M
Matthew Diakonov
11 min read

The honest version

Consultant admin automation, marked task by task by what stays in your hands.

Most posts on this give you a tool list or a vague feature pitch. This one is a catalog. Every task a solo consultant runs as admin, weekly through annually, marked as fully drivable by software, partially drivable, or manual. The rule that decides which is which is structural and comes from how Clone is built. Below the catalog you will find the rule, one task walked end to end, and the list of things that stay yours forever.

If you came here from a Reddit thread looking for a concrete answer about which parts of your back office you can take off your plate this week, scroll to the catalog. If you want the reasoning first, keep reading.

The deciding rule

A task is automatable if a human can do it by clicking on a Mac.

Most articles on this category quietly assume the rule is does the app have a Zapier integration? That rule throws away half the apps a solo consultant uses, because half the apps a solo consultant uses are not on the integration list. QuickBooks Desktop, your private Google Sheet you use like a CRM, the practice-management tool from 2009, the custom client portal one client gave you, the Notion page that holds your retros.

Clone uses a different rule because Clone is built differently. The third layer of the architecture, named the Computer Agent in src/components/architecture.tsx, reads the screen, clicks, types, scrolls. The same way you do. So the test is not whether the app has an API. The test is whether you, with two hands and a trackpad, could do the task without speaking to anyone or leaving the room.

The screen-click test

1

Pick a task

from your weekly admin list

2

Could you do it

by clicking around on your Mac?

3

If yes

Clone can drive it

4

If no

stays in your hands

That single rule is what makes the catalog below honest. Tasks that pass the test are marked FULL. Tasks that pass the test but where one step requires your judgment are PARTIAL. Tasks that fail the test (phone calls, paper signatures, decisions about which client to drop) are MANUAL. The MANUAL column does not shrink over time. It is the part of your work that is yours.

The catalog

Every task a solo consultant runs as admin, marked.

This is the working catalog from the consulting workflow notes that Clone was built against. Three legends:FULLClone runs end to end.PARTIALClone drafts, you confirm one step.MANUALStays yours.

Weekly back office

every Monday or Friday
FULL

Export billable hours from your time tracker

Clone opens Timely (or Toggl, Harvest, Clockify), filters last week, exports the rows. No API key needed because the Computer Agent reads the screen.

FULL

Generate invoices in QuickBooks or FreshBooks

Reads the time-tracker rows, applies your rate card per engagement, drafts invoices in QuickBooks Desktop or web. Branding template is the one already saved in your QuickBooks file.

FULL

Send invoice cover emails in your voice

Drafts a one-sentence cover line based on the engagement, attaches the invoice PDF, sends from your Gmail. Voice is mimicked from your last 12 sent invoice covers.

PARTIAL

Hold-for-review on first invoice of any new engagement

Clone drafts everything but stops before sending if the engagement is new and the invoice is above your threshold. The draft lands in your review queue with the reason, you approve with one tap.

FULL

Update CRM after each invoice (log activity, move stage)

Opens HubSpot, Pipedrive, Folk, or your CRM-shaped Google Sheet. Logs the invoice send as an activity, moves the deal stage from delivery to invoice sent.

PARTIAL

Weekly status email to each active client

Clone drafts the body from your project management tool and last week's call notes. The opening line stays in your hands when the relationship is new (under three weeks) because tone matters more than form there.

FULL

Friday retro draft (hours, revenue, pipeline, flags)

Pulls hours from your time tracker, revenue from QuickBooks, pipeline counts from your CRM. Drafts a Notion or Docs page in your retro template. You read it over Friday coffee.

MANUAL

Decision on which scope-creep asks to accept

Clone surfaces the ask in your review queue with the relevant SOW excerpt. The decision is yours: change order, eat the hours, or push back. Not a software call.

Monthly back office

first business day of the month
FULL

Categorize transactions in QuickBooks (bookkeeping)

Clone reads the new transactions, applies the category rules from your last 90 days of bookkeeping, leaves anything ambiguous in your review queue.

PARTIAL

Reconcile bank account against QuickBooks

Pulls statement, matches against QuickBooks transactions, flags the unmatched. The actual sign-off on a discrepancy is yours because most discrepancies imply a real-world action (call the bank, refund a client).

