Anonymized under-retail business · Door-to-door reconciliation.

A Evidence level

Datawhale FDE Case 100 | No. 10: From manual checks to intelligent reconciliation for offline retail

Publisher: Datawhale FDE100 · Official case collection PDF · Direct sources at the case level

Claim origin: Disclosure by the author of the case · Independent verification: No · Accessed: 2026-09-19

Evidence level measures whether a source can be located and reviewed; it does not mean vendor-reported claims were independently audited.

Standard case | Useful reference with remaining gaps 8 / 12

Business context1 / 2

Transformation workflow1 / 2

Technical workflow2 / 2

Human roles & governance1 / 2

Measured outcomes1 / 2

Source traceability2 / 2

Remaining gaps: Business context, Transformation workflow, Human roles & governance, Measured outcomes

The key to this case is not the single-point use of AI, but the re-establishment of the “door reconciliation” into an enforceable, verifiable production stream with boundaries of human responsibility.

Business problem

Multiple shops, multiple payment channels and multi-system bills require manual checking by finance staff.

Solution

Uniform billing, automatic matching of transactions and the assignment of discrepancies and anomalies to finance staff.

Technical architecture & production workflow

Step 01

Invoices/documentation/billing/operational materials

→

Step 02

OCR/ field extraction + business classification

→

Step 03

LLM/rules judge business types and risks

→

Step 04

Determination matching/verification/reconciliation

→

Step 05

RPA/API in ERP/OA/Financial System

→

Step 06

An anomaly, a manual review, a rewriting of the results.

Key technology & infrastructure components

OCR/Info extractionLLMRules EngineRPA/APIERP/OAHuman-in-the-loop

Human roles & accountability

The finance staff only handle unmatched, unusual and high-risk transactions.

FDE delivery actions

Shawowing finance, decomposition of “Looking at the material-judgment-checking-checking” into atomic action.
Distinguishing semantic judgement, mechanical execution and responsibility clearance of three types of tasks
Make it visible and create an unusual type of old employee's covert audit rules.
Priority is given to the existing ERP/OA without re-engineering of the system
Set conditions for manual upgrades such as low confidence, monetary thresholds, rule conflicts, etc.

Reusable delivery patterns

AI is responsible for judging, RPA/API is responsible for determinative enforcement, and people are responsible.
Automate the standard and concentrate on the anomalies.
The core of the finance category projects is not chatting, but auditable programming

Business outcomes & delivery results

The objective was to move from a paper-by-written check to an anomaly; public information was not disclosed as a unified ROI.

Evidence boundaries & verification notes

Public information confirming business processes; technical components are abstract architecture based on public description
The source can be directly located in the case; the value remains disclosed by the source and does not represent an independent audit.

Primary source: Datawhale FDE Case 100 | No. 10: From manual checks to intelligent reconciliation for offline retail ↗ Additional sources: Datawhale FDE100 case webpage | No. 10 ↗ Additional sources: Secondary collation or aggregation source used in the old version ↗