I spent over 10 years at CD Baby, the last seven and a half in product, through two acquisitions. I built the platforms, tools, pricing and policies that independent artists and 100+ partners counted on, and I led the company's first AI quality control system. I start with the teams closest to the customer, because they know where things get stuck and are usually too polite to say "I told you so." I would happily bring that approach to anything with customers, payments, risk and a lot of complicated rules (yes, I enjoy them).
Partner integrations: Spotify, Apple, Amazon, YouTube, Meta, TikTok, Peloton, Blizzard and 100+ more.
How I work
Years of interviewing and shadowing support, fraud, finance and outside customers shaped the platform relaunch.
Plain-language copy, warnings instead of blocks, and consent flows that say exactly what someone is agreeing to.
Superuser co-design and train-the-trainer sessions turn doubtful power users into champions.
Flagship case study
Turnaround, drawn to scale
I led CD Baby's first AI-assisted quality-control system for release review. Humans checked every release against dozens of hidden partner rules, artists waited up to two weeks, and review was our biggest source of support tickets.
Chapter 1
The queue held about 11,700 reviews with about 1,600 new ones a day. Artists wanted hours. Operations wanted strict checks. Some leaders wanted to cut review as a cost center, but dropping below partner scorecard thresholds put platform access at risk, so I argued for fixing the process instead.
Chapter 2
Hive.ai scanned each release's metadata and artwork against the same partner rules humans used and returned a pass or fail confidence score in 45 to 60 seconds. It did not replace full-listening judgment. Low-risk releases skipped review, high-risk ones went to people, and accounts with three or more clean releases auto-passed. Artists also saw warnings at upload so they could fix issues themselves.
Chapter 3
V1 was internal only, with inspectors using it and support beta-testing it. V2 showed artists only validated, high-accuracy errors. Before anything auto-approved, I agreed the trust metrics with Data Science: flag rate, false positives, rework and escalations. I set the threshold cautiously, because a bad release slipping through hurts partner report cards far more than an extra review.
Case study 2 · How I use AI day to day
Freelance film and photo crew kept saying a payroll platform "underpays overtime". I used AI for the research, the math, the spec and the code to find out why, and designed a fix that shows the rule when it matters. I've left the platform's name out.
Chapter 1
I searched public crew and producer threads for the platform. About 85 to 90 percent of results were job posts, which left eight user threads and six I read in full. Four themes came out: overtime that doesn't match the deal, support that's hard to reach, small productions unsure who would take them, and insurance declines for stunts. Overtime was the only one that appeared for both crew and producers over two years, named the platform directly, had a cause I could explain and could be fixed with something small. AI couldn't reach the site, so I ran the searches and pasted the threads back.
Chapter 2
I was new to film payroll, so I had AI build me a visual crash course: one worked example, who is involved, pay hours drawn as blocks, a live explorer, vocabulary and a quiz. The calculator shows the core idea. Crew divide a $1,100 day rate by 10 hours. Payroll divides it by 11 "pay hours", because California counts hours 9 and 10 as overtime. Every overtime hour is paid on the lower base, so four 12-hour days come up $120 short, the same figure a crew member reported.
Chapter 3
Checking the platform's own help pages changed my direction. A setting that pays what crew quote already exists. Producers pick the setting per worker at deal setup, crew never see which one was chosen, and admins see a number without the reason. The problem was visibility, not math. So I stopped designing a calculator and started designing a way to show the existing setting in the product.
Chapter 4
Pay isn't calculated any differently. The rule is simply shown where the mismatch begins, reusing patterns the product already has.
The warning leads with the impact, explains why in one line, then offers the fix. I rewrote it after reading it as a new user.
The overtime setting, the base-rate math and "your quote" appear in the deal summary crew already use.
An Issues strip on the weekly timecard shows the gap and the affected days, with a version for roles that can't see rates.
The standalone tool that reproduced the $120 gap from a real thread, kept as supporting evidence.
AI did the research, math, synthesis, spec and code. I chose the direction, gathered what it couldn't reach, and pushed back when its output drifted from reality.
This was a take-home exercise, so these are proposals and not results.
Selected work
Six projects with the results up front. Open any card for the full story, or dig into more below.
Owned a 0-to-1 service for high-value artists from concept to launch, including positioning, buyer messaging and go-to-market. About $375K gross revenue in H2 2024.
FastForward gives priority front-of-queue review, expedited delivery and priority support for $24.99 a release. It launched in October 2024.
I retired a publishing product that couldn't scale and launched its replacement: a $39.99 service bundling MLC, SoundExchange and sync royalties.
I oversaw Pro publishing's global expansion, then advised against scaling it for two reasons: the partner could not handle the volume, and the outdated program needed a major overhaul that would have cost more resources to modernize than to rebuild. When it wound down, existing Pro clients were fully grandfathered and received Boost at no extra cost. For new users, Boost offered a more flexible a la carte selection with more advanced royalty choices and payouts. Artists could opt in one songwriter or artist on a track and not another, and set a different royalty percentage for each opted-in person, track by track. The old product only allowed track-by-track selection, with no percentage per songwriter or artist.
