trigger
HubSpot
deal stage + activity log
Pipedrive
alternate CRM target
QuickBooks
invoice draft target
Stripe
invoice send + dunning target
Calendly
next meeting invite
Notion
action items + status drafts
DocuSign
contract_signed webhook
Google Drive
client folders, summaries
Toggl
timesheet pull for invoices
Slack
approval queue DMs
The four-step rollout
How to apply the framework on Monday.
Do not try to redesign your back office in one weekend. The honest sequence is four steps; you can run the first three on a Monday afternoon and start the fourth the same week.
The four-step shape-per-loop rollout
- 1
Step 1: Inventory
List your loops. Tag each one as clean, fuzzy, calendar, or human-judgment.
- 2
Step 2: Match shape
For each loop, pick the shape that genuinely closes it. Most consultants pick one shape for all seven; that is the failure mode.
- 3
Step 3: Drive the fuzzy ones
Post-call admin, milestone invoicing, weekly status. These need content matching; they need Drive.
- 4
Step 4: Glue the clean ones, keep judgment human
Onboarding webhook can be Glue. Quarterly pipeline review stays with you. Dunning is Drive with human approval over a threshold.
The unit of progress is “one loop running on its right shape,” not “the whole back office automated.” You will get to four loops on the right shape inside a month, and the remaining three at the next time their trigger fires (the next contract signed, the next end-of-month, the next quarter end).
Where each shape genuinely wins
The honest fit for each of the four.
When Replace is the right answer
Your practice has one repeatable engagement on one cadence: proposal Monday, deposit Tuesday, deliver Friday, invoice on signoff. Weddings, fixed-scope brand work, single-modality coaching. The seven-loop framing does not apply to you in the same way; you have one loop, repeated. Buy HoneyBook or Dubsado; do not write seven instruction files.
When Glue is the right answer
The trigger is clean (a webhook, a payment, a signed contract) and the action is small. DocuSign signed -> Drive folder + welcome email is one Zap and it works for years. Glue starts to fail the moment the if-this-then-that branching tree has to read content rather than respond to events.
When Delegate is the right answer
The chore is genuinely human-judgment: the sensitive client tone, the rate-review conversation, the vendor phone call held by a human voice, the once-a-quarter pipeline meeting where you decide which dormant clients to chase. A part-time VA or fractional ops person is the right shape; do not try to Drive judgment.
When Drive is the right answer
The trigger is fuzzy (content in an email, a phrase in a transcript, a number that has to be cross-checked against a SOW) or the cadence has to be tuned per invoice / per client. Post-call admin, milestone invoicing, tone-graduated dunning, weekly client status. Four of the seven canonical loops sit here. Most of the wins from automating a consulting back office come from getting these four onto the Drive shape; the rest is Glue and a part-time VA.
“The reframe that broke the loop for me was: stop asking which tool to use, start asking which shape each chore wants. Once I tagged my seven chores by trigger type, the answer wrote itself. Onboarding stayed on a Zapier graph. The quarterly pipeline review stayed with me. The four fuzzy ones (post-call, milestone invoice, dunning, weekly status) all moved onto plain English instruction files, and Friday afternoons stopped being a back office block.”
Want help mapping your seven loops onto the four shapes?
Twenty minutes on Zoom. We sketch your seven loops, tag each by trigger type, and pick the shape that closes each one. You walk away with the first instruction file already firing.
Common questions about how to automate a consulting back office
What does it mean to automate a consulting back office, in one sentence?
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).
What are the four shapes of automation in plain terms?
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.
Why do four shapes matter? Why not just pick the best one and use it everywhere?
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.
Is the 'Drive' shape just a fancy way of saying 'AI agent'?
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).
How is 'Replace' different from just buying a CRM?
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.
When is 'Glue' (Zapier or Make) the right answer?
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.
When is 'Delegate' (a virtual assistant) the right answer?
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.
When does 'Drive' actually win, then?
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.
Can I mix shapes? Or do I have to pick one?
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.
Where does Clone actually fit in this framework?
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.
What is the smallest version I can run on Monday?
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.
What about cost? A VA is $3-6K, Drive is $49/month. Is the comparison fair?
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.
More on which shape closes which loop in a real consulting practice
Keep reading
Consulting Back Office Automation
The seven loops, named, with their close windows and decay rates. The companion piece that goes deeper on the specific loops once you've picked the shapes.
Post Call Admin Automation
The fuzzy-trigger loop that pays back fastest under Drive. Walked second by second from zoom_call_ended to the held-for-review queue.
Solo Consultant Batched Admin Block
The Friday batched admin block is what happens when you apply one shape to all seven loops. Diagnostic plus a 30-minute working version.