Marketing Operations & AI Automation

One person running what normally takes four hires.

At most agencies the phone system, the call QA, the content publishing and the creative are four separate people. I run all of it for a US marketing agency, because the repetitive part is handled by systems I built.

Seven years in B2B sales and revenue operations. I started on the phones, which is why the automation gets built around how buyers actually behave on a call rather than how a flowchart says they should.

Portrait of Christian Perales

Christian PeralesRemote, US Central hours · English and Spanish
Available for contract and fractional work

What I actually do

Phone systems and AI voice agents

Inbound and outbound agents that answer, qualify, handle objections and book straight into the calendar. Transfer rules, call scripts, the lot.

CRM and automation

GoHighLevel end to end, Make.com, webhook and API integrations, lead routing, data hygiene. The plumbing that makes everything else work.

Call auditing and reporting

Every call transcribed and scored against a rubric, turned into coaching reports a client can act on. Runs on a schedule, not by hand.

Content, ads and creative

Publishing that posts itself on schedule, plus short-form video ads and landing pages produced in house. Paid social and search, SEO, Local Services Ads.

Web and mobile app development

Concept to shipped product. A dating app for the Spanish-speaking market and a booking platform for a driving academy are both live and clickable further down, alongside five more sites and products. Production software, not tool configuration.

Post-call automation flow diagram
Phone systems
The outbound dialer on the Make canvas
CRM & automation
A page from an automated call coaching report
Call auditing
Still from The Graveyard, a 67-second concept film
Ads & creative
Verbana dating app landing page
Apps & sites

Every one of those is further down this page, full size and clickable. Nothing here is a stock photo.

Hire me for one of these or for all of it. Most people start with the one that's bleeding.

It's one machine, not four skills

Every piece feeds the next. That's the whole reason one person can hold it.

01 · Capture

AI voice agent answers, qualifies, books straight into the calendar.

02 · Route

CRM tags, routes and rescues anything that slipped through.

03 · Score

Every call transcribed and scored, coaching reports generated.

04 · Feed

Content engine keeps the top of the funnel full on schedule.

05 · Watch

Monitoring catches a failure before it costs a lead.

Built on GoHighLevel, VAPI, Make.com and custom pipelines, wired end to end rather than bolted together.

Talk to an AI phone agent I built

Click the button and it answers, in your browser, out loud. It'll ask what you do, work out whether there's a fit, handle whatever you throw at it, and try to book you a call. That's the same job the agents I build do on live customer calls all day. This one is pointed at me instead of a client, so it's the work sample rather than a description of one.

Runs in your browser, needs a microphone. Ends on its own after five minutes.

Worth trying
  • Ask what he actually does
  • Ask if he could run your phones
  • Ask what he charges
  • Say you're not hiring right now
  • Book a time

Call ends in 5:00

Transcribing what it hears you say

Your side of the call appears here, live, as you speak.

4hires this replaces
20+partner accounts audited
48modules in the dialer
7products shipped, all live
$200Kmonthly revenue reached, up from $10K

What each piece replaced

The short version, before the detail. Six things a growing agency would otherwise hire for, and what does each one now.

Inbound rep & after-hours answeringAI phone agents on live customer calls
The follow-up nobody has time forA callback loop that keeps the promise
A QA analystCall auditing across 20+ partner accounts
A social media managerPublishing that runs on schedule by itself
A video editorAround thirty short-form ads, made in house
Nothing — most teams just don't have itMonitoring, so automations fail loudly

AI phone agents on live customer calls

Problem

Inbound calls hit voicemail after hours and outbound follow-up depended on whoever had time. Leads died in the gap.

Built

Voice agents covering inbound and outbound, with qualification logic, rules for handing off to a human, and calendar booking wired directly into the CRM. Prompts and scripts tuned against real call transcripts, not guesses.

Outcome

Calls get answered and qualified around the clock, and booked appointments land in the CRM without anyone retyping them.

Diagram of the post-call automation: the call ends, the system reads what happened, creates or updates the contact, saves the call notes, then sorts into ten outcomes including booked, not a fit, voicemail and stopped at reception.
A call ends and nobody has to write anything down. The system reads how it went, files it against the right contact, then sends it down one of ten paths: booked, not a fit, not ready, do-not-call, voicemail, stopped at reception. Where a callback was promised, it schedules that too. 31 steps, and the alternative is a rep remembering every one of them after every call.
Diagram of the outbound dialer's opening sequence: the run kicks off, pulls today's appointments, goes through them one by one, notes who is already booked, marks who still needs a call, checks for promised callbacks, then decides whether to call or skip.
This is what decides who gets called today. It pulls the day's appointments, works out who is already booked, checks who was promised a callback, and only then dials the people who still need it. 48 steps, none of which depend on somebody remembering to check the calendar first.

Plain English on top, the real module name underneath. Both diagrams are drawn straight from the live scenarios, so those names are the actual ones in the build.

The whole outbound dialer: about forty-five modules ending in a router that fans out to separate AI call agents.
The whole dialer on the Make canvas — 48 modules, ending in a router that hands each contact to the right agent for that call type.
Post-call automation: a router fanning out to a separate branch for each call disposition, each one writing back to the CRM.
The other half. A call ends, the outcome is classified and written back: booked, not interested, not qualified, follow-up, do-not-call, voicemail, gatekeeper. This is the part that decides whether a call turns into anything.

Click either to open it full size.

