Customer support and service, and the AI I build on top of it
Twenty years of being hired for one job and growing it into a bigger one. Six times, twice on teams the company itself called a special project.
What the work produced
96%
of tickets touched by the support AI I built, across 38,249 tickets in ninety days
92%
CSAT through 2024, all four quarters above 90, straight through a full automation rollout
48hrs
translation turnaround, down from three to five days, on the platform I shipped in August
Every figure here is computed from production data and recorded against its source. Ask the agent below to push on any of them.
You are not hiring a rehearsed pitch. You are hiring a person: driven, honest, harder on himself than any boss. I hear a problem and my first thought is how do I fix this now.
Every job here started as one thing and turned into another, because I kept doing the work next to the work. Car floor to internet manager. Call queue to special projects. Support lead to building the AI. Same move, twenty years running, and it is the honest answer to whether I can do a job I have not done yet.
Showed an aptitude for the phone and online side nobody had staffed, and got moved to it. Ran lead generation for the whole floor and led a team of 10. Online went from an afterthought to the strongest channel, in a market that was shrinking.
Got pulled onto the Verizon Offer Team as a special project, then worked with corporate to take it national. Also negotiated a remote work agreement in 2012, years before that was a normal thing to ask for.
Earned a B.S. in Game Design from Full Sail University while working those jobs full time. Not a computer science degree. A degree about how interactive systems get planned, built and shipped, taken at night by someone already working sales floors and call queues. The instinct to build did not start at COROS. It started here.
Went up to Tier 2 across Mac, iOS and Accessibility, supported lower-level advisors, and took rotational roles specifically to learn how the company worked outside my own job.
Make sure customer commitments were met on time.
Started building the reporting. Weekly, monthly, quarterly and yearly analysis of impact and trends, written for advisors, managers and business leaders so they could actually plan against it. Ran the process improvements that came out of what the data showed.
The real bridge. This is where I stopped just working the operation and started instrumenting it.
Run customer support.
Pursued a BPO to scale coverage. It was shut down on cost and setup, so I pivoted and built a direct hiring model in the Philippines instead, defining the hiring standards it still recruits against. Before that COROS hired only in the US, EU/UK and China. Grew the org to 32 people across four regions and 8+ languages, held 92% CSAT through 2024 across a full automation rollout, and taught myself to build the tools the team did not have.
A role that did not exist before: AI and Automation. I hired and trained my own replacement to get there.
Build automation for the support organization.
Shipped CARA, RELAY, SENTINEL and a mail-routing classifier, co-authored the enterprise AI playbook, and run the production infrastructure myself. Several shipped before they had funding.
The thing I am actually selling: someone who runs the operation and builds the system that scales it.
I like to be the lever,
not the wheel.There is nothing wrong with being in the wheel. Most of the work happens there. But sometimes you have to get the thing moving yourself, and that is doubly true when you are running a team. People need to know you will go to work just as hard as you are asking them to. You cannot ask for that with a speech. You earn it where they can see you.
A decade in customer support and service taught me the thing no tool teaches. Operations are people first. Get that right, and the technology finally has something worth amplifying.
Grew it to 32 people across the US, EU/UK, China, and the Philippines. The Philippines was mine: I pursued a BPO, it was shut down on cost and setup, and I pivoted to a direct hiring model there instead, writing the hiring bar it still hires against. Then I handed the whole org to the successor I hired and trained, and it has kept growing since. Hiring, structure, coverage, and culture across four time zones and eight-plus languages.
I ran CX as the thing the operation is built around, not a queue to clear. 92% CSAT through 2024, all four quarters above 90, straight through a full automation rollout, because service quality was the constraint I protected.
I get fired up when someone on the team levels up because of something I built. I lead by building trust, not carrying a big stick. The measure is whether the people around me grow.
When the highest-leverage work moved to building, I hired and trained a successor for the people-management side and handed it off cleanly. The team was stronger for it.
Named the way they actually come up, not by what I called the software. My title sits under Support, but the systems do not: the mail classifier routes for 22 audiences companywide, the translation pipeline serves marketing, and the feedback platform pushes consumer insight to product. Open one to see what I did about it.
Every support org hits this. The usual answers are hire more people, or buy a bot that annoys customers into giving up.
Built an autonomous support AI on the Claude API, wired into the ticketing system with a domain knowledge layer, multi-step reasoning and quality scoring on every response. Shipped it before it had headcount approval.
It touches 96% of tickets in the support queues it serves, 38,249 of them over ninety days, and closes a meaningful share without a human. The number that matters is the quality one: 84.5% CSAT on tickets CARA resolves, against 86.9% human-handled. Statistically even. The ceiling is data-privacy policy, not capability.
Successor in-house architecture beats the incumbent vendor on 82.5% of 1,074 replayed real tickets at about three cents each, built to retire a $43,200 a year contract.
Support data is locked behind query syntax. If you cannot write the filter, you file a request and wait, so most questions never get asked.
Built Support Intel, a plain-language layer over ticket and service data. Ask it a question the way you would ask a colleague and it answers from the real records, showing the SQL it ran so the answer can be checked rather than trusted.
