[N] weeks | Trading firm
From a trading journal to a trading system
The situation
A small trading firm ran on its principal's swing-trading method. The method was well documented, but only as written notes and annotated chart screenshots. An earlier tool built by an outside contractor was no longer usable by the firm.
The firm believed the system was "mostly built." The first question was what was actually running.
What we did
We started with a two-week audit. The honest finding: the method was excellent, and almost none of it was automated.
The first request was an AI agent that would pick trades. We recommended against it. The method depends on judgment, and an unreliable auto-signal would put real money at risk. The firm's own notes said the thing that mattered most was discipline.
So we treated each trade as the unit and built around it:
- Rulebook: the firm's risk and grading rules, written down precisely for the first time.
- Intake and review: a pre-trade checklist that runs every position against the rulebook before it's opened.
- Engine: chart alerts for the parts of the method that can be defined exactly.
- Delivery: a structured trade journal and a weekly report card against the firm's targets.
The trade decision stays with the trader. The system makes it hard to skip the rules.
What changed
[Outcome metrics, e.g. % of trades checked against the rulebook, rule breaches caught, hours of weekly review saved.] The firm owns the code, the rules and the data.
Example build | Bookkeeping firm
Payables exceptions, per exception
Illustrative. This shows what an engagement looks like; it is not a client result.
The work
A bookkeeping firm handles payables for [40] small-business clients. Most invoices match their purchase orders. The rest (wrong amounts, missing POs, duplicate bills) are chased by hand by senior staff.
The unit
One resolved payables exception.
What we'd build
Invoices and POs arrive through the firm's existing accounting software. Agents match them and draft a resolution for each exception: the likely cause, the vendor email, the correcting entry. The rulebook holds each client's quirks and approval limits. Low-value, high-confidence fixes go straight out; anything above a set amount goes to a senior bookkeeper. Clients see open exceptions on a dashboard.
What it changes
Senior staff time moves from chasing mismatches to reviewing the few that matter. The firm can take on more clients without hiring at the same rate, or offer exception handling as a priced service.
Example build | Freight forwarder
Customs classification, per shipment
Illustrative. This shows what an engagement looks like; it is not a client result.
The work
A freight forwarder classifies goods and prepares customs paperwork for [N] shipments a month. Every line needs the right tariff code. A wrong code means delays, penalties or overpaid duty.
The unit
One classified, document-ready shipment.
What we'd build
Shippers upload invoices and packing lists. Agents extract each line item and propose a tariff code with the reasoning behind it. The rulebook holds past rulings, the forwarder's own classification history and known problem categories. Lines that match a previous ruling go through; new or ambiguous items go to a licensed broker. Shippers track status in a portal.
What it changes
Brokers spend their time on the hard lines instead of retyping the easy ones. Turnaround drops from [days] to [hours], and every decision has a written reason when customs asks.