The 42% adoption figure is measured on the cart upsell: of the users shown it, the share who chose to add Boost to their regular release submission.
CD Baby was the first distributor the MLC worked with by API. My workflow became the template for every partner after.
I made the gate-versus-open decision, refused imperfect-data and no-opt-out bars, and handled millions in unmatched royalties.
I owned large sections of CD Baby's platform relaunch, designing user flows and the rules for moving accounts and releases to the new platform.
It was an all-hands effort, and I owned large parts of it. I designed the user flows and built the data taxonomy, migration and logic rules for account migration, release migration, operations rules and partner rules (Spotify, Apple and others), based on years of interviewing and shadowing operations, support, fraud, finance and outside users. I was laid off before it launched.
An end-to-end rebuild replacing legacy Dynamics and DAX tooling used across support, finance, fraud and legal. Email throughput rose 23%.
Unified search, tags and holds, an audit trail and CSV bulk actions across millions of accounts. The phone queue was later retired.
A gamified progress tracker and intent-based onboarding lifted account creation. Moving checkout earlier with fewer fields lifted payments.
Album and single payment conversion rose 12% after moving checkout earlier in the release flow with fewer clicks and fields. Task completion on the progress tracker rose about 12%.
I settled a standoff between finance and support with a phased plan and a 90-day cutoff. Only about $9K was left unpayable.
Finance wanted 30 days and support feared ticket volume. We stopped new check setups first, then added a banner and direct messaging, with projected ticket numbers for support. ACH and PayPal options replaced checks.
I helped build the foundation of an internal program for high-earning and high-potential artists: the tagging, account identification, reporting, support alerts and account banners behind it.
Stages was an approval-curated list of artists with strong earnings or potential, who were pitched for extra promotion on platforms like Apple and Spotify and sometimes received white-glove service. I built the data connections, tagging, account identification and reporting for the team managing the list, the Backstage banners that flagged high earners, and the Zendesk cohort groups that alerted support when one of these artists wrote in. I also helped curate the cohort lists. The company later rebranded the program and promoted it publicly. It is separate from FastForward, which was a paid service anyone could buy.
I said no to a one-click opt-in and insisted on a mid-friction flow so artists understood what they were authorizing.
I worked on an early catalog launch into TikTok's commercial audio library and a later UGC scanning-rights transition. I also wrote the deck for engineering, legal and leadership and the plain-language artist copy.
Shadowing the team showed the real need was fast matching, not real-time alerts, so I scoped a batch check and cut the bigger build.
Fuzzy weighted matching handles lists of up to about 500K records against user and payment data. Results are emailed with flags highlighted and are not stored.
A public Slack escalation channel where senior reps answer juniors' questions, so customers get resolved without callbacks. CSAT rose 10 points.
A later backlog analysis found about 60% of a two-month ticket queue was password and account-recovery requests, which pointed the platform work at the real cause.
Classical releases took about six times longer to review than standard ones and produced under 1% of revenue, so I led removing the delivery.
Some classical inspections took 30 to 40 minutes each. I led the removal with approved customer copy, a Help Center article and support macros, and it shipped on October 21, 2025.
I owned the relationship with our payout partner, Trolley, and the process for royalties that could not be matched or paid.
I took over the Trolley relationship (platform changes, contracts, integrations, escalations and roadmap asks) after the product manager who owned it was laid off and the next one left. I also owned royalty matching for unmatched and unpayable royalties: setting guardrails and matching frameworks with data, quantifying where reasonable assumptions could be made, and making the calls with finance and legal on what to escheat.
I cleared a large inspection backlog by auto-approving accounts with a clean track record, so partner quality scores stayed safe.
I ran two flushes at different times, each using the same clean-account auto-pass logic as the AI quality-control system: accounts with a clean history skipped manual review, and risky releases still went to a person. The first took the queue from about 8,200 to about 4,600. The queue later climbed back, and the second flush started from close to 10,000 and cut it by about 40% again. An earlier April flush also cut about 40% and generated only about 14 Apple tickets, which showed the approach was safe before I repeated it. Separately, I cleared an aged support queue of about 17.8K tickets: a purge of resolved, duplicate and stale tickets, each sent a “reply if you still need help” message, removed about 90% of it, and the platform and automation fixes stopped it from refilling.
I worked on sign-in security and consent across the platform: SMS and multi-factor login in 221 countries, and the first cookie policy.
I built the cost model for sending SMS and WhatsApp codes across 221 countries and made the route and screen decisions that followed from it. I worked through international sender-ID registration so codes arrived from a recognized name. With legal, I helped put the company’s first cookie policy in place and handled data-subject requests under GDPR for artists in the EEA. I am a fast learner of regulated domains, not a compliance specialist, and these projects were product work alongside the people who were.
Security and support
Prototypes
Working screens help engineers, designers and stakeholders react to something real. V0 even generated full requirements for every page from my admin portal prototype.
A goth curio-cabinet site concept with fashion, creatures, crafts, journal, confessions and music sections. I used it to explore dark, luxe layouts and clear navigation.
An internal tool prototype for user search, bulk operations and a global catalog, the kind of screen support and operations teams use all day.
