Operations at Scale
The specifics of this work are confidential, and keeping them that way is part of the job.
An essay on the craft of running very large operations, told at the altitude discretion allows. The specifics stay inside. The thinking is mine to share.
At a glance
- What
- A craft essay on running Trust & Safety operations at scale
- Scope
- One quality standard · 10 global sites · 800+ moderators
- Receipts
- 45% fewer processing errors · 20% faster reporting
- Note
- Specifics confidential — by design
0
Global sites, one standard
0+
Moderators in scope
0%
Fewer processing errors (Legal Ops)
0%
Reporting efficiency gained
Why Operations
I fell into operations. I stayed on purpose.
Nobody grows up dreaming about quality frameworks. I landed in operations sideways — and then discovered it's where my favorite kind of problem lives: systems design with people inside it. Code does what you tell it. An operation is hundreds of humans across ten sites, a policy that keeps evolving, and a queue that never sleeps — and it still has to produce one consistent, defensible answer, every time.
Making that happen quietly is a craft. When it's done well, nobody notices, which is exactly the point — the best compliment an operation can get is silence. This page is about that craft: the parts I can show, from the work I mostly can't.
The Craft · 01
The tools people actually use are designed around the people.
Most operational tooling is designed around the process and inflicted on the humans. I work the other way: watch where people actually stumble, then design the tool around the stumble. It's design thinking applied somewhere unglamorous — not a product launch, a workflow — and that's precisely where it pays off, because a small amount of friction multiplied by hundreds of people and thousands of repetitions is not small.
The clearest proof I can share: I rebuilt an error-management workflow for legal operations, and internal processing errors fell 45%. The insight wasn't a clever algorithm. It was sitting with how errors actually happened and refusing to blame the people for a process that made errors easy.
The Craft · 02
One standard is easy. One standard across ten sites is the job.
Here's the uncomfortable truth of quality work: a standard is not what's written in the document. It's what hundreds of different people, in different countries, with different first languages, actually do with the document at 3 a.m. their time. Those two things drift apart the moment the ink dries — and the drift compounds. Calibration is the discipline of pulling them back together, continuously, without ever being in the room.
It sounds abstract until you feel it. So feel it — takes about thirty seconds:
Rating Rubric v1.0
Rate each swatch: how blue is it? (1 = not blue, 5 = fully blue)
- Blue means blue-blue: sky, cobalt, navy.
- If it leans green, it caps at 3.
- If it leans purple, subtract at least 1.
Written with clarity as the intent. So is every policy.
Swatch A
Swatch B
Swatch C
Swatch D
The Craft · 03
You get exactly what you measure. Choose carefully.
Metrics are steering wheels, not report cards. Measure speed alone and quality quietly erodes; measure quality alone and the queue grows; measure both without context and people optimize the number instead of the outcome. Designing a quality metric is designing behavior — which means the metric designer holds more influence over the operation than almost anyone in it, and had better take that seriously.
I build the instruments too: custom SQL dashboards that surface where performance actually bottlenecks, built to answer questions instead of decorating slides. One rebuild improved reporting efficiency by 20% — which, in an operation, means decisions land a day earlier, every day, forever.
The Craft · 04
The system can't stop while you change it.
Operational change has a constraint most change frameworks politely ignore: the queue doesn't pause for your rollout. Every improvement ships into a system that is already running at full speed, staffed by people who have a job to do today. So change becomes a craft of sequencing — pilot small, calibrate the pilots, absorb the lessons, then scale the version that survived contact with reality. Rolled out across ten sites, a change isn't one change; it's ten local changes wearing one name, and each site's context has to be respected for the standard to hold.
The newest chapter of that work is moving safety upstream — integrating generative AI into moderation pipelines so problems get caught earlier in the product lifecycle instead of cleaned up after the fact. Same craft, new instrument: the hard part still isn't the technology. It's changing a running system without dropping what it carries.
Reflection
The best operations work is invisible. That's the point — and the price.
Builders get to show the thing. Operators get to show the absence of disasters — which looks, from the outside, like nothing. I've made peace with that trade. Quiet is what a healthy operation sounds like, and building toward quiet — fewer surprises, fewer heroics, fewer 3 a.m. escalations — turns out to be some of the most demanding design work there is.
Discretion isn't a limitation on this portfolio. It's a qualification in it.
Instruments
- Quality Frameworks
- Calibration & Consistency
- Metric Design
- SQL & Dashboards
- Process Design
- GenAI in Workflows
Judgment
- Change at Scale
- Influence Without Authority
- Design Thinking in Ops
- Cross-Cultural Leadership
- Discretion
- Systems Thinking
Want to talk shop?
I can't tell you the specifics. I can absolutely talk craft — quality systems, calibration, and how to change an operation without dropping what it carries.
Operations at Scale · Trust & Safety · YouTube (Google)