Qeasy Smart Reconciliation · Running in production at large-scale e-commerce companies
Multi-platform statements, refunds, subsidies, commissions, shared expenses — every penny a platform pays you is automatically matched line by line against your own books. Whatever doesn't match comes with the business reason and a suggested action; whatever does flows straight into Kingdee, Yonyou or SAP. No more 'manual exports + row-by-row Excel checks' for your finance team.
Marked as 'matched' on the books, yet the difference was never truly explained — that's the reality for most e-commerce finance teams. Reconciliation was never as simple as subtracting one column from another.
At month-end, three people eyeball six platform statements against one supply-chain export, row by row
Statements from every platform are collected automatically and matched order by order — people only review what doesn't match
Platforms only give you the net amount after deductions — subsidies, coupons and compensation all lumped together
Every deduction type is itemized; each difference is auto-attributed to a concrete business reason with a suggested action
Refund overhang and cross-period settlement: money refunded before goods return, this period's cash for last period's orders
Refund-only, instant refund, FBA refund and cross-period income are auto-identified and attributed to the right accounting period
Shared expenses allocated by gut feel — nobody knows which SKU actually makes money
Shared expenses are auto-allocated to stores and SKUs, revealing true per-item profit after allocation
Reconciliation is only the starting point. Every reconciled datum feeds fine-grained accounting across five dimensions — store, SKU, gross profit, tax and cash — ready for business decisions.
Ten JD stores, twenty Douyin shops, three Amazon marketplaces — all unified into one financial view, with cross-store consolidation and combined statements in one pass.
Trace every order down to each item: per-SKU revenue, cost and allocated expense, with per-item profit precise to the cent — loss-making SKUs at a glance.
Seven deduction layers from GMV to net profit: net receipts → platform commission → service fees → ads → logistics → refund losses → shared expenses. Every layer evidenced.
Platform-withheld product tax, shipping tax and promo-rebate tax are separated automatically; cross-border multi-currency (USD / EUR / JPY), EU VAT and export rebates covered.
Sales collections, refund recoveries, platform rebates, procurement, logistics, commissions, taxes — four inflows and four outflows, finely modeled, with collection forecasts and close cadence under control.
From reconciled to posted: sales, refund, commission and adjustment vouchers generated automatically; provisional receivables pushed down to financial receivables; out-of-the-box integration with Kingdee, Yonyou and SAP.
Many platforms, one set of books
JD nets whole orders; Douyin matches in five rounds; Alipay has three statements (aggregate settlement, account flow, Yulibao); Amazon reconciles receipts and refunds separately; AliExpress has five bill types plus fully-managed JIT. Every platform's rules are built in as ready-to-use capabilities — new statement formats are adapted within days, so your finance team never has to study each platform's statement documentation again.
Promo-season surges, handled
Before the campaign: pre-run reconciliation rules, stress-test data, simulate collections. The first day after: results ready. The platform processes a million statement lines a day and aggregates 100k-order reconciliations in seconds. With the CFO cockpit, sales, margins and collection cycles are monitored in real time, and anomalies like failed matches and refund overhang alert automatically — the 'reconciliation surge' is routine for the system, and no longer overtime season for finance.
AI finance assistant
'How many unreconciled differences are left on Douyin this month?' 'Adapt the new platform's statement format for me.' Business users state needs in natural language; the AI assistant reconnoiters, writes and tests on its own — humans confirm once before launch. When a platform changes its rules, you respond the same day instead of queueing for development.
Measured across 28 mid-sized e-commerce rollouts in 10 industries (annual GMV ¥50M–¥5B): returns come from labor, error and compliance savings, plus the cash-flow benefit of a faster close.
Source: smart-reconciliation rollout sample, Aug 2026 (28 companies across 10 industries: beauty, 3C electronics, apparel, home, mother & baby, food, sports, pet, auto aftermarket, appliances).
One-click import from every platform — no manual exports, no missing statements
Three-way match across platform statements ↔ orders ↔ your books; matches pass automatically
Mismatches are auto-tagged with 23 business reasons, each with a suggested action
Shared expenses auto-allocated to stores and SKUs, writing back true profit
Results pushed to Kingdee / Yonyou / SAP in one click, vouchers auto-generated, fully audited
Supports 'daily clear, daily close': daily reconciliation, weekly margins, monthly P&L, quarterly balance sheet, annual tax settlement — the cadence follows your management needs, not just a month-end sprint.