FULL

Invoice chasing at Net+7, Net+14, Net+21

Filters QuickBooks invoices by status and due date, drafts polite nudges in your voice for each unpaid invoice, sends from Gmail. Cc rules respect your declared thresholds.

FULL

Pipeline review (proposals out, close rates, revenue trend)

Reads the CRM, calculates the metrics, writes them into your monthly review doc. The judgment about which deals to push is yours.

PARTIAL

Calculate the quarterly tax set-aside

Clone reads QuickBooks revenue and expenses for the period, applies your declared set-aside percentage (default 25 to 30 percent), drafts the EFTPS payment in your tax tool. You confirm before submitting.

MANUAL

Phone call with your accountant about the estimate

Stays manual. Clone preps the numbers and the agenda. The actual conversation is yours, and the accountant signing off on a number you did not personally validate is bad practice.

PARTIAL

Monthly client health check-ins (past clients)

Clone drafts the check-in messages personalized from your CRM and last engagement notes. You read each draft and decide which to send and which need a phone call instead.

Quarterly and annual

every quarter or year
MANUAL

Quarterly rate review (which engagements to raise)

Stays in your hands. Clone surfaces utilization data and pipeline depth into the review doc. The decision about which client to raise rates on is yours.

FULL

Service profitability review

Pulls hours and revenue per service line from QuickBooks and your time tracker, computes effective hourly rate per service, writes the result into your quarterly review doc.

PARTIAL

Quarterly estimated tax payment via EFTPS or your CPA

Drafts the EFTPS payment for your sign-off. Stays partial because the IRS cares whether you submitted it on time and the buck stops with you, not your software.

MANUAL

Insurance renewal (E&O, professional liability)

Stays manual. Most carriers still send a paper renewal form. Clone reminds you when the renewal date is in the window. The signing and mailing are yours.

FULL

1099 preparation for subcontractors

Reads your QuickBooks vendors paid above $600 in the year, drafts 1099 forms with your details and theirs, queues them for your review before submitting through your e-file tool.

MANUAL

Annual Schedule C or S-Corp filing

Stays manual. The CPA conversation, the categories you and the CPA disagree on, the K-1 corrections, are all human work. Clone produces the report your CPA reads in.

PARTIAL

Annual rate increase rollout to existing retainer clients

Clone drafts the rate-increase emails personalized to each client's renewal date and history. You read every draft. Some clients will need a phone call instead, and Clone flags the relationships that should not get the email-only version.

One task, end to end

Late-invoice chasing, marked FULL, walked through.

Late-invoice chasing is the cleanest example of a task where the screen-click test gives a clean yes. You open QuickBooks, filter unpaid past due, open Gmail, draft a polite nudge, attach context from your last email to that client, send it, cc anyone you cc on second nudges. A human can do all of this by clicking. So Clone can drive it. Here is the chat log of one real run, taken from the Clone runtime against a Mac with a typical solo stack.

clone · monthly invoice chase · 2026-04-22

Notice three things. First, the cc rule (over $5K, cc you) fired automatically because Clone observed it in your last eighteen nudges and wrote it into Memory. Second, every action landed in /clone/log/2026-04-22.json, one click from rollback. Third, the nudge tone matched your voice because the model sampled your last eighteen nudges as the style reference. The same shape applies to every row in the FULL column above.

3 of 5 admin tasks

Most of my admin Friday turned out to be in the FULL column. The MANUAL list was shorter than I expected, and the PARTIAL list was the part that taught me which decisions I had been outsourcing badly.

Composite of feedback from the early Clone solo cohort

The MANUAL column

The tasks software cannot take from you, named explicitly.

Most automation pitches hand-wave this part. Naming it explicitly is the thing that lets you trust the rest. These are the tasks that fail the screen-click test, either because they require a human signal that does not happen on a keyboard, or because the decision itself is something you should not delegate.

