Two of six phases. The other four are still yours.

The taxonomy

A consulting business is six phases, not one workflow.

The six-phase shape below is the structure inside the Clone source repo's own consulting business workflow doc, a 241-line internal reference. It is the structure of any independent consultancy from one person to about ten, regardless of niche, because each phase has a different cadence, a different output, and a different question being answered. Phase 1 happens once, phase 2 runs all year, phase 3 fires per client, phase 4 fires per engagement and per week, phase 5 fires at engagement end, phase 6 runs on monthly and quarterly clocks.

The honest map

Two phases AI touches. Four phases it does not.

When you put the six phases above next to the six features Clone ships in its features.tsx file, the shape is uncomfortable for a marketing site but clean for a consulting buyer: zero features touch phases 1, 2, 3, or 5, four touch phase 4, two touch phase 6. The feature list will not pretend otherwise because the product is architected not to pretend otherwise. The first founding principle (runs on your machine) and the third (tool agnostic by design) are the architectural reasons the product cannot quietly slide into the unautomated phases without consent.

What it does

The two phases, walked.

Phase 4 (delivery) and phase 6 (ongoing operations) share one structural property: they fire on recurring events with predictable triggers and consistent close windows. A Zoom call ends. A milestone gets signed off. Friday hits at 9am. The first of the month rolls over. An invoice ages past Net+7. Each event has the same handler week to week, only the inputs change.

That repetition is the property that lets a recurring AI loop close the work. The handler does not have to invent the right action, it only has to apply the right rule to fresh inputs. Below is what a Monday at 8am looks like once the phase 4 weekly admin loop and the phase 6 monthly cycle are running.

Each line is a real operation against a real app on your machine. Toggl gets read. QuickBooks gets opened. Six invoices get drafted at the rates declared in your engagement records. Five get sent because their amounts match the SOW expected number, the sixth holds in your inbox because the variance is over ten percent and the rule says hold for review. HubSpot gets six deal updates with notes drawn from last week's actual activity. Six status email drafts wait in Gmail. The client health board in Sheets refreshes with current pipeline, current outstanding, and the next thirty days of renewals.

This is one event firing one handler. The other recurring events look the same. Zoom call ended fires the post-call loop, summary plus CRM log plus follow-up draft inside sixty seconds. Milestone signoff fires the invoice loop. Net+7 fires the dunning ladder. None of these handlers tries to do phase 1, 2, 3, or 5 work because none of those phases produce a recurring trigger.

What it does not do

The four phases that stay yours.

The temptation when you build a tool that automates two phases is to claim coverage of all six. The reason this product does not is that the value of the relational phases compounds in ways that are damaged, not helped, by AI involvement.

Phase 1 is one-time and judgment-dense. The questions are which structure (LLC for tax flexibility plus liability protection in most cases), what professional liability coverage your niche actually needs, what your three service tiers should be, where you set the rate. None of these are recurring decisions and getting them mechanically wrong by automation costs you more than the time saved. The market rate for advisory-only work in most niches is $150 to $300 per hour, the same range you would pay a CPA or a lawyer to give you direct guidance on these questions for a few hours total. Pay them.

Phase 2 is where AI looks most appealing and is most actively damaging. You can absolutely use a tool to write 200 cold emails an hour, the response rate floor on those is approximately zero. The honest mechanism is fewer emails, a personalized first three to four sentences each, a specific observation about the prospect's business in line one, and the discipline to walk away after four touches. Personalized cold outreach in this category lands roughly five to fifteen percent reply rates. That is a per-prospect ten-minute research investment, not a recurring AI loop.

Phase 3 is where consultants leave the most money on the table when they hand work to a tool. The leverage in a discovery call is the silence after you ask a real question. The leverage in a proposal is restating the prospect's problem in their own words in line one of the executive summary. The leverage in proposal presentation is stating the price and stopping talking, then handling the objection that comes with a specific scope adjustment rather than a discount. None of those compress into a prompt.

Phase 5 is closeout and it is the highest-leverage phase per minute spent in your entire business, because it is where retainers, testimonials, referrals, and case studies all originate. The closeout meeting question (did we accomplish what you set out to do, what worked, what could be improved, what is next for you) opens three doors at once: the retainer pitch, the referral ask, and the case study angle. A bot pinging a satisfaction survey closes all three.

The six features

Where each feature lands on the phase map.

The six features below are the literal contents of the marketing features file, src/components/features.tsx, lines 13 through 62. Each one is annotated with which phase or phases it actually touches.

The covered surface

Specifically, the recurring events that close themselves.

Six handlers fire on six recurring events. None of them require a workflow graph editor and none of them require a CRM migration. Each handler is a sentence of plain English that runs against the apps you already use.

Why this scope, not more

The product is architected to stop at phase 4 and 6.

In the architecture doc that ships on the marketing site, the four founding principles are: runs on your machine, your workflows your voice, tool agnostic by design, always reviewable. Each one is a design constraint, not a feature.

Runs on your machine means client files never leave your computer, which means there is no shared cloud where a generic phase 1 LLC-formation flow would live anyway. Your workflows your voice means the product observes how you already work and mirrors that, which is a design specifically tuned to the recurring phase 4 work, not the one-time phase 1 setup or the per-prospect phase 2 outreach. Tool agnostic by design means the product uses the apps you already pay for, which is the operational property of phase 4 and phase 6 work but not of the relational phases. Always reviewable means every action is logged and reversible, which is an asymmetric protection for the recurring loops where a single mis-fire could cascade across an inbox.

Together those four constraints describe a system that can only honestly automate the recurring back-office loops in phases 4 and 6. It is the most useful 33 percent of a consulting business because that 33 percent is the percent that runs every week regardless of where you are.

When to use what

Where this fits against the other options.

A graph automation tool (Zapier, Make, n8n) is the right answer when you have one stable two-step automation, a webhook fires, a row gets created. The phase 4 and phase 6 loops have seven independent triggers and per-loop hold rules, which tips the cost shape away from a graph editor and toward a sentence of instruction. A graph editor will work for one of the seven loops, the second through seventh accumulate maintenance.

A vertical practice-management tool (HoneyBook, Dubsado) is the right answer when your back office is a single repeatable engagement on one cadence, a fixed-scope service with a contract on Monday, deposit Tuesday, deliver Friday, invoice on signoff. Most consulting practices are not shaped that way because each engagement has a different cadence and a different deliverable, which is why the seven phase 4 and phase 6 loops have to fire independently rather than rolling into a single client portal.

A virtual assistant at $3,000 to $6,000 a month is the right answer when phase 2 (finding clients) is the bottleneck and you have continuous queue depth (a dedicated outbound list of 500 prospects, a content calendar, a CRM hygiene project). For phase 4 and phase 6 work the queue is too lumpy for headcount, the events fire on uneven cadences and a recurring loop runs at $49 a month and fires the handler the second the trigger lands.

A non-trivial number of solo consultants run all three: a graph tool for one stable webhook (calendar booking to CRM, for example), a phase-4-and-6 recurring loop for the back-office work, and a fractional VA for one phase 2 project at a time. The point is to match the mechanism to the cadence, not to find one tool that pretends to do six phases.