Hired to run analytics for a regulator. Left behind a product practice — the first one the agency had.

The agency didn't need more data. It needed a discipline for knowing what the data was for. That took about ninety days to see clearly, and another five years to build. I wrote a proposal — a team, a way of working, a place to put it — and a few commissioners took the bet.

Building the team

The analysts already there were the first hires, in a sense. They knew the agency — its history, its quirks, its people — and that kind of knowledge is most of the job in a regulator. Throwing it out to bring in an outside team would have cost a year I didn't have. Familiar faces using new methods opened doors a transplant couldn't.

The outside hires came in a deliberate order. A data scientist, an engineer, a senior analyst — people who could ship something visible inside of a quarter. Then, once we'd banked enough credibility to spend, a deputy director who'd run data as a product, a designer, a researcher. By then we weren't an analytics team anymore.

Choosing what to work on

Prioritization at TLC worked the way it does most places that are operationally stretched. A commissioner heard a complaint at a community board. A project got launched. Whether anyone could later say what number it moved was a question that rarely got asked, and when it did it tended to be retrospective.

We changed the posture: nothing got scoped without a forecast attached. If we do this and it works, what changes, and by how much. Wait times. Processing days. Resolution rates. The trick wasn't asking the question — it was answering it ourselves, in advance, and bringing the answer to the conversation. Asking stakeholders to justify their requests would have gone over badly. Showing up with a forecast didn't.

What the team actually shipped

Customer service is the one I'd point to first. TLC had a call center, a web portal, an email inbox, and a walk-in center that felt like a DMV. Four channels, four sets of dashboards, no shared view of who was contacting the agency or why. We pooled the data.

What surfaced was the same person, working on what was effectively one issue, bouncing between units for weeks. Nobody was handing off badly. The units behind the channels just weren't talking to each other, and the only place that showed up was in the data once it sat in the same place.

Because the team had a designer and a researcher by then, we could do something with what we'd found. Single-owner cases. Shared history across units. Cleaner routing. Wait times dropped. The bounce pattern thinned out. None of it required restructuring the agency.

One ask became the distribution methodology for the Medallion Relief Fund, the city's response to driver debt during the medallion market collapse. We weren't the obvious team for it. We were the team people thought of.

The resistance, and how it resolved

The pushback, when it came, was always a version of why are you asking so many questions, just build the thing. Fair enough. In an institution wired for execution, discovery looks like overhead, and the people pushing back weren't wrong to notice. It didn't resolve through argument. It resolved through wait times going down. Once a few operating units were quietly telling other units we were worth the trouble, the request pattern flipped.

The decision I still think about

The team lived in an operating division. Not IT, not the commissioner's office. I'd love to say that was strategic. It wasn't — operations was where the sponsor with budget and appetite happened to sit, and government windows close fast.

It turned out to be the most consequential choice in the whole story. Down the hall from the people whose jobs we were trying to make easier, we saw the messy parts. Two floors up, we'd have been treated as an inspectorate, and shown the version of the work that was ready for inspection.

What survived, and what travels

When I left, the role was split, and one of the pieces became the agency's first chief product officer — a seat that didn't exist when I started and emerged because the work had made it obvious one should.

Three things travel. Take the opening that's close to operations, even when a tidier org chart would put you elsewhere. Sequence the team around credibility, not capability. And expect discovery to read as overhead until you have a portfolio of wins to point at — then it reads as the reason the wins happened.

A van-sharing pilot, built in weeks, to change the conversations the company could be in.

Bandwagon was a shared-mobility startup splitting taxi rides between riders heading the same way from New York's transportation hubs. The product worked. The strategic problem was who could buy us. The natural exits — TNCs, taxi vendors — all had their own pooling tech or were building it. We were a feature, not a company, from those buyers' point of view. Vans were the angle we picked to get into a different conversation.

What I want to write about is what we did next — a partnership with Transdev to launch a van-sharing pilot called Bandwagon Transit, built and deployed in weeks, designed to do something more specific than test whether van-sharing worked.

What the pilot was actually for

Chariot was building momentum as a commuter-van service, and the category was open enough that a credible van play could put a new kind of buyer at our table. The pilot wasn't really a market test. It was a vehicle, in both senses, for being in a conversation we hadn't been able to get into.

That changes what "ship" had to mean. We didn't need to optimize the route, refine the pricing, or scale the operation. We needed something real enough to generate press, conversations, and a credible claim that we operated in this space. Anything else was deferrable until one of those conversations turned into something.

How we shipped it fast

Transdev provided the drivers and the Mercedes Sprinter vans. That single decision saved us months — we weren't building a fleet, hiring drivers, or wrangling commercial insurance. We were a software and operations layer on top of someone else's infrastructure.

For the customer-facing layer, we repurposed almost everything Bandwagon already had. The on-the-ground sales staff working the airport queues moved to the new pilot. Existing signage and palm cards got light edits and a fresh visual treatment. We bought swag online.