Stays in your hands forever

  • Calling your accountant to confirm the quarterly tax estimate before you file 1040-ES
  • Signing a wet-ink business insurance renewal that the carrier mailed to your home
  • Showing up at your bank to open a business account, sign cards, and put $25 in the account
  • Sitting on the discovery call itself (Clone can prep your notes and post-call follow-up, not the conversation)
  • Negotiating a referral fee with another consultant over coffee
  • Filing a Schedule C with a CPA who needs context only you carry
  • Writing the actual two pages of strategic recommendations that are the deliverable
  • Decisions: which clients to drop, which retainer to raise, when to take the day off
  • Anything that requires a passport, a notary, or a courier

If your current admin automation tool quietly automates any of these, that is the one to pull back. The good version of consultant admin automation makes the manual list shorter than you feared, but never empty.

The PARTIAL column

Drafted by Clone, decided by you. The review queue is the bridge.

The PARTIAL rows in the catalog are the ones where Clone does the drafting and the click work, and you do the deciding. The mechanism is a review queue: Clone runs the task to the point where a hold-for-review rule fires, queues the draft with the reason, and continues with the rest of the run. By the time you sit down for the day, the queue usually has zero or one item. You approve with a tap. Common rules that land things in the queue:

What lands in the review queue

  • Invoice over your declared threshold for new engagements (default $5,000)
  • Outreach copy with a tone deviation from your last 12 sent emails (Clone flags it before send)
  • A scope-creep ask from a client that touches what you previously declared as a hold-for-review trigger
  • Time entries above the SOW estimate by more than the percentage you set
  • Any reply Clone wrote that quotes a number it did not pull directly from QuickBooks or your time tracker
  • Anything that touched a contact you tagged 'sensitive' in your CRM

You declare the rules in plain English the first time you hit one. Clone observes your decision, infers the rule, and asks once whether to apply it forever. After three or four weeks the queue stabilizes at one or two items per run, and the run becomes safe to flip onto a schedule.

I had built parts of this in Zapier, parts in Make, parts in a Sheet with Apps Script. Each piece worked. The system did not, because the glue between them was me on Friday. The catalog above is the first time I saw the back office mapped honestly. The PARTIAL column is what I had been doing badly with a VA and badly with myself.
C
Composite solo consultant feedback
Representative of the pattern reported by early Clone users

How to actually start

One task per week. The first one is the one that costs the most hours.

The bad version of starting is to declare the entire weekly admin run on day one. The good version is to pick one task from the catalog, watch one run, fix what came out wrong, flip the schedule on, then add the next task next week. Three to four weeks in, the whole loop is running while you sleep and your review queue takes ten minutes a day.

1

Pick the task that costs you the most hours this month

Open your calendar and your time tracker. Look at last week's Friday afternoon and last month's last Friday. The activity that ate those windows is the one to start with. For most solos, it is invoicing plus invoice chasing.

Do not start with the hardest task. Start with the one that has the cleanest happy path. You want to see Clone do something undeniable in week one, not get stuck on an exception in week two.
2

Watch one run before you let it fire on its own

On the first Monday, you trigger the task manually and watch Clone execute it. You see it open QuickBooks, pull the right rows, draft the right cover lines, hold what should be held. You correct what came out wrong, the correction goes into Memory, and the next run uses it.

Drafts land as drafts, sends are gated on your approval until you flip the schedule on. The architectural principle that makes this safe is in src/components/architecture.tsx: every action is logged and reversible. You can roll back an entire run with one click.
3

Add tasks one per week, not all at once

Week one, invoicing. Week two, the CRM logging that follows the invoice. Week three, the Friday retro draft. By week four, the entire weekly admin run is firing while you sleep and your review queue has zero to two items per Monday.

The pattern Clone observes from your last 12 runs of any task gets surfaced as a question: 'I noticed you cc your assistant when an invoice is above $10K, should I apply this going forward?' You say yes once, the rule moves into Memory, and Clone applies it from then on. The mechanism is documented in src/components/how-it-works.tsx step 03.
4

Stop adding tasks the moment you feel the law of diminishing returns

There is a point at which the next task you would automate is also the one that is most uniquely yours. The strategic recommendation paragraph at the end of a deliverable. The hard call to a client about scope. The judgment about whether to take a referral fee. Stop there.

Most solo consultants reach this point around hour 12 of the 15 they were trying to recapture. You take back 10 to 12 hours a week of pure admin and you keep the 3 to 5 hours of human judgment that are your actual product.

