Sundry Systems

What I build

The first thing most firms ask about is transaction categorization.

Every month someone on your team opens a client's bank feed and codes a few hundred transactions, most of them the same vendors as last month. Then they spend the rest of the time chasing the handful nobody recognises.

I automate the repetitive part and make what is left faster to work through. The output is not a pile of coded transactions you have to take on trust. It is a shorter list of actual decisions, each one with a written reason attached explaining why it needs you.

Firms also ask for client query lists, statement chasing, close checklists, and moving data between two systems that were never built to talk to each other.

What it will not do

Most of what matters in this work is knowing where automation should stop.

Nothing posts because a model said so.

Every transaction coded automatically is coded because the client's own history says that vendor goes there. The model only touches rows your team has never seen before, and when it does, its job is to write the explanation rather than to make the posting.

Nothing auto-posts to a suspense account.

Anything landing in Ask My Accountant or its equivalent is forced to human review regardless of how confident the system is. A transaction nobody can identify is precisely the transaction a person needs to look at.

A person approves everything before it posts.

Your review is your professional exposure. The tooling is built around it, never through it.

Who I am

I am Jeff Lee. I build these systems myself, with a background in finance and software.

I have working OAuth integrations running against both QuickBooks Online and Xero, tested against live sandbox data, including the parts that tend to break quietly in production: token refresh, and rate limits under real concurrency.

Where things stand

I am taking on a small number of firms at no charge right now.

I will spend thirty minutes going through how your team actually works, then send you a written breakdown of what is automatable, what is not, and what I would sequence first. No charge, and no pitch at the end. I am building case studies and I need real firms to look at.

Sometimes that breakdown says categorization is the wrong place to start, because it needs your transaction history clean enough to learn from first. That is the sort of thing it is better to know before anyone builds anything.