td finance
What a large bank looks like from the inside is not what it looks like from any external description of it. The org chart is real but incomplete. The processes that keep things running are partially documented at best, and in many cases not documented at all — they live in the people who have been in their roles for five, ten, twenty years, inherited from predecessors who did it the same way and never had occasion to write it down. In 2022 I moved from the capital markets side of TD into the finance function, working on the modernization effort ahead of automation deployments. The first thing that became clear was that 'modernization' presupposed a level of visibility into existing workflows that often didn't exist.
The SOP sessions were where this revealed itself. You'd sit down with someone who had held a role for fifteen or twenty years — someone who was genuinely good at their job, who knew every edge case, every counterparty quirk, every exception that the official process didn't account for — and you'd ask them to walk you through what they did. And what you got was not a process. You got a person. The documentation you were supposedly transcribing didn't exist anywhere except in how they moved through the workday. They'd never been asked to explain it before, not because anyone was hiding it, but because it had never needed to be said. It was just the job.
What those sessions surfaced, consistently, was structural misalignment that had accumulated over years without anyone being accountable for it. A role whose scope had grown past what one position could reasonably carry, the overage absorbed quietly because the person in it cared about the work. Adjacent roles doing reconciliation that only existed because an upstream system had been built with a gap in it, and the gap had been patched by adding a human step. Three positions whose combined function could have been one if the department boundaries had been drawn differently. None of it was visible from the org chart. It only appeared through the act of sitting with people and asking them to explain what they did and why.
The automation work — RPA, the tooling we were mapping workflows for — had an implicit precondition that we kept running into: it could wrap a stable, well-understood process and execute it reliably. But stable and well-understood were the things we were still establishing. An automation built on an incompletely documented workflow fails at the edges of the documentation. The bot does exactly what the SOP says, and the SOP was silent in the places where the practitioner had been improvising for years, because the improvisation was the tacit knowledge that had never made it into any formal record. When something broke in production it looked like a technical failure. Usually it was an epistemological one.
This was 2022 and 2023, which meant it was happening just before AI made conversations about process documentation and role restructuring feel newly urgent in the broader market. The large consulting firms would spend 2024 and 2025 selling the insight that large organizations had significant latent inefficiency that new tooling could identify and restructure. We were already living in the working conclusion of that insight, except the path to it ran through the people in those roles, not through any algorithm. The distinction that mattered wasn't technical. It was about what you thought the goal was — whether the SOP work was preparation for reducing headcount, or whether it was an attempt to understand the full value of what those people actually carried. The colleagues who became our guides through those sessions were often the most capable people in the function. They'd been keeping things running because the system around them wasn't well enough designed to run on its own.