A consulting follow-up has to do five jobs at once. Drip sequencers do one of them honestly. Here is the rest of the decomposition.

It means closing each of the five separable jobs of a follow-up (citation, voice, timing, reply parsing, rule application) on the right surface for that job. The mistake almost every guide makes is treating the work as one task and recommending one tool. A drip sequencer is honest at one job out of five. A virtual assistant is honest at four out of five but pays in hours and lag. Software that drives your existing apps from a plain English file is honest at all five and runs on your clock, not anyone else's.

Because the architecture of a drip sequencer is a template body plus merge fields plus a calendar cadence. That architecture is honest about exactly one of the five jobs (parsing a clean reply and pausing the cadence). It cannot cite the call (the transcript is not in scope), it cannot sound like you on a Wednesday morning two months from now (the template is frozen at creation time), it cannot fire on the Friday board meeting the client mentioned (only on day-N arithmetic), and it cannot apply your implicit rules without you turning each rule into an explicit branched workflow. None of these are bugs, they are the architecture. Asking a sequencer to do those four jobs is asking the wrong tool.

Because the VA's clock is the VA's working hours, not the client's signal calendar. If a follow-up should fire two days before the board meeting the client mentioned, and the VA is part-time Monday to Wednesday, the email goes out the wrong day in a tone that has already drifted. A VA is excellent at the other four jobs (they will read the call, write in your voice with a little training, parse a reply with judgment, and apply your rules). Cost and response time are the trade. A 24/7 software handler closes the timing job that a part-time human cannot.

It observes the implicit rules in your sent mail and surfaces them for one-time confirmation, then applies them forever. The how-it-works panel on the home page (src/components/how-it-works.tsx, step 03) shows the literal pattern: 'Noticed a pattern in your last 12 kickoff emails: you start with a personal line, you always attach the SOW PDF, you cc your assistant when the contract is above $10K. Should I apply this template going forward?' That observation is rule extraction, not rule writing. You confirm once. The rule moves to a plain markdown file in ~/.clone/memory/ and runs on every same-stage send after that. You never built the path. You never wrote the trigger. You said yes once.

The instructions file declares the trigger in the same plain language you would use to describe it to an employee. Examples from real consulting practices: 'two business days before any board meeting the client mentioned on the last call', 'the morning the new GC starts, if the start date appeared in any thread', 'fourteen days before the contract renewal date in the SOW PDF', 'three days after a milestone signoff phrase appears in inbound mail'. Clone reads the instruction, watches the calendar and the inbox for the named events, and fires when the event happens. There is no path-builder UI. The instruction file is the source of truth.

Plain markdown, eight to twenty lines per stage, lives at ~/.clone/memory/rituals/. A typical solo practice declares five to seven stages: discovery, proposal sent, mid-engagement pulse, post-engagement reactivation, ghosted nudge. Each stage names what to pull context from (the call transcript, the proposal PDF, the last weekly status thread), what voice source to mimic (your last twelve same-stage sends), and any conditional rule (cc assistant above ten thousand, skip if replied in the last thirty days). Editing the file is how you change the behavior. There is no dashboard. That is the point.

Then Job 1 (citation) is harder. The follow-up can still cite the inbound email thread, the proposal PDF, the SOW, the last weekly status, the last quarterly review. Citation does not require a transcript, it requires the conversation surface to be in scope. The transcript is the cleanest surface for post-call follow-ups, but the proposal text is the cleanest surface for proposal-sent follow-ups, and the prior thread is the cleanest surface for ghosted nudges. The point is that follow-ups need a real surface to quote from, not merge fields.

Yes. Clone drives Gmail and your existing CRM through the screen, not through their APIs. HubSpot, Pipedrive, Folk, Copper, or a Google Sheet you treat as a CRM all work. Switching CRMs is a one-line edit of the rituals file. The product is explicit that it does not own the data layer (architecture.tsx lines 56 to 59: '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.'). Removing Clone leaves your business intact.

The other guide (/t/consulting-follow-up-automation) goes deep on Job 1 (citation, specifically the post-call transcript-lift mechanism) and Job 2 (voice mimicking from the outbox). This page is broader and more of a decomposition: it argues that follow-ups are five jobs, names them, and shows which automation shape closes each. If you already know you want transcript-lift specifically, read that one. If you have not yet decided whether a drip tool is the right shape at all, this page is the upstream question.

/t/consulting-milestone-followup-automation is about Job 3 (timing) for the specific case of post-deliverable cadences (day three, day seven, day fourteen) with timeline-shift language. This page covers Job 3 across all signals (board meetings, GC starts, contract renewals, deal-stage moves), not just calendar arithmetic after a milestone. Both can run together; in a mature practice, they typically do.

Pick one stage. Just one. The most common is post-call to existing client. Write four lines of a rituals file: which transcript tool to read from, what to quote, which voice source to use (your last twelve same-stage sends), and one conditional rule (the cc-above-threshold rule is a good one to start with because almost every consultant has an implicit version of it already). Run the ritual once after your next call. Read the draft. Tweak the one line that was off. The fix goes back into the file. By the end of week one, that one stage is producing drafts you do not change. From there, add a stage a week.

It is honest on the four jobs where they overlap (citation, voice, reply parsing, rule application). The VA can do all four with judgment and a learning curve. The software can do all four for one or two orders of magnitude less, on a 24/7 clock. It is not honest on the rare follow-ups where the right move is a human voice on a call (a sensitive client renewal, a contract dispute, a hard pivot conversation). For those, keep a part-time VA or a fractional ops person and let the software handle the routine four jobs. Most healthy stacks are $49 to $129 per month of software plus a part-time human for the human-shaped follow-ups, not one or the other.