Lead Intake Automation: A Build Breakdown

This is a build breakdown, not a client story. Catalyst Consulting is new, and we won't dress up an illustrative example as a named result — when we have a client case study, it will be a real one, with their permission. What this post does instead is show you exactly how a lead intake automation gets built, so you can judge the work on its mechanics.

Lead intake is the workflow we most often recommend as a first project. It's frequent, it's painful when it's slow, and the rules are clear enough to build cleanly. Here's the whole thing, start to finish.

The problem being solved

A small business gets inquiries through several channels — a website form, a shared inbox, maybe a phone call logged by hand. Each new lead needs the same handling: capture the details, figure out who it is, decide how urgent it is, send a fast first response, and log it somewhere the team will actually look.

Done by hand, this depends on someone being at their desk and remembering. Response times swing from minutes to days. Leads get dropped. The fix is not "respond faster" — it's removing the dependency on a person remembering.

The workflow, step by step

The automation runs as a defined sequence. Here's each step and the decision behind it.

1. Capture. A single entry point receives leads from every channel — form submissions, forwarded emails — and normalizes them into one consistent shape: name, contact, message, source, timestamp. Normalizing here means every later step can be simple, because it only ever sees one format.

2. Enrich. The raw lead is thin. An enrichment step adds context — company, rough size, anything publicly available that helps qualify. This is bounded on purpose: enrichment is useful, but it's also where cost and creepiness can creep in, so it pulls a defined, minimal set of fields and stops.

3. Classify. Here the language model earns its place. It reads the message and classifies it: is this a real prospect or a vendor pitch, roughly how urgent, which service it points at. The prompt is strict — classify into these categories, never invent a value, flag anything ambiguous for a human. The output is structured so the next step can branch on it.

4. Draft a response. For genuine leads, the model drafts a first reply — acknowledging the specific thing they asked, not a generic auto-responder. The draft is written in the business's voice, from a small prompt library built during the engagement. Crucially, it's a draft.

5. Route and log. The lead and its draft land where the team works — a CRM record, a channel notification, an email. Urgent leads get a louder signal. Everything is logged, so nothing depends on memory and nothing gets dropped.

Where the human stays in the loop

This is the part that separates an automation you trust from one you don't.

The system does not send the response by itself. It prepares everything — the classification, the enriched context, the drafted reply — and puts it in front of a person who approves, edits, or rejects with one action. The slow, error-prone parts (noticing the lead, gathering context, writing a first draft) are automated. The judgment part stays human.

That's deliberate. A fully autonomous version is buildable, but for a first project most businesses want to watch the system be right for a few weeks before they let it act on its own. The approval step makes that possible — and once trust is established, removing it later is a small change, not a rebuild.

What the client gets at handoff

The build is only half the engagement. At handoff, the business receives:

• The workflow itself, running in their accounts — not ours.
• Plain-language documentation: what each step does, what the common failure modes look like, how to fix them.
• The prompt library, as editable assets they own, so the classification and drafting can be tuned as the business changes.
• A training session walking the team through operating it, including how to read the logs and what to do when something looks wrong.

After that, the support window covers the inevitable early edge cases. When it closes, the business runs this without us. That's the deliverable — not the workflow alone, but the workflow plus the ability to keep it.

The honest caveats

A breakdown owes you the limitations too.

This works because lead intake has clear rules. A workflow that depends on unwritten judgment is a different, harder build. Enrichment quality depends on what's publicly available — for some businesses it adds a lot, for others a little. And the classification step needs a few weeks of real leads and review to tune well; the first-week version is good, not perfect, which is exactly why the human-approval step exists.

None of that changes the shape of the build. It just means the first month is a tuning period, not a victory lap — and an honest engagement is scoped that way from the start.

If lead intake is a bottleneck in your business, this is the kind of workflow a discovery call is built to scope. Bring how leads reach you today, and we'll map what this would look like for your operation.