23 built-in e-commerce variance reasons. Every unmatched line is no longer 'difference ¥245.04, handled manually' but a clear reason plus a clear suggested action — readable by new hires, acceptable to auditors.
Platform-funded coupons and instant discounts, auto-accepted within ¥0.01
Platform compensates the buyer after dispute rulings and deducts from you — auto-linked to the order
This period's refund against last period's order — income reversed in the right period
Refund overhang — money returned, goods not — flagged and tracked
Platform pays in installments — settled and pending amounts booked separately
Postage / price make-up orders identified, differences auto-billed as expense
Zero-price gift lines flagged, with a reminder to check for missed receivables
Product, shipping and promo-rebate taxes itemized separately
Finance states needs in plain language; the assistant reconnoiters, writes and tests on its own — humans confirm once before launch. Four assistants, each with its own job, dispatched from one entry point.
bill-parse-agent
When a statement arrives in a new platform or format, describe it and the assistant learns to read it — validated before it goes live.
reconcile-script-agent
Runs line-by-line matching under your rules, explains every mismatch in business terms, and verifies results before you rely on them.
expense-allocate-agent
Allocates shared expenses to stores and SKUs under your rules, writes back true per-item profit, with verifiable results.
general-assistant
'How many unreconciled differences on Douyin last month?' Ask in plain language — it queries, delegates and answers from one entry point.
The screens below come from the real system running in production (demo account) — what you see is what you get.
Domestic coverage spans JD POP, Douyin Shop, Pinduoduo, Xiaohongshu, WeChat Channels, VIP.com and the Alipay funds system; cross-border coverage spans Amazon, Temu, Shopee, Lazada, eBay, AliExpress, Shopify storefronts and Newegg. Every platform's reconciliation rules are built in, with new platforms onboarding continuously.
JD POP
Douyin Shop
Alipay
Amazon
Pinduoduo
Xiaohongshu
WeChat Channels
VIP.com
Temu
Shopee
Lazada
eBay
AliExpress
Shopify
NeweggPost results in one click: provisional receivables pushed down to financial receivables, expense variances auto-billed as payables, vouchers generated under your chart of accounts. Kingdee Cosmic receivables can join reconciliation at line level — a closed loop across platform statements ↔ business orders ↔ ERP accounting, with no manual data shuttling between systems.
Continuously updated from the production frontline: multi-platform reconciliation manuals, variance-handling playbooks, fine-grained accounting methods, cross-border specials and customer cases.
金蝶云星空、用户系统里的对账结果要生成应收单、应付单，自研 ERP 要的是另一套字段……这套「目标系统千差万别」的中间产物，怎么在 Prisma schema 里只占 4 张表、不随集成数线性膨胀？答案就是「transform_documents × payload JSONB × columnSchema × editableFields」这一组通用骨架——骨架字段固定、目标系统差异全收敛进 JSONB、列声明与编辑白名单随批次快照落库、金额字段一律禁改。
同一句话「对账」，在京东 POP、抖店、支付宝、亚马逊、速卖通五个平台跑出来的是五种不同的判定逻辑——京东用整单轧差把同一订单的收款和退款合并成净额再判、抖店沿用同一种整单轧差但把判定轮次刻意做少靠多证据链单点爆破、亚马逊坚持分向不轧差让收款与退款各对各的、支付宝靠三张表分别告诉你「平台给了多少」「在哪个账户」「生了多少利息」、速卖通用 4 个资金 billType 拆开两套资金账户再叠 1 张不入金蝶的 JIT 维表。这篇文章不是平台对比表，是把这五种「判定权威」背后的工程逻辑拆给你看——读完你会知道为什么亚马逊的对账失败率能从 20% 压到 5% 靠的不是多写代码而是认清「分向不轧差」是账单本质决定的。
