The AI Field Guide for Regulated Finance
Most AI advice for finance stops at the demo. It shows you a clever answer and leaves out the part your compliance function actually cares about: can it run the same way every time, can you show where the answer came from, and can you hand it to a colleague without being in the room.
This field guide is the opposite. It is a five-part, practical series on building AI inside a regulated finance firm, from the first reusable skill to the infrastructure underneath it all. Each part stands on its own, and together they describe how a workflow goes from one analyst's clever prompt to something the firm owns, governs and can evidence.
You can unlock and download the full five-part series as PDFs (one email unlocks all five), or read the summaries below.
01 · Skills: capturing a process so it runs the same way every time
A skill is a captured process: the steps, the sources, the checks, written down so the system runs them the same way on every instance and leaves something behind you can show a regulator. This is the difference between a prompt that works when the expert runs it and a method the firm can rely on. Part one is about turning tacit judgement into a reusable, inspectable unit.
Read AI skills for finance.
02 · Connections: wiring Claude into systems you already govern
An answer is only as defensible as its source. Part two is about connections: wiring the model into the systems you already govern, through MCP, so the answer comes from the record rather than from a spreadsheet somebody exported last Tuesday. Permissions, data boundaries and provenance are the point, not an afterthought.
Read Connections: wiring AI into systems you already govern.
03 · Plugins and agents: something a firm can actually deploy
A skill on your laptop is not a capability the firm has. Part three is about packaging skills, commands and connectors into one installable unit, with its permissions attached, so a colleague can run it without being in the room when you built it. This is where individual technique becomes something a team owns and deploys.
Read Plugins and agents: packaging AI a firm can deploy.
04 · Prompting: getting a defensible answer out of a probabilistic system
Prompting is the most visible skill and the most misunderstood. Part four is about getting a defensible answer out of a probabilistic system, and, just as important, knowing which parts of the answer you are still obliged to check yourself. The goal is not a clever prompt. It is an answer you can stand behind.
Read Prompting for a defensible answer in regulated finance.
05 · The Frontier: what we are building underneath all of it
The last part is the one most vendors skip: the infrastructure. Attestation so every action carries its source and sign-off, knowledge graphs so context does not reset, adapters that learn from corrections, and an honest account of the part nobody has solved yet. This is the engineering that turns the first four parts into something durable.
Read The Frontier: attestation, knowledge graphs, and what is not solved yet.
The through-line
The series is really one argument told in five parts: AI becomes a capability, rather than an activity, when the method is captured, connected to governed systems, packaged for the firm, prompted for a defensible answer, and built on infrastructure that leaves proof behind. That is the same discipline we describe in our operating model for AI where the output has consequences, applied to the specific tools a finance team uses.
If you work in law rather than finance, the same series exists for you: the AI guide for legal practice.
Want the whole thing on your desk? Download the five-part finance series, or send us a workflow and we will tell you plainly whether it is worth building.