billhazelton.com
Co-founder, SelfScale · Los Angeles

Bill
Hazelton

I come from three generations of entrepreneurs. I've taken a couple of real jobs along the way, but I always drift back to my own thing.

I'm a builder, which mostly means I keep starting things before I'm qualified to.

My grandparents built a successful employment agency and a multi-million-dollar real estate portfolio in Southern California. My parents ran a small chain of Subway shops and sold it. My own record is more mixed. Some things worked: a search agency I started in 2004 and grew to 150+ clients, and a credit card affiliate business that did $1.2M. Some didn't, like AudienceMap, and that one's on me. Right now I'm all in on SelfScale.

See SelfScale LinkedIn X Contact
Under the hood

What SelfScale is made of

Most AI writing tools make everyone sound the same. SelfScale goes the other way. It builds a measurable model of how one specific person thinks, writes and talks, then scores every draft against it. If a post doesn't sound like you, it gets flagged before it ships. Here's what's underneath it, ten months after the first line of code.

10,909commits to main in the first nine and a half months
19,300regression tests across 1,670 files, wired to CI
817API routes
611database migrations
954defects logged in a registry, 786 of them fixed
87background and scheduled jobs
8social platforms with publishing integrations built
7required checks before anything merges

Not bad for a guy who isn't a developer. A lot of these were learned the hard way, and there are a few migrations in there I'd rather you didn't look at.

Build records

Things I built

The pattern, for better and worse: I spot a problem from the marketing side and start building before anyone can talk me out of it.

01

SelfScale

Oct 2025 – present Started it · co-founder

Social media taught everyone to perform, and generic AI made everyone sound the same. SelfScale learns how you write and talk, so you can post more without sounding like a press release.

I started researching the idea in October 2025 and wrote the first line of code that December. In April 2026, David Iya joined as my co-founder and we formed SelfScale, Inc. He's far better at design and go-to-market than I am, and I'm glad I don't have to fake those anymore.

  • A voice-profiling engine that learns from what you've already written and said, plus filters that flag when the output starts sounding like generic AI.
  • The boring guardrails early AI products tend to skip: every customer's data walled off from everyone else's, audit logs, and secrets scanning on every build.
  • Usage-based billing that doesn't charge you when a generation fails. Seemed only fair.
  • 19,300 tests and 7 required checks before anything merges, because I've broken enough things to know I need them.
Next.jsSupabase / PostgresVercelStripeElevenLabsHeyGenClaude Code
02

SelfScale MCP server

Early · still testing Built it

The idea: let any MCP-compatible AI assistant use SelfScale's YouTube-analysis and content tools directly, without opening the app. The scaffolding is built. Real-world testing is next, so treat this one as a work in progress.

  • 17 tools, running stateless on serverless, with token auth and the security protections the spec calls for.
  • Designed so every customer's requests stay isolated, with rate limits tied to their plan.
  • One config that works across Codex CLI and Antigravity, without ever exposing a secret.
MCPOAuth 2.1-style authRedisServerless
03

AudienceMap

Jul 2022 – Oct 2025 Founder Didn't survive as a business

A data-as-a-service agency. The package was identity resolution, built on a third-party platform, combined with cold email outreach to the contacts it turned up. I built the offer and the business around it, ran the sales calls, and handled the enterprise security reviews, which I don't recommend as a hobby.

Why it failed

I built the whole business on someone else's technology, and it didn't hold up. On top of that, the service was complex enough that every sale was long and hard to close, in a market where privacy rules varied by vertical and by state. Bottom line: selling data as a service in the U.S. was a headache, but most importantly, I never made it easy to buy.

Data-as-a-serviceAgency modelThird-party identity data
04

CreditCardAssist

2005 – 2013 Founder This one worked (twice)

An affiliate business in the credit card category that I built and ran entirely on my own. Not glamorous, but it earned $1.2M in affiliate revenue over six good years.

Then 2008 happened. Card issuers stopped approving applications and my income disappeared basically overnight. The business sat dormant for two years, came back fast when issuers started approving again in 2010, and I ran it alongside the agency until declining traffic and rising acquisition costs made me wind it down around 2013. I sold the domain in 2016.

05

Optimum Interactive → Spirecast → Everlast Ventures

2004 – 2021, with detours Founder This one worked too

I started an SEO and paid-search agency from scratch in 2004 as Optimum Interactive. When the 2008 crash wiped out my affiliate income, I took a job at Increase Visibility to keep the lights on, and got a lot better at SEO there. In 2011 I relaunched it as Spirecast, bringing some of the original clients with me, and in 2017 I founded Everlast Ventures. Across all three, we ran organic and paid search for 150+ clients.

  • At Everlast, built a demand-generation engine with a Salesforce-integrated attribution model that improved client lead volume 675%.
  • Clients saw a 94% average year-over-year revenue increase, and I ran our martech platform as a product with its own roadmap.
A bug I'm weirdly proud of finding

My CI was lying to me.

Three required checks were quietly skipping on large diffs and reporting success anyway. Green builds, nothing actually tested. That's how unreviewed code reaches production, and it's the kind of failure nobody notices until a customer does.

I found it, fixed it, and now I make every check fail on purpose before I trust it. Paranoid? Sure. Cheaper than an outage, though.

The day jobs

When I worked for other people

Mostly marketing: helping companies like Home Depot, Farmers Insurance and Atkins get found and get chosen, across $50M+ in ad spend. It's the part of building I'm actually trained for.

PatientsLikeMe

2021–2022
Sr. Director, Marketing & Community +74% qualified leads

And +82% pipeline opportunities, on a 20% smaller budget. Held a vote on roadmap priorities through the product relaunch.

Sober College

2016–2017
Chief Marketing Officer +217% revenue

Managed an $11.4M media budget, and built lifetime-value tracking by campaign that improved media efficiency 87%.

Earlier

2000–2004
Digital ad sales Google AdWords, early

Started in digital ad sales at About.com, 24/7 Real Media, iWon and Fastclick, back when "search marketing" needed explaining.

California State University, Northridge

B.S. Marketing
Cum laude, which I mention because there's no other place to mention it.
Elsewhere

Find me

LinkedIn has the full résumé. X and Facebook are there too, if you want the less buttoned-up version.

LinkedInin/billhazelton
X@billhazelton
Facebookbill.hazelton.9
SelfScaleselfscale.app
bill@selfscale.app