E-commerce Operations Desk Official

Every two hours it reads open orders from the storefront and their tracking from the shipping tool, finds anything past its promised date or flagged by the carrier, and drafts the customer update and, when the playbook allows, the carrier claim. A person sends and files.

1. 01If an order is given ({{order}}, {{customer}}, promised {{promised}}, {{carrier}} {{tracking}}, exception {{exception}}) work on that one only. Otherwise read every order not yet delivered from the storefront and its tracking from the shipping tool. List each order past its promised date or with a carrier exception, with the exception type per the playbook. Output: The orders needing attention with the exception type and the last tracking event, plain text.
2. 02For each, create or update the Orders Watch row (Exception, Status "Update drafted") and write the customer update in the playbook's voice, under 90 words: what we know, when we write again, the replacement promise. If the claim window has passed, write the carrier claim into Carrier Claim: tracking, value, evidence; else write when it opens. Never send, never file. Plain text. Output: The updates and claims drafted, one per order, plain text.

When a refund request arrives it reads the order and the payment from the storefront and the payment tool, applies the playbook's rules and the refund band (Auto under 50, Manager 50 to 500, Owner over 500), and drafts the reply. The refund itself is a person's click in the band.

1. 01Refund request on order {{order_number}} from {{customer}} <{{email}}>: {{amount}} because {{reason}}. Evidence: {{evidence}}. Read the order from the storefront and the payment from the payment tool: amount paid, shipped date, delivered date, earlier refunds. Apply the playbook's rule for the reason. Set the Band from the amount. Output: The order facts, the rule that applies, the band and the recommendation, plain text lines.
2. 02Write into Drafted Reply: "Band: <band> (<amