Field notes.

The thing we actually built was legitimacy

A large bank's in-house agency made printed things. Over a decade, a few of us built the digital side of it into a multi-million-dollar business. Almost none of the difficulty was technical.

Context

The agency existed to produce print. Brochures, signage, statement inserts, direct mail, in-store. It was good at it, and everything about the place was shaped accordingly: how work was estimated, how it was approved, how long it took, who signed off, what a finished thing looked like, what counted as a real project.

Meanwhile the bank's customers had already moved online. The demand for digital work arrived before anything existed to absorb it, which is the normal order of events and still surprises everyone every time.

Friction

The obstacles were almost entirely structural, and each one looked reasonable from the inside.

Digital work didn't fit the estimating model, so it was hard to price and easy to treat as a favor. It didn't fit the approval model, which assumed a fixed artifact reviewed once rather than something that keeps changing after launch. It didn't fit the calendar, because print has a press date and digital has a release schedule, and those are different kinds of deadline with different consequences for being wrong.

There was also a quieter problem. Every early digital win was attributed to the individuals who pulled it off rather than to a capability the organization now had. That distinction sounds academic. It decides whether you get budget next year.

Insight

We were treating this as a skills problem when it was a legitimacy problem.

The bet everyone else was making was that if the work got good enough, the organization would adapt around it. That isn't how organizations behave. Good work buys you attention, once. What converts attention into permanence is being legible to the systems that allocate money and headcount, which means being estimable, reviewable, staffable, and boring in the specific ways an organization needs things to be boring.

So the real project stopped being the work and became the scaffolding around the work.

Action

We proved it on things people already cared about rather than on a showcase nobody had asked for. Visibility came from being useful on a priority, not from a demo.

Then we made it repeatable. Estimating models that fit iterative work. A review process that senior people could actually run without being trained in a new discipline. Standards for what good looked like, written down, so quality stopped depending on who happened to be staffed.

We hired ahead of demand, which felt reckless and wasn't, and we hired leadership rather than only practitioners, which is the step most teams skip. And we pushed to get the capability into the annual planning and funding conversation, so it was a line in the budget instead of a thing that happened when someone was generous.

Outcome

An award-winning digital creative practice serving consumer bank marketing across every digital channel, which grew into a multi-million-dollar business with a team, an operating model, and a seat at the table it hadn't had.

The more durable outcome was that it stopped being fragile. It survived reorganizations, budget cycles, and the departure of people who had been essential to it, including eventually me.

Hindsight

The most important thing we built wasn't a website or a digital team. It was legitimacy.

What I'd underline for anyone standing where we stood: the stage everyone underestimates is not proof and not sponsorship. It's the one where you have to hire someone to lead the thing instead of hiring more people to do the thing. Founders of a capability are usually reluctant here, because the capability still feels like theirs. That reluctance is how a promising practice quietly stays a two-person heroics operation for four years.

New capabilities rarely fail because the technology doesn't work. They fail because the organization never changes enough to let them take root.

The five stages, and where each one dies

Where AI actually fits in a research organization

A research and development group with real interest in AI, a lot of vendor noise, and no clear answer to where any of it belonged in how discovery work actually got done.

The friction was that every conversation started from the tool catalog. Which model, which platform, which vendor. That framing makes it impossible to judge anything, because a capability only has value relative to a specific step in a specific workflow, and nobody had written the workflow down.

The insight was that the informal parts of the process were the important parts. The documented process was a clean sequence. The real one was full of judgment calls, side conversations, and knowledge that lived in a handful of heads. Those are precisely the steps where a model either helps enormously or produces confident nonsense, and you cannot tell which without looking closely.

What we did was map the end-to-end service, including the backstage operational work nobody documents, and then test candidate opportunities against it. Some collapsed immediately. A few got substantially more interesting once the surrounding process was visible.

What came out was a blueprint the team could act on and, more usefully, a shared picture of which opportunities were real and which were demos waiting to disappoint.

What I'd tell you now: if you can't describe how the work happens today, including the informal parts, you have no basis for judging where a model helps. The knowledge architecture comes before the tooling. Most AI strategy work fails because it starts from the catalog and reasons backward toward a problem.

Institutional clients, no agency behind me

I started my own firm and went after work that normally goes to shops with a floor of people behind them.

The friction was obvious and entirely fair. On paper I could not match an agency on capacity, process maturity, or the comfort of a recognizable name on a procurement form.

The insight was that those weren't the things the buyer was actually anxious about. What they wanted was someone who understood the business problem well enough to be trusted alone in a room with their stakeholders. Capacity is a real constraint, but it's rarely the thing that loses the work.

So I competed on judgment and proximity. Fewer clients, closer relationships, and a willingness to say when the thing being asked for wasn't the thing that would help.

The result was brand, identity, and digital work for the Carlyle Group, ProFunds, the Mortgage Bankers Association, and Reed Business Information. The firm still operates.

What I'd tell you now: running your own shop teaches you things being good at the craft never will. How buyers actually decide. What a proposal has to accomplish before anyone reads the work. Why the better option sometimes loses, and what you could have done about it three weeks earlier.