Transforming a Securities Firm Into an Ai-Native Firm: From the margin book to the whole firm - Book One - The Margin Core

That letter opens this book, and answering it in under a minute, to the tick, is what the book builds. On the way it names the seven design errors sitting on almost every margin desk - an advance-rate table nobody owns, a call ladder with no liquidity term, concentration measured per client but never per firm, a risk grade that enters no formula, one rate for every collateral book, ratios computed on yesterday's close, and a close-out order improvised under stress. None is a technology failure. Each is a business decision that was never put to anyone who could take it.

The repair is a limit that is the minimum of five independently computed ceilings: lending value, client capacity, concentration - per stock, per sector and across the whole firm - firm limits, and the liquidity horizon. A worked account runs through all five: a portfolio worth $1,560,000 on paper supports a $594,000 loan by value - and only $494,000 by exit, because one holding needs six days to sell. That $100,000 gap between paper value and exit value is where margin books die, and the fifth ceiling is the only one that sees it.

Inside:

• Five capability layers, from routing to close-out, each passing one test: switch off every model and the layer still works;


• A call ladder whose triggers rise and whose grace shrinks as a portfolio's exit time lengthens - and a committed close-out that sold 4,600 liquid shares where the improvised desk sold 150,000 illiquid ones;


• Seven playbooks and a case map with proved reductions: 27,648 combinations to 51 cells to a 335-case acceptance suite that Book Two must pass;


• Findings you cannot see without the sweep - including why the per-stock concentration cap mathematically cannot protect you on illiquid collateral, and why at institutional size every name is a small name;


• Governance for a machine that decides intraday, personal data by design, decisions that replay to the cent - and information barriers built into the data model, assistant included;


• The full engine in an appendix: six small files, 72 columns, offline, with tests reproducing every number in the book.

The honest position, stated on page one. Every parameter is marked illustrative. A signature page names what your head of margin risk, your trading desk and your board must review before any figure is used - and the appendix ends with that table, waiting for initials. The method transfers as it stands; the numbers are yours to set.

For chief executives, heads of risk, architects and regulators of securities firms - and for anyone who wants to see what AI-native means when the output is a forced sale of a client's property and usually right is not good enough.