Callbacks that actually happen when they were promised

Problem

"Can you try me after four?" is where most outbound dies. Someone writes it on a sticky note, or nobody does, and the lead quietly goes cold. It's the single most common way a warm prospect is lost.

Built

The agent hears the time the prospect asked for, that commitment is captured off the live call, and the callback is scheduled for exactly then. If they don't pick up, it doesn't stop. It enters a retry ladder with its own rules per outcome, and every attempt writes back to the CRM so nobody is called twice or forgotten once.

Outcome

A prospect who says "call me at four" gets a call at four, without anyone remembering to make it happen.

The callback loop automation: a scheduling chain feeding a large fan-out of retry branches.
The callback loop. 45 modules of scheduling and retry logic, because "try me later" has a dozen different endings and each one needs its own handling.

Automated call auditing across 20+ partner accounts

Problem

The agency needed to show partner businesses how their staff handled inbound leads. By hand that meant listening to hundreds of calls per account, so in practice it never got done.

Built

A pipeline that pulls recordings, transcribes, scores each call against a defined rubric, and produces client-ready coaching reports as finished PDFs. Every score is verified against the actual transcript, because a model will hand you a confident wrong answer that reads perfectly well.

Outcome

Runs on a recurring schedule across 20+ partner accounts. Days of listening became a pipeline run plus a human review pass.

Page one of a monthly call handling report: scored call-handling skills, a summary takeaway, and per-skill averages.
A page the pipeline produces. Client details anonymized — the layout, scoring and copy are the real output. Open the PDF.

Content publishing that runs itself

Problem

Daily posting is the first thing that slips when a team gets busy, and it slips quietly.

Built

A calendar built weeks ahead with slots held open for real-time content, feeding a scheduler that publishes to Instagram, Facebook and Google Business Profile automatically.

Outcome

Publishing continues on schedule whether or not anyone is at a desk, and a failed post raises an alert instead of going silent.

Ad creative produced in house

Problem

Outsourced creative is slow and expensive, which means you test far fewer angles than you should.

Built

A production pipeline for short-form video ads and landing pages, including AI-generated cast and scenes where live shooting isn't practical. Roughly thirty finished pieces to date.

Outcome

New angles ship in days instead of weeks, so testing is driven by what the numbers say rather than by the production budget.

Monitoring, so automations fail loudly

Problem

Automated systems break quietly. You find out from the damage rather than from an alert, and by then it's been days.

Built

Health checks and alerting across every live system, including off-box heartbeats, so a server going down is itself detected instead of producing a silence that looks like everything is fine.

Outcome

Failures surface immediately instead of after they've cost leads.

Before the systems

Scaling a remote B2B sales operation

Role

Ran the B2B sales and CRM operation at an early-stage B2B service business while monthly revenue grew from roughly $10K to over $200K. Owned the CRM build, lead routing, automation sequences, scripts and rep coaching.

What I'd do differently

I rebuilt the scripts from call recordings far too late. We ran close to a year on assumptions about why people said no, and the recordings told a different story than the pipeline did. That's why call review is now wired in from day one, and ultimately why I built the auditing system above.

Ad creative, produced in house

Written, cast, cut and shipped by me. No agency, no editor, no shoot day. Where you see people, the cast is AI generated, which is how a new angle gets tested in days instead of weeks.

The Graveyard · 67s concept film
A man walks a cemetery at night reading headstones. Each one is a dead company, with the cause of death on the stone. Rapid Response Restoration. 1.8 stars, four reviews. Est. 2015, closed 2022. "Nobody could find us online." He writes each one down.

Entirely generated: the location, the character, the fog, the crow. Same man and same graveyard across the whole run, which is the part that usually falls apart. The pitch is never spoken. You work out what's being sold from what killed everybody.
Size Matters · 43s spot, credit repair
Two women at a bar, mid-conversation, and you think you know exactly what they're talking about. Then the card lands: Size matters. Credit matters more. Written for a friend's business.
The whole ad is one bait and one turn. It works because it borrows an assumption the viewer brings with them, spends forty seconds letting them sit in it, and spends the payoff somewhere else entirely.
Same Name, Same Script, Sold Four Times · 33s
Cut from real podcast footage rather than generated. The format does the work: two people at microphones reads as a conversation you're overhearing, not an ad you're being served, and that buys enough patience to land an actual argument.

The argument being that the leads these owners pay for are resold to their competitors. It names the resellers, shows the price difference, and closes on "stop buying competitors' leftovers."
You Become Who You Surround Yourself With
Opens cold on a black card, then hands the rest to real customers.
The Marketing Graveyard
The same idea as the film above, cut as motion graphics. One concept, two executions.
Keyword Theft
Explainer. Turns an invisible problem into something you can watch happen.
Lead Relay
AI-generated cast and scenes, cut as a straight direct-response spot.

How I work

On AI: I use it heavily and I check its work. Nothing a model produced ships without a pass against the source. That review step is the job, not an afterthought to it.

Sites and products, shipped and live

Designed, built and deployed. Every link below is live right now — click any of them.

Seven live products across dating, driver training, med spa, credit, construction and marketing. I ship production software, not just tool configuration, which is usually the difference when the thing you need doesn't exist yet.

Get in touch

Available for contract and fractional work in marketing operations, automation and conversational AI.

perales_c@ymail.com

Client work is kept general here by design. Happy to walk through any of it in detail on a call.