2024 年 1 月，财政部《企业数据资源相关会计处理暂行规定》正式施行，把"数据资源"从治理议题推到会计科目——而决定一项数据能不能从"原始流水"走到"无形资产"入表的，是企业数据治理的成熟度水平。同样是 5 亿条订单数据，治理成熟度 3 级的电商企业可以做出 95% 自动对账的核算体系，治理成熟度 1 级的同行依然困在"每月对账 8 天、差异靠 Exce…
电商对账不是「两列数字相减」，而是一套**合规风险控制系统**——5 大雷区（收入确认时点错配、退款倒挂、平台补贴未入账、跨期收入错配、代扣税未计提）每年让中型电商公司产生 0.3%-1.2% 的「假利润」，审计一来就得做红冲、做追溯、做说明。本文不讲术语堆砌，从一份真实的月结错账复盘出发，把每条雷区的触发场景、合规口径、规避动作拆给你看。读完你会知道：合规对账的护城河不在 Excel 表里，而在**机读差异标签**（diffReason）+ **沙箱分摊反写** + **物料级追溯链**这三件事上。
AI 系统的知识库不是「写一次吃一辈子」，它会随代码迭代、随业务更新、随 md 文档修正而持续变化。这篇文章不讲 RAG 原理，而是讲一个更接地气的问题：怎么让「拉代码 + 重启服务」一行命令搞定知识库更新，让运维不需要关心这次改的是 md、是 seed、还是 schema 注释？答案是一个 70 行的 Node 脚本 + 一份调用纪律：post-deploy-knowledge。
2026 年 9 月的一次电商财务沙龙现场调研，47 位业务负责人举手回答"你能熟练用几个数字化工具"——**62% 的人只举了 1-2 个手指**（Excel + 1 个内部系统），能举到 4 个以上（CRM + ERP + BI + 对账系统全用上）的只有 14%。但当主持人追问"那你懂你公司从平台收钱、到 ERP 入账、到 BI 出报表的整条数据流吗"，**举 4 个以上工具的那 14% 里也只剩 4% 能讲清完整链路**。差距不在"会用几个软件"，在 **3 件事**——会用 4 类核心工具、懂端到端数据流、懂与 IT 协作边界。本文给业务人员一份 **3 层能力阶梯**（L1 会用工具 / L2 懂数据流 / L3 懂 IT 边界）+ **每层的能力判定标准**+ **30 天速成路径**。不是教你怎么点鼠标，是教你怎么从"工具操作工"升级成"数字业务骨干"。
Your ERP keeps your internal books, but platform statements live outside it — how much commission was deducted, which coupons were funded, how much tax was withheld. The platform reconciles platform statements, business orders and ERP entries three-way, then pushes results back into the ERP for posting. It doesn't replace your ERP; it covers the part your ERP can't see.
Fifteen platforms are covered: JD POP, Douyin Shop, Alipay (Tmall), Pinduoduo, Xiaohongshu, WeChat Channels, VIP.com, Amazon, Temu, Shopee, Lazada, eBay, AliExpress, Shopify and Newegg. Onboarding a new platform is a routine capability, not a custom project — statement formats are adapted within days, with the AI assistant doing most of the work.
No. Adaptation rules are visual and version-managed, and the AI assistant helps with the change: describe what changed, adjust automatically, validate before enabling — business users respond the same day instead of queueing for development. Every change is audited and rollbackable.
The AI assistant has no direct write path to your books: all validation runs in an isolated environment; every real execution requires human confirmation; every step is audited and rollbackable; and permissions for query / authoring / testing / execution are strictly separated. The AI proposes; humans decide.
We recommend a single-platform pilot: the 'import → auto-match → variance attribution' loop runs within weeks, then expands platform by platform each month, and finally connects to your ERP for automatic posting. Companies completing the full rollout typically see 8–12× efficiency gains, error rates dropping from ~5% (manual) to around 0.1%, with full payback in a median of 26 months.
Our core product Qeasy focuses on enterprise data integration and application building. This smart reconciliation platform is Qeasy's flagship enterprise AI deployment, serving real financial reconciliation at large-scale e-commerce companies — not a proof of concept, but a pipeline running in production every month, turning reconciliation from a month-end sprint into a daily business routine.
Monthly reconciliation pipelines running in production at large-scale e-commerce companies
15 platforms' rules built in; 23 variance reasons codified
Auditable steps, rollbackable versions, tests isolated from real books
New platforms and rules adapted fast; the AI assistant keeps lowering the bar
