AI automation systems for service businesses
AI agents, workflow automations, CRM systems, and smart websites — built so your team stops doing robot work and leads stop slipping.
What I build
Six ways to take manual work off your team's plate — each scoped around a business outcome, not a tool.
Find the highest-leverage automation before you spend a dollar on tools.
Your tools, connected into end-to-end pipelines that don’t drop things.
Every lead gets a fast, correct first touch — automatically.
Agents that survive production: evals, guardrails, human handoff.
A website wired into your operations — it produces handled leads, not traffic.
Your systems get an owner: monitoring, fixes, and steady improvement.
Tool-specific work: n8n · Make.com · Zapier · GoHighLevel — or see all services.
What changes when the systems are fixed
The same lead, two very different journeys.
Selected case studies
Real systems in production, with the numbers they moved.
A tier-1 triage system that classifies, drafts, and routes support tickets — so humans approve instead of typing.
An OCR → extract → reconcile pipeline with a full audit trail, replacing a four-person manual document operation.
A repo-aware coding agent that plans, edits, and opens pull requests with passing tests — no hand-holding chat sessions.
A booking flow that converses — negotiating dates, answering questions, and upselling — instead of abandoning visitors to a calendar grid.
How I work
Four phases. Each ends with something you can use — even if you stop there.
- 01
Discover
→ A prioritized automation roadmapAudit the current workflow. Map data, decisions, and edge cases. Identify the highest-leverage automation surface — scored by effort and payoff, so we fix the most expensive problem first.
- 02
Architect
→ A build plan with budgetsPick models, tools, and infrastructure. Draft eval criteria for anything AI-driven. Define cost and latency budgets up front, so there are no invoice surprises later.
- 03
Build
→ A working system on real dataShip a thin slice end-to-end first, then iterate with real data. You see progress weekly, not at a big reveal. Claude Code does most of the typing; judgment stays human.
- 04
Harden
→ A documented, monitored systemEval suite, observability, error handling, and fallbacks. Documentation your team can operate from, training for the operator, and an optional retainer so the system keeps an owner.
Who this is for
- Service businesses, agencies, and consultancies with real lead flow
- Founders and lean teams who want systems, not more headcount
- Teams ready to change a process, not just buy a tool
- Businesses that want AI applied practically, with guardrails
- Teams looking for prompts or advice without implementation
- Projects with no internal owner or decision-maker
- Businesses that want AI for the press release, not the operations
What clients say
Erick rebuilt our doc pipeline in 6 weeks. We retired a 4-person manual process. Still working perfectly 9 months later.
He treats AI systems like real software — evals, observability, runbooks. Unusually rigorous.
The kind of contractor you keep on retainer. Ships fast and the systems hold up.
Leads used to sit in an inbox overnight. Now the right tech gets them in under a minute, with follow-up running on its own. Booked jobs went up the first month.
Our onboarding paperwork ran through four tools and two people. Erick collapsed it into one n8n pipeline with alerts. Nobody re-types anything anymore.
We white-label his GoHighLevel builds for our clients. Snapshots, documentation, naming discipline — it’s the cleanest GHL work I’ve seen.
Our managers ask the copilot questions they used to queue behind the data team for. Five-minute waits became twenty seconds, and the data team finally does real work.
The CRM rebuild paid for itself before it was finished. Follow-up stopped depending on anyone’s memory, and my forecast finally matches reality.
Reps approve researched, personalized outbound instead of writing it. Reply rates jumped and nobody wants the old way back.
Booking, reminders, no-show recovery — all automated, all working. Front desk calls dropped by a third.
He untangled two years of Make scenarios nobody understood, added error handling, and cut our task bill in half. Then he documented everything.
The audit told us what NOT to build, which saved us more than the fee. The two systems we did build run every day.
Clear scope, weekly demos, budgets set up front. It’s the least stressful technical project we’ve ever run.
Tools of the trade
The stack on the left — what it looks like in use on the right.
Full stack notes on /tools · versatile across LLMs, default to Claude.
// live replay: real build sessions — Claude Code, n8n, evals. Open the full playground →
Notes from real builds
Implementation thinking, not tool roundups.
Most workflow problems don’t need AI. Some don’t even need automation. Here’s the three-question filter I run before recommending anything.
The words get used interchangeably in sales calls, and the confusion causes real project failures. The distinction is simple: automations follow paths, agents choose them.
Nobody budgets for the deals that die in a stale pipeline. The cost is real, it compounds, and the fix has a correct order of operations.
Find the highest-leverage fix first.
Book a call and I’ll help you identify the workflow, automation, or systems bottleneck worth fixing first — before you spend on tools or builds.
30 minutes · no pitch deck · reply within 24h if you write instead