How a distributor turned a WhatsApp group into an audited order ledger

A consumer goods distributor takes orders through a WhatsApp group. The same conversation also carries payment updates and follow-ups to earlier orders. Someone had to monitor every message before a valid order could enter accounting. The agent now classifies new messages and writes clear orders to a structured ledger linked to the source message. Accounting receives the messages that need a person.

The situation

Order details arrived in chat, and follow-ups often referred to earlier orders without naming them. Turning that conversation into accounting records required continuous monitoring and manual re-entry. The team performed this clerical step outside the accounting system.

Documented rules

We turned the company's order-entry requirements and data controls into instructions the agent follows.

01The ledger database stores orders, their line items, review items and approved sales targets.Data
02A clear current order becomes a ledger record. An ambiguous message becomes a review item for accounting.Order entry
03Payments and changes to earlier orders keep their own classification and do not create new orders.Order entry
04The ledger records each source message once and uses that record to prevent duplicates.Data
05The audit trail stores counts and internal identifiers. Personal details remain in the operational database.Privacy
06People edit sales targets through the dashboard, and the agent reads the approved values from there.Targets
Rules held by the team

Conversations with the sales and accounting teams surfaced the judgment calls missing from the written requirements.

A follow-up may refer to an earlier order without naming it. The agent uses the surrounding conversation to identify the customer.Source: Sales teamOrder entry
Names used in the group often differ from the official customer and product records. Accounting validates the match.Source: AccountingData
The agent excludes messages sent before the agreed cutover point from order entry.Source: ManagementProcess
If a selected customer fails validation, the agent stops and sends the item to accounting.Source: AccountingData
Private reminders about sales targets require explicit approval before they are sent.Source: ManagementTargets
Authorized employees can ask questions in the group by mentioning the agent by name.Source: ManagementAccess

How the agent went live

The team first defined what creates an order and what requires accounting review. The agent tested those rules against past messages before it began processing new conversations.

Timeline
Phase 1
Connect the workflow
The team connected the WhatsApp group to a dedicated ledger and an accounting review portal.
Phase 2
Test the order rules
The agent classified past messages using the new rules. Processing began from an agreed cutover point.
Phase 3
Limit the operating scope
The agent handles order entry and accounting questions. It declines requests outside that scope.
System integration

The agent connects to the tools the team already uses, so sales and accounting kept their existing interfaces.

W
WhatsAppSource messages and employee questionsConnected
N
Neon PostgresOrder ledger, review state and sales targetsConnected
A
Accounting portalItems that require human reviewConnected
D
Sales dashboardCurrent reporting for managementConnected
Guardrails

One message, one source record

The ledger records each message once and uses that source record to prevent duplicate orders.

Personal details stay out of the audit trail

Audit records contain counts and internal identifiers. Names, phone numbers and message text remain in the operational database.

What the agent does

The agent checks new group messages every hour and records each one once. A clear order becomes a ledger record linked to its source message. If context is missing, accounting receives a review item with the original conversation attached. Other message types keep their own classification, so the agent does not invent an order.

Hourly

Order entry

A valid order message creates an order and its line items in the ledger. The record preserves a link to the message that created it.

Live

Management dashboard

The dashboard reads sales and target data from the same ledger that receives new orders, giving management a current view of performance.

Skills in production
Order entryturns valid chat messages into linked ledger records
Accounting reviewsends ambiguous items to a person with the source context
Dashboard reconciliationchecks management reporting against the order ledger
Target reminderscalculates progress and waits for approval before sending

What changed

Daily workload
24/day
scheduled group checks
one each hour
9.6/day
messages classified
each recorded once
1.8/day
orders created
entered directly from chat
5/day
review items created
sent with source context

The recorded messages fell into five operating categories:

ClassificationShareResult
New orders19%created in the ledger
Needed a person25%sent to accounting with context
Payments and order changes11%kept in their own workflows
Follow-ups needing context21%sent for review
Not an order24%recorded and left alone

The agent covers about 65 hours of recurring order-entry work each month.

What changed for the team

The sales team keeps working in WhatsApp. The conversation now feeds an audited ledger, and accounting receives a review queue with the source context attached.