Before I automated it, my client intake looked like everyone's: an inquiry lands, I skim it, open the website, poke around LinkedIn, start a notes doc, copy details into it, and eventually assemble a proposal from the last proposal with the names changed (carefully). Perhaps forty-five minutes per serious inquiry, spread across interruptions, with quality depending on how my afternoon was going. Now the same inquiry arrives and, by the time I sit down to it, there is a structured brief and a draft proposal skeleton waiting. Here is the pipeline, honestly, including the parts that stay human.
The Pipeline
Every automated step produces a draft artifact. Nothing reaches the prospect without the last, human step.
Step 1: Structured arrival. The single highest-leverage change is making inquiries arrive structured instead of as prose. On this site, that is the fit quiz and the contact form: they capture business type, team size, the problem in the prospect's own words, and rough budget expectations. A form tool, or fields in your existing contact flow, does the job anywhere. Unstructured email can still enter the pipeline; it just goes through the parser first.
Step 2: Parse and structure. A workflow in n8n catches the submission and sends it to Claude via the Anthropic API with a fixed prompt: extract who they are, what they are asking for, stated constraints, and anything ambiguous worth asking about on a call. Ambiguities are gold; they become my first three discovery questions.
Step 3: The research pass. The workflow pulls what is publicly available (the company website, mainly) and asks the model to summarize what the business appears to do, who it serves, and where the requested automation would plug in. This pass is deliberately shallow and clearly labeled as machine-gathered in the brief, because its job is to save me the first twenty minutes of tab-opening, not to be right about everything. I verify anything that matters on the call.
Step 4: Brief plus proposal skeleton. The model assembles everything into two artifacts: a one-page brief (who, what, constraints, open questions) and a draft proposal built from my actual service structure, with scope options mapped to what the inquiry described. The proposal draft is maybe 70% right, which is exactly the point: editing a 70% draft takes ten minutes; producing it from a blank page took most of the original forty-five.
Step 5: The human part. I read the brief, take the call, and rewrite the proposal where judgment lives: pricing, sequencing, and the honest "here is what I would not do" paragraph that closes more deals than any feature list. The pipeline also books the call itself; the quiz hands its results to Cal.com so the context arrives attached to the meeting.
What This Actually Changed
Three things, in order of importance. Response time collapsed: a structured inquiry gets a substantive reply the same day, because the twenty minutes of prep I used to defer no longer exists, and speed is the most underrated close-rate lever a service business has. Consistency went up: inquiry number forty gets the same quality of attention as inquiry number four, because the pipeline does not have bad afternoons. And discovery calls got better: walking in with the open-questions list means the call spends its time on judgment instead of information gathering.
What it did not change: the close. Prospects buy from the paragraph the machine cannot write, the one where a human demonstrates they understood the specific business. The pipeline's job is to make sure I have the time and context to write it.
Build Notes If You Want This
Start with steps 1 and 4 only: structure the arrival, and template the output. That pair alone, no research pass, no enrichment, captures most of the value for an afternoon of setup. Keep every output a draft, log every run, and follow the same review discipline from the 30-day plan: read everything for the first month, then relax review where the error rate has earned it. The intake workflow is also the single best first agent for a service business, because every input is already someone asking to give you money.
Sources
- n8n, workflow automation platform used for the pipeline orchestration. n8n.io
- Anthropic, API documentation (parsing, summarization, and drafting steps). docs.anthropic.com
- Cal.com, scheduling platform receiving quiz context via booking URL. cal.com
Want this pipeline pointed at your inbox?
Client intake is the workflow we build most often, because it pays for itself with the first inquiry it saves. We wire it to your forms, calendar, and proposal format.