Small bosses keep their books in WeChat. I turned that into an AI product, and then...

Preface

This account has been registered for almost a year now, and the first article is finally going out 😂

Today let's briefly talk about the first version of the Zhang Xiaozhu product.

Introduction

Zhang Xiaozhu is an AI product I built at the end of last year. The original idea was simple: solve a real problem.

First, the scenario.

In businesses like renovation and construction, money in and out often isn't in accounting software—it's in WeChat.

A client transfers a renovation payment and tosses over a screenshot; materials bought on the job site, a photo of the receipt on the phone; wages settled after a couple of chat messages. The work is done—but the books? Many small bosses either flip through chat history at month-end to reconcile the hard way, or just go by memory.

Zhang Xiaozhu was originally built for exactly this.

Logic

People always say the programmer-startup starter pack is bookkeeping, a diary, and vocabulary flashcards 😂

But Zhang Xiaozhu really isn't yet another “professional bookkeeping app.”

The original audience was small renovation bosses, small construction crews, people without a dedicated accountant—what they actually need isn't a chart of accounts, but: which job site this money belongs to, and whether this project is actually making money.

Usage is equally straightforward: however the business happens in WeChat, that's how you hand it over.

• Send a line like “received Zhang San renovation payment 5000”
• Or just drop a transfer screenshot or an invoice photo
• It recognizes the content and tries to file it under the corresponding project ledger
• If the information isn't enough it asks you a question, instead of recording blindly
• After that you can ask anytime: how much revenue this project has, how much cost, roughly what the profit looks like

Development

The earliest version had very simple features.

Later I went to several users' shops and did on-site research of real user scenarios.

From the results, what they wanted wasn't “AI + bookkeeping chat,” but “project lifecycle management + product cost details + accounts receivable / collections management + project P&L + low-friction AI data entry”

You could call it a simple ERP.

AI's job is to turn spoken language, screenshots, and scattered descriptions into structured drafts; the actual calculation, posting, permissions, audit, and reminders are all handled by a complete backend system.

So the new version's goal became helping users answer four questions at any time:

• Is this project making money right now?
• How much remaining payment hasn't been collected on this project?
• How much did we make on each brand/model/spec of tile?
• Did restocking, returns, freight, and handling fees eat the profit?

Dilemma

Once it reached this level, development costs rose sharply.

Subscriptions from just a few users couldn't cover costs at all, unless we expanded the user base and sold the product to more customers.

But that involves the sales problem...

So this project seems to have stalled here.

Still, I later found a new direction for it. I'll introduce the details in the next article—stay tuned~