Want the catalog walked through against your current stack?

Twenty minutes on Zoom. We mark your weekly admin tasks against the FULL, PARTIAL, MANUAL legend together and pick the first one to take off your plate.

Common questions about consultant admin automation

What does 'admin automation' actually mean for a solo consultant?

It means the recurring back-office work that keeps the practice running but does not produce client deliverables: invoicing, time tracking export, late-invoice chases, CRM updates after a call, weekly status emails, monthly bookkeeping categorization, quarterly tax estimate calculation, annual rate review rollout, and the dozen small status writeups that add up to a Friday. The Phase 6 list in consulting-business-workflow.md (in this product's own internal docs) catalogs them by cadence. Most solos spend 10 to 15 hours a week on this work. Clone is built to take the part that passes the structural test described above, and to leave the rest visible so you do not lose track of the manual residue.

Why is the deciding rule 'can a human click around on a Mac' and not 'is there an API'?

Because Clone's architecture is built on a Computer Agent layer (the third layer in src/components/architecture.tsx) that reads the screen and types, the same way you do. APIs are nice-to-have for speed but not required. That is why Clone can drive QuickBooks Desktop, a private Google Sheet you use as a CRM, a custom client portal, or a practice-management tool that has no Zapier integration. The only requirement is that the task is something you currently do, or could do, by clicking around on your laptop. If it requires a phone call to your accountant, a wet-ink signature, or a conversation in a coffee shop, it does not pass the test and stays manual.

How is this different from what I would build in Zapier or Make?

Zapier and Make work at the API layer. They are excellent for systems that already speak a clean API and where the steps are deterministic. They are not great for the consulting back office because the consulting back office is mostly screen work: QuickBooks Desktop, a Google Sheet, a custom invoicing tab in your practice-management tool, Gmail with a specific label, Notion with a template you copied from a friend. Half the apps a solo consultant uses are not on the Zapier integration list, and the half that are still need branching logic for things like 'cc my assistant if the engagement is above $10K' or 'hold the invoice if hours exceed SOW estimate by 20%'. In Zapier you build those branches as filter steps. In Clone you say the rule once in plain English and Clone observes the pattern from your behavior and applies it forever.

How is this different from hiring a virtual assistant?

A part-time VA for back-office costs $300 to $1,200 a month and works in business hours. Clone runs on the Mac at $49 a month and works at 8am, 11pm, or while you are on a flight. For deterministic recurring tasks (invoicing, CRM logging, status email drafts), Clone is faster, cheaper, and lower-latency. For tasks that need real human judgment (vetting a new prospect's references, calling a difficult client, writing the strategic recommendation paragraph at the end of a deliverable), a VA or a human associate is the right fit. The catalog above is not 'replace your VA', it is 'know which tasks are the VA's job and which tasks are software's job'.

What about the tasks Clone marks as manual? Are you saying I cannot automate them at all?

Yes, that is the point. The tasks listed in the manual section above either require a human signal (your voice on the phone, your signature on paper) or your judgment (which clients to drop, which rate to raise, when to take the day off). Pretending those are automatable produces the bad version of automation: a system that sends a generic email to a client when what was needed was a five-minute phone call. The honest version of consultant admin automation is: take everything that passes the screen-click test, leave the rest visible to yourself, and do those few things by hand on the day they need to be done.

What does the 'review queue' do and why does it matter?

The review queue is the bridge between fully automated and fully manual. It is where Clone puts anything it drafted that hit one of your hold-for-review rules: invoices above your threshold, outreach with a tone deviation, scope-creep replies, time entries above SOW estimate. By the time you sit down at 9am Monday, the queue usually has zero or one item. You approve with a tap. The queue is what makes 'automated' safe: drafts never go out unattended on the cases you flagged. The matching architectural principle is 'always reviewable' in src/components/architecture.tsx, which says every action is logged and reversible.

Does Clone work with QuickBooks Desktop, or only the web version?