The point is not the model. It is that a team lead can check a hunch in thirty seconds instead of opening a ticket with someone like me and waiting two days. Temperature is pinned at zero and common questions run hardcoded SQL, because a number that changes between two identical questions is worse than no number.
Live at its own domain. By my own assessment it needs another pass: natural-language querying is only as trustworthy as its worst answer, and that is the part I would tighten next.
Order history in one system, device and warranty in another, the conversation in a third. Every answer starts with four tabs.
Built a customer intelligence panel that pulls orders, devices, warranty and history into the agent's existing workspace. Then reflowed it from nine look-alike accordions grouped by data source into the sequence the job is actually done in.
Handle time is the metric every support org already tracks, and this attacks it where it happens: at the agent's desk, in the tool they already have open.
Live. Also caught and fixed a warranty rule that had been quietly telling customers their device was expired during its final month of coverage.
Onboarding was tribal knowledge and shadowing. Ramp time was whatever the nearest senior agent had time for.
Built a training platform with structured courses, quizzes and progress tracking, so ramp is a process instead of a favor.
Ramp time is a support leader's problem, not an L&D problem. If it takes ninety days to make an agent productive, every hire is a ninety day bet.
Live, with the product curriculum rebuilt after an audit found courses teaching specs that had shipped wrong.
The support stack was three vendors deep and the renewal was coming. The honest question was whether any of it needed to be bought at all.
Researched the incumbents properly, then built a complete replacement solo: 52 route modules across roughly 516 endpoints, 20 background workers, 80 database migrations, the full ticket lifecycle, and a simulator for the AI layer.
Then I did not deploy it. It works and it is validated, and the business case for cutting over did not clear the bar at the time. It is held on purpose, pending further evaluation of development and budget.
This is the one I would most want to be asked about. Building it proved I understand the domain. Not shipping it is the part that took judgment.
The work nobody assigns. Fraud monitoring, mail routing, inventory forecasting, competitor tracking. None of it was on a roadmap.
These run outside the support org. SENTINEL watches Certificate Transparency logs and DNS permutations to catch lookalike domains within hours of registration. A Claude classifier routes one inbound mailbox serving 22 audiences across 8+ languages, and learns from human corrections.
The mail router is the clearest example of the habit: nobody assigned it, a pipeline was mis-routing daily, so I replaced it. A Claude classifier now triages one inbound mailbox serving 22 distinct audiences across 8+ languages, and every human correction feeds back as training. The interesting engineering is the failure handling, because a classifier that quietly starts mis-routing looks exactly like a quiet inbox, so corrections are logged and reviewed rather than assumed.
All running. None of them were assigned.
Every piece of content that needed another language waited on an outside specialist. Three to five days depending on the material, and a bill attached to each one. Marketing planned around the delay rather than the work.
Built the translation platform from scratch, including a terminology lexicon so product names and technical terms stay consistent, and per-market register so the output reads native rather than translated. Several flows were tested before one earned its place.
The model was never the hard part. Getting the process right was: who submits, what the tool guarantees, and where a human still reviews. That is why the version that shipped is the one everyone involved actually wanted to use.
Live and in daily use. Most content now turns around in 24 to 48 hours instead of three to five days, and it has cut billable hours with the outside specialist. This one sits with marketing, not support.
Prefer to interrogate this rather than read it? There is an AI agent below trained on how these were built. It has been told not to quote figures that were never sourced, so if you push it on business value it will say so.
Ask it anythingKnowing systems is one thing. Knowing systems and people, so they actually work together, is the whole job. Here is where it shows up.
CARA takes the repetitive share so the team spends its time on the hard, human half: judgment, relationships, the edge cases. The capacity it creates is not headcount removed, it is headcount freed to do work that actually needs a person. We held 84.5% CSAT on AI-resolved vs 86.9% human-handled straight through the rollout.
A tool nobody adopts is a hobby. So the systems ship with the other half. I co-authored an AI Playbook with leadership and ran hands-on training. The build and the adoption both get my time, on purpose. That is why the systems actually stick.
When building became the higher-leverage work, I did not just walk. I hired and trained a successor for the people side, handed off cleanly, and moved into a full-time AI and Automation role. The measure of the work is whether the people around me are better for it, not whether I am indispensable.
Iron Man works because Tony Stark is inside it. AI amplifies what good people already do, and exposes what weak process was hiding. Without the human judgment inside, the technology is just code.
I can excel and lead. I can also burn out and have a bad day. Those are human things. I do my best work where people get that, and where I can be the one they count on to get it done.
It is one thing to know a system. It is another to know systems and people, and get them working together. That overlap is where I live.
I have never been the guy who interviews easy or shows off well. What you get instead is honest, driven, and steady. Find the real problem, build the thing that solves it, bring the people with you, measure everything, move on.
I want roles where operational leadership and hands-on AI building are not two different people. Customer experience, support operations, or automation with real stakes. Also open to select consulting through Infinitus Digital.