2026-04-11
A walkthrough of how Five8Five used Claude Code to build an automated client intake system — what we built, how it works, and what it cost.
One of the questions we get most often is: what does an AI implementation actually look like in practice? Not the theory — the actual steps, the actual timeline, the actual result.
This is one of those walkthroughs.
A professional services firm came to us with a straightforward problem: their client intake process was eating 3–4 hours a week of staff time, and it was still producing inconsistent results. New clients filled out a form, someone manually read it, copied the relevant fields into their CRM, sent a confirmation email, and notified the right person internally. Every single time. For every single client.
We built them an automated system in two days. Here’s exactly what we did.
The firm runs a boutique advisory practice — around 15–20 new client intakes per month. Their intake form collected everything they needed: name, company, service area of interest, timeline, budget range, how they heard about the firm, and a free-text field for context.
The issue wasn’t the form. The issue was everything that happened after someone submitted it.
A staff member would open the response, read it, manually enter the data into their CRM (HubSpot), create a deal record, draft a confirmation email to the client, and send an internal Slack notification to the right advisor. Start to finish: about 12 minutes per intake. At 20 intakes a month, that’s 4 hours of manual, repetitive work — done by someone whose time is worth significantly more than that.
They’d also been burning leads. When the responsible staff member was out, intakes piled up. A prospective client who submitted a form on Friday sometimes didn’t hear back until Tuesday.
The system we built runs this workflow automatically:
Form submission
↓
Claude reads the intake response
↓
Extracts structured fields (name, company, service, timeline, budget, source)
↓
Creates/updates CRM contact and deal record in HubSpot
↓
Generates personalized confirmation email → sends to client
↓
Posts internal Slack notification to the right advisor channel
↓
Flags any incomplete or unusual submissions for human reviewThe whole chain runs in under two minutes from form submission to CRM entry and client confirmation.
The core of the system is Claude processing each intake response. Rather than rule-based field extraction — which breaks when responses are unexpected — Claude reads the free-text responses intelligently.
If a client writes “we’re looking to get started sometime in Q3, budget is somewhere around 50k but we have flexibility,” Claude extracts that as timeline: Q3 2026, budget: ~$50,000, flexibility: yes. A traditional parser would fail on that sentence. Claude handles it the same way a human would.
That extracted data populates the HubSpot contact and deal via API. The confirmation email is generated based on the specific service area the client mentioned — someone inquiring about tax advisory gets a different email than someone inquiring about M&A support — and sent automatically.
The Slack notification goes to the channel for the relevant practice area, tagging the advisor who handles that type of work. If the submission is outside normal parameters (no budget given, unusual service request, something that looks like it needs a human judgment call), it gets flagged to a review queue instead of auto-processed.
Day 1 — Discovery and build.
Day 2 — Integration and testing.
Total build time: approximately 14 hours across two days.
One month after launch:
Time saved: The staff member who managed intake went from spending 4 hours a week on the process to spending roughly 20 minutes — reviewing the flagged exceptions and spot-checking a sample of processed intakes.
Response time: Average time from form submission to client confirmation email dropped from 6–18 hours (depending on when someone got to it) to under 2 minutes, 24/7.
Data quality: CRM records were cleaner and more complete than before. Manual entry introduced inconsistencies — abbreviations, missed fields, different formats for the same data. The automated system applies the same structure every time.
Missed leads: Zero in the month following launch, versus a handful the prior month.
Implementation: one day of consulting time. Ongoing operational cost: API calls to Anthropic’s Claude API, which at their volume runs approximately $15–25/month.
The staff time saved in the first month alone — roughly 12 hours at a fully-loaded rate — returned the implementation cost in full. From month two onward, it’s pure margin.
Not every implementation goes this smoothly. A few things made this particular project fast and clean:
The workflow was already well-defined. The client knew exactly what the process was supposed to be — we just needed to automate it. When clients come in with vague processes, there’s a definition phase before any building starts.
The inputs were consistent. Their intake form was structured and the fields were clear. Claude still handled the natural language well, but it wasn’t starting from scratch with unstructured text.
They were ready to trust the output. The client was willing to let the system run without approving every record — with exceptions routed to a review queue. Clients who want manual approval on every output can get it, but it slows the ROI.
Want something similar for your practice or business? The intake automation pattern works across professional services, agencies, healthcare, and anywhere that currently has a human reading form responses and manually entering data. Book a free review and we’ll map out what it would look like for your specific workflow.
Or if you want to understand the broader landscape of what AI agents can do, our complete guide to AI agents for small business is the right place to start.
Book a free AI review and we'll map out exactly where to start.
Book a Free Review