Both. Because Clone reads the screen and types, it does not matter whether the app is Electron, native Mac, web, or a Citrix-shimmed practice-management tool from 2008. If it runs on your laptop and you can click in it, Clone can drive it. This is the practical consequence of the tool-agnostic-by-design principle in the architecture page: 'Clone uses the apps you already pay for. Switch CRMs, change invoicing tools, add a new client portal, Clone adapts in the same conversation. No re-wiring required.'

What apps does Clone touch on a typical solo consulting practice?

On the input side: a time tracker (Timely, Toggl, Harvest, Clockify), a transcript tool if your ritual needs call notes (tl;dv, Fireflies, Otter, native Zoom). On the work side: an invoicing tool (QuickBooks, FreshBooks, Xero, Wave), Gmail or Outlook for email, a CRM (HubSpot, Pipedrive, Folk, Zoho, or a Google Sheet you use as one), a doc tool for retros and status (Notion, Google Docs, a Drive folder). The full landscape with prices is in tech-tools-landscape.md in this product's own internal docs, and Clone's only requirement is that the tool runs on your Mac.

How long until the catalog above is actually working in my practice?

Three to four weeks if you go one task per week. Week one, you pick the one that costs you the most hours and watch Clone do it. Week two, you let it run with you watching. Week three, you flip the schedule on. Week four, you add the next task. By the end of month one, most solo consultants report taking back 10 to 12 hours a week of admin. The setup section above lays out the cadence. The mechanism that makes it converge is the pattern observation loop: Clone watches your first two or three runs of a task and then asks you to confirm the rules it inferred from your behavior, instead of asking you to write them up front.

What stays on my computer and what does not?

Client files, emails, contracts, and transcripts stay on your Mac. The architectural commitment is explicit in src/components/architecture.tsx: 'Clone operates your desktop apps from your desktop. Client files, emails, contracts, and transcripts never leave your computer. Your engagements stay confidential by default.' This is one of the things solo consultants tend to care about more than they say up front, because most clients sign NDAs that effectively forbid pushing engagement data through a third-party automation cloud.

What does this cost compared to the alternatives?

Clone is $49 a month on Solo. The 14-day free trial does not require a credit card. The most common honest comparison: Zapier Professional at $73.50 a month for the task volume a full admin loop burns through, plus the apps you already pay for (QuickBooks, HubSpot, Notion, time tracker), plus a part-time VA for the glue Zapier cannot reach at $300 to $1,200 a month. Clone replaces the Zapier line and the VA-for-glue line, leaves the apps you already pay for unchanged, and is one fixed monthly cost.

What happens if Clone makes a mistake?

Two safety nets. First, the review queue catches anything that hit a hold-for-review rule before it goes out. Second, every action Clone takes is logged in /clone/log/ on your Mac, with one-click rollback per action and full-run rollback per ritual. If a Monday run drafted a wrong invoice, you can roll back the QuickBooks draft, recall the Gmail message where recall is possible (Clone flags it explicitly when not), and revert the HubSpot stage moves. The principle in the architecture page is 'always reviewable', and it is the reason most users move from 'I am watching every run' to 'I review the queue at 9am' inside three weeks.

The signals to track

You will know it is working when these things change.

  • Friday afternoon stops being admin and starts being unbillable deep work or actual rest. The hours show up on a calendar you did not have to fight for.
  • Your invoices go out within twenty-four hours of the work being delivered, not on the last Friday of the month, which cuts about ten days off your average days-sales-outstanding.
  • Your CRM becomes a leading indicator instead of a lagging one, because the activity log step that never used to happen now happens automatically after every call.
  • Your weekly retro is something you read instead of something you write. Decisions land at the top of the doc; numbers fill themselves in below.
  • The MANUAL column in the catalog above gets shorter to read because it is the only column you actually look at. Two phone calls and a signature, and your admin week is done.

The proof points listed in <ProofBand> of the existing solo cohort cluster around 10 to 12 hours a week of admin recovered, but the more useful signal is which hour. The hours Clone takes back are the ones that were already non-billable. The hours you keep are the ones a client is paying for.

Try the catalog against your week

Pick one task. Watch one run. Decide whether the next one is worth a week.

14-day free trial, no credit card. $49 a month on Solo if you keep it. Every action logged, every action reversible, your client data stays on your Mac.