Evan Lemoine

ReplyGen AI internal tooling

Problem

Every new client started the same way: a discovery call, then half a day turning it into a positioning statement, an ICP definition, a Clay table that returns the right accounts, and copy that matches the offer. Real judgment work, but the same judgment work every time, and it pushed the first campaign out by days while the client waited.

Solution

A library of Claude Skills that takes the call transcript and lead form as input and produces the offer, ICP, Clay table structure, and outbound copy, then builds the test campaign itself. The hard part is not generating copy; it is encoding the sequence of judgments an operator makes between a sales call and a launched campaign.

Value

Setup went from roughly four hours to under thirty minutes per client, producing a live test campaign with 1,000 verified contacts. Campaigns now launch the same day as the discovery call rather than the following week.

• Claude Code
• Claude Skills
• Claude API
• Clay
• Instantly
• MillionVerifier

commercial real estate investment & sourcing 5.0 rating

Problem

Deal sourcing ran almost entirely by hand: pulling property and owner records from CRE databases, filtering in spreadsheets, looking up contacts one at a time, retyping into a CRM, and sending outreach individually. Sourcing from those databases also means competing with everyone else who bought the same list.

Solution

Two ingestion paths feeding one schema. The structured path pulls the CRE databases; the proprietary path reads deal news and blog coverage over RSS, where a Claygent agent classifies investment type, scores the opportunity, and extracts the parties involved before the deal reaches anywhere purchasable. Lightbox exposes an API while CoStar and Crexi offer only exports, so normalization happens after ingest rather than at the source.

Value

Eliminated 20 hours per week of manual sourcing, and moved first contact ahead of the databases competitors subscribe to.

• Clay / Claygent
• n8n
• RSS
• CoStar
• Crexi
• Lightbox
• Apollo.io
• Hunter.io
• Close
• Instantly.ai

self-built and operated medical-aesthetic devices

Problem

A capital medical device can only be bought by a practice that owns its own purchasing decision. Hospital systems, national chains, PE rollups and group practices cannot, so the qualification that matters most is ownership structure, and no commercial database exposes it. ZoomInfo and Apollo run $1,000+ per month per seat; the medical-device-specific tools are five figures a year. None of them answer the one question that decides whether a rep should dial.

Solution

A pipeline that scrapes private practices out of Google Maps, disqualifies the ones that can’t buy, reads the owner’s name off the practice’s own website, and delivers the survivors into a local calling app with a map, a call queue and a pipeline board.

The engineering that actually moved the cost is the work done before any paid API is called. Group practices are flagged against the free CMS NPI registry, so no money is spent qualifying a lead that was never sellable. Enrichment tries the site’s own sitemap.xml over plain HTTP first and only falls through to the paid scraper when a site genuinely needs rendering. 76% of them don’t. A SQLite ledger means a practice is never paid for twice, and pagination only requests another page of results when the previous one came back full.

Value

• $0.061 per qualified, enriched, ready-to-call lead: measured spend divided by delivered leads, not an estimate
• $60.26 of lifetime spend produced 3,619 practices scraped, 2,557 surviving disqualification, 1,254 enriched with an owner name, and 982 callable today
• Scraping was 96% of the bill ($57.71); search was $2.54, which is why the free-first layer was worth building
• On the most recent batch, the free-first layer hit 76% and avoided $4.55 of API cost, more than the batch itself cost ($3.67 for 1,069 new practices)
• Projected all-in cost per enriched lead going forward: $0.016–$0.024. This is a projection from measured hit rates, not a completed batch

• Python / Flask
• SQLite
• Serper (Google Maps)
• Firecrawl
• sitemap.xml / HTTP
• CMS NPI registry

cross-client capability · Manage Digital reused across client campaigns

Problem

Client databases were full of leads that had already been paid for and were no longer being worked. Reactivating them by hand across email, SMS, voicemail and chat is repetitive and lands inconsistently, and for a consultancy running the same play for several clients, rebuilding those workflows for each new engagement was its own drag on margin.

Solution

One reactivation engine that treats email, SMS, ringless voicemail and AI chat as channels inside a single journey rather than four campaigns running beside each other. A lead that answers on one channel stops being contacted on the rest, which is the difference between a reactivation campaign and a complaint. The infrastructure is built once and redeployed per client.

Value

• 10+ new customer appointments in under 7 days, from one campaign against a database that had gone cold
• Four channels coordinated on a single journey, not four separate campaigns
• Reusable infrastructure: built once, redeployed per client rather than rebuilt
• A repeatable path from dormant record to conversation to booked appointment

• email
• SMS
• ringless voicemail
• AI chatbots
• lead automation
• multi-channel orchestration

medical software company · sells to healthcare providers HIPAA · SOC 2 · BAA-backed contracts

Problem

The revenue lifecycle ran as manual handoffs across eleven disconnected systems. Prospects saw the product before NDAs were executed. Contracts were assembled by hand against a pricing matrix in a shared drive. Payments landed in Stripe but reached QuickBooks without line-item revenue recognition, so subscription and onboarding revenue were never separated and reconciliation broke.

Solution

One orchestrated lifecycle across the whole stack, each stage gated so the next cannot begin until the prior one completes. HIPAA meant no PHI in automation logs or test data and SOC 2 meant real access control and audit logging, so compliance shaped the design rather than being reviewed at the end.

Value

• 11 systems reduced to 1 gated lifecycle, lead intake through renewal
• Line-item revenue recognition now splits subscription from onboarding revenue, so reconciliation holds
• NDA execution gates product exposure: the legal gap closed structurally rather than by policy

• GoHighLevel
• Make.com
• Stripe + Stripe Tax
• QuickBooks Online Advanced
• Slack
• Zoom
• Google Workspace

entrepreneur · 3 brands real estate · modular development · financial services 4.0 rating

Problem

One operator ran four different sales processes (a property transaction, a development project cycle, a service onboarding, and a lending underwriting flow) across three brands, with leads arriving from five form types, two Facebook pages, three Instagram accounts and multiple websites. Nothing was structured, attributed, or automated.

Solution

A single GoHighLevel environment with separated brand workspaces and four pipelines, each modeling its own real process rather than forcing one funnel onto four businesses. Two of the websites were platform-controlled franchise sites that could not be edited, so lead capture ran on standalone branded landing pages with short URLs instead of embedded forms.

Value

• 4 business lines on 1 attributed system: every lead routed to the correct pipeline and tagged by source and brand
• 11 lead sources consolidated, including 2 sites whose code could not be touched
• Client operating the platform unaided after handoff, following recorded training and a reference guide

“Good experience and solid work, will use again.”

client · 4.0 · endorsed: Committed to Quality, Clear Communicator, Detail Oriented, Accountable for Outcomes

• GoHighLevel
• multi-brand workspaces
• custom pipelines
• fields & tags
• forms & funnels
• landing pages
• Meta Business Suite
• FB/IG Lead Ads
• SMS + email automation
• Loom

sales consultancy · end client's HubSpot same-day-urgent brief 5.0 rating

Problem

A consultancy needed Clay and HubSpot connected in both directions for an end client, and needed it that day. Reading out of HubSpot was the easy half. Writing back was the risk: without a lookup layer, every sync run manufactures duplicates in a production CRM, and duplicates are far more expensive to unwind than they are to prevent. The native integration could not do the field mapping or the conditional logic the job required.

Solution

Four production Clay tables orchestrated through Make.com and direct HTTP calls to HubSpot’s API: two for full import, two implementing the lookup-and-match layer. Ambiguous matches skip rather than merge, because a bad merge is more expensive to unwind than a missing record.

Value

Delivered end to end in roughly 31 hours from contract start, with duplicate-safe bidirectional sync live in the client’s production HubSpot instance across four tables.

“Evan was great to work with, knowledgeable, fast, and clearly experienced with Clay and Make.com. He quickly set up complex HubSpot integrations exactly as needed. Highly recommend for any Clay work.”

client · 5.0

• Clay
• HubSpot Contacts / Companies / Search API
• Make.com
• direct HTTP/JSON

digital publisher · multiple brands $500/month operating ceiling 5.0 rating

Problem

A digital publisher was capturing leads across several brands, but the process broke down after submission. Each brand needed its own identity, its own inbox and its own response path, and different lead types needed different follow-up. Routine pre-purchase questions were being answered by hand, while legal questions and pricing exceptions genuinely needed a person. The whole system had to run under $500 a month.

Solution

Source-tracked forms identify the originating brand and the lead type, then route each prospect into the correct CRM path and follow-up sequence. An AI conversational layer handles routine questions and early-stage follow-up.

Legal questions and pricing exceptions escalate to a human by rule rather than letting the model improvise, which is the decision that makes an AI layer safe to point at inbound leads at all. Each brand sends from its own authenticated domain, so one brand’s reputation cannot damage another’s.

Value

• First response and follow-up automated across every brand, where both had been manual
• Routine pre-purchase conversations handled without a person in the loop
• Legal and pricing exceptions escalated by rule, never by model judgment
• Separate sending infrastructure and domain authentication per brand
• Delivered inside a $500 per month operating ceiling
• 5.0 client rating

“Evan was really easy to work with, and I would hire him again in the future!”

client · 5.0

• AI conversational layer
• CRM
• source-tracked forms
• multi-brand email infrastructure
• SPF / DKIM
• automated sequences
• human escalation logic

marketing agency 11-week engagement

Problem

A marketing agency could not trust its own funnel data. Lead sources were recorded inconsistently, campaign tracking was not standardized, forms and CRM were disconnected, and the Pipedrive data model did not match how the business actually ran. Leadership had no reliable recurring view of acquisition performance. The reporting problem was a data architecture problem.

Solution

Rebuilt the foundation underneath the reporting instead of rebuilding the reports. Standardized the lead-source taxonomy and campaign tracking, redesigned Pipedrive fields and mappings around the actual sales process, and connected lead-capture forms straight into the CRM.

Weekly reporting was built last, on a data model that could support it. Everything was documented on video so the client could run it without me, which was the point of the engagement rather than an afterthought at the end of it.

Value

• Attribution standardized across every acquisition channel, from inconsistent per-campaign tagging
• Lead-capture forms connected directly into the CRM
• Pipedrive data model rebuilt around the real business process
• Recurring weekly performance reporting established on clean data
• Delivered over 11 weeks, with the client operating it independently at handoff

• Pipedrive
• Make.com
• Zapier
• web forms
• Google Sheets
• UTM & campaign taxonomy
• Loom documentation