A prototype for support and operations teams: find any account, change many at once, act safely on one, and manage a global catalog. It uses mock data, identifiers are redacted, and V0 generated a full requirements document for every page from the finished screens.
Pick the change, upload a CSV of accounts, configure it. The apply button stays off until at least one account matches, and tags are grouped by severity.
Green for stages, amber for legal and monitoring, red for compliance and fraud, so risk is visible at a glance.
Tag banner, identifiers with one-click copy, balances, payouts and a menu of ten guarded actions.
Combining accounts is a focused, explicit step that can be cancelled with no side effects.
Required sections are marked, edits happen only after unlocking, and a section menu keeps a very long form navigable.
Wildcards, combined terms, sortable results, status badges and a result cap that asks you to refine.
A prototype for the customer side of a music platform: how someone adds a songwriter, producer or performer to their account. One decision per screen, a branch on who the person already is, and the rights details tucked into optional sections so a first-timer is never blocked. The screens use test data.
Performance rights organization, IPI and ISNI sit in a collapsible section beside social links and bio. Everything is optional, and the screen says it can be updated later, so a missing number never stops the flow.
Each answer is explained in plain words. People already on streaming services connect their profiles, and behind-the-scenes contributors skip straight ahead.
The confirmation names the new contributor and offers the two likeliest next moves, so people keep going instead of landing on a dead end.
Prototype screens for the part of a release where money gets divided: songwriter shares that have to total 100%, payment splits on an artist profile, and a confirm or decline step for splits on releases someone else owns. All data is mock and the names are placeholders.
Each songwriter gets a contribution share, a contribution type and a publisher, with shares split evenly by default and a running total that has to reach 100%.
Each release shows who is paid what as percentage bars, whether the split is pending or confirmed, a nudge for slow approvers and a way to request a change.
When someone else owns the release, you can confirm or decline the split you are assigned, but you cannot edit it.
A Regency dynasty sim, in progress and not yet released. I am using it to explore game-systems design: how traits pass between generations, what a family can lose through scandal, and how marriage and inheritance shape a long game. Built in V0 with generated families and portraits.
Reputation, fortune and stats for charm, wit, accomplishment and propriety, with skills, appearance and wardrobe tabs.
Each family has a rank, a seat and an income, and their eligible members are who your children can court.
Every family is generated, with portraits that inherit features from both parents.
Status: early prototype. Courtship and events are still being built.
The form reacts to the song type. Choosing Cover swaps the songwriter fields for a mechanical license requirement, with a way to get one or confirm you already have it.
One artist can have several aliases, each with its own streaming platform IDs, so revenue and analytics land on the right person.
A three-step sign-up: email, verification, then password, designed to keep each screen to one job.
Reorder tracks, lock the 15-second preview clip that streaming platforms require, and offer mastering right where the artist is already choosing options.
A reskin of a B2B data consultancy site: a light paper look, forest-green and lime palette, serif headlines, audience paths and a mobile menu. The client is anonymized here.
Player and creator
I have shared 26 community Workshop items across two management games, from habitats and scenery to original characters. Players subscribe to them and put them in their own worlds, so I get direct feedback on what people actually want to use.
Themed builds designed to work in any biome. Subscribers per item:
Named colonists with their own looks and stories, including families and twins. Subscribers per item:
Counts come from Steam's public Workshop pages and change over time; this snapshot is from October 2026. Subscriptions are not unique people.
Path
“I've sat on the front line at Netflix support, managed the front line at iSoldIt, and designed for the front line at CD Baby.”
Aug 2018 – Jan 2026
Platform, partners, rights and royalties, internal tools, onboarding and checkout.
Aug 2015 – Aug 2018
Frontline operations, QA, process work and the Brain escalation channel.
Apr 2010 – Jun 2015
About 100 calls a day at the Real Time Expert level, plus coaching reps on other teams.
Aug 2006 – Nov 2009
Ran an eBay storefront with a team of 5 to 10.
Toolkit
I write requirements and prototype in Figma and V0, then work closely with engineering and data teams who ship the production code. My SQL is intermediate.
Credentials
Twelve badges and certificates from three issuers, all earned in 2026. Every one links to its public verification page.
Product analytics, experimentation and data hygiene.
Product judgment and building with AI.
Working with Claude and Claude Code.
Consulting
Open to creator and entertainment products, and to any platform with customers, payments, risk or internal tools. I bring 20+ years of working directly with customers, from theme park registers to Netflix's contact center to product at CD Baby.
Shadow customers and frontline teams, turn messy asks into scoped requirements, and prototype before engineering starts.
Rebuilds that cut handle time: search, audit trails, bulk actions and the rules behind them.
Confidence-based routing, trust metrics agreed with data teams, and phased rollouts.
Licensing and royalty flows, payout changes without support spikes, and opt-in design that keeps creators in control.
0-to-1 services, upsell placement, onboarding and checkout.
Get in touch
I'm looking for my next senior product role on a platform for creators, makers or entertainment audiences, and I take consulting and contract work. Tell me the scope and timeline and I'll reply with next steps.
Need a break?