That single call got us operating without months of permitting, and the donation data told us what riders actually valued the service at when they had the choice not to pay at all. For the route, we analyzed gaps in the MTA's network and picked Penn Station to Grand Central — a corridor where a meaningful number of trips weren't well-served by the existing transfer-heavy subway connection.

The whole stack — vans, drivers, brand, signage, route, regulatory posture — was assembled in weeks. It wasn't beautiful. It was real, and the difference between "we have a deck about a van service" and "we have vans running between Penn and Grand Central, here's the press" was the entire point.

What the work was, and what the company was

The pilot did what it was built to do. The press landed. The conversations we'd built it to create happened. We sat at tables with partners we hadn't been able to reach before. Those conversations didn't ultimately convert into the deal that would have saved the company. Bandwagon didn't make it.

Conflating those — claiming the pilot failed because the company didn't survive, or claiming the company would have survived if the pilot had been better — would be dishonest in either direction. The deal that didn't close was a separate decision, made by another company, on factors that had little to do with whether van-sharing was a viable category. What I took from it isn't a startup-survival story. It's a muscle.

What travels

If the conversation you're hoping for isn't presenting itself, you can build something to create it. Not a deck. Not a memo. Something real, on the ground, that changes the facts the conversation is based on.

This is the muscle that gets least respected in formal planning environments, where the default is to study, model, propose, debate, approve, plan, and only then deploy. I wish more transit planning worked the way Bandwagon Transit was built. Too many decisions get made top-down and committed to permanently when the right move would be a six-week pilot that generates real ridership data, real operational learning, and real political room to maneuver. Building fast doesn't mean building badly. It means building enough to put yourself in a different conversation than the one you'd have had otherwise.

Turning a pile of good artifacts into a coherent case the right people could act on.

Mentoring has been a through-line for a decade — managing teams at Ad Hoc and TLC, advising early-career technologists through Coding it Forward and Technologists for the Public Good. This is one written-up example.

The mentee was an early-career data professional inside a major regional transportation authority. Engineering undergrad, master's from a strong urban planning program, several years of analytical work. He'd shipped real things — a fare-evasion dashboard pulling 100M+ rows from five sources, a federal funding allocation tool. The kind of work that's hard to fake.

The premise of most mentoring writing is that the mentor brought something the mentee didn't have. That's not what happened here. The substance was already there — decks, system diagrams, intern templates, ideas about how the team he wanted to build should operate. None of these were bad. But they were a pile of artifacts rather than a case, and they were pitched at a level of technical detail that wouldn't move the people who could actually unlock resources for him. The gap wasn't engineering. It was the framing that turns a portfolio of good thinking into something a director can champion upward.

The advisory work

The first move was refusing to start where he wanted to start. He came in with a goal stated as build the data warehouse and a question about how to pitch it. I held us on the prior question — what is the warehouse for, and what does landing it actually get you — for longer than was comfortable.

By the end of that conversation the goal had been restated: build the case that gets me the team and the IT relationship I need to do this kind of work for the next decade. The warehouse is one piece of evidence in that case.

From there the work organized itself around three deliverables, each pointed at a different audience. A data operations document — the technical operating manual, written for him, his interns, the IT team, any future data hire — that defined how data flowed through the department as a system rather than a series of one-off projects. A data strategy one-pager, written for his Assistant Director and Director, that translated the technical reality into a leadership story. And a management framework for the interns — collaborative goal-setting, a 1:1 cadence, a retrospective process — designed to be tested with the unpaid undergraduate intern starting in February and refined for the paid graduate hire arriving in June.

The artifacts weren't new ideas. The work was deciding which audience each one was for, what each had to do, and in what order they got built.

What changed

Once build the warehouse became build the case the warehouse sits inside, sequencing decisions stopped being arbitrary. The strategy one-pager had to come before the warehouse pitch, because IT couldn't be asked to plan against the warehouse without leadership having committed to the broader investment first. The management framework had to be tested in the spring, not the summer, because the paid graduate hire was the higher-stakes engagement. The IT relationship had to be approached through the Assistant Director, not directly, because IT bottlenecks don't unblock for analysts — they unblock for the directors analysts work for.

None of those calls were obvious from inside the work. They became obvious once the goal had been restated.

Where it landed

He's actively pitching the case to leadership now. The strategy one-pager exists. The ops doc is in draft. The undergraduate intern is onboarded against the management framework, which will get its revision before the summer hire arrives. The substance was always going to come from him. What the engagement did was compress the time between "I have the pieces" and "I have a case I can take to leadership." Government windows are short. An analyst who arrives at a coherent pitch six months later than they could have, often arrives after the moment to make it has closed.

What travels

The case isn't landing because the artifacts are pointed at the wrong audience, in the wrong order. The pattern isn't about data, or government, or early career — it's the gap that shows up wherever someone technically capable is making an institutional case for the first time. I've done close to the same exercise with a startup CEO trying to get a board on board with a pivot, and with a foundation officer trying to get a program approved internally. The setting changes. The shape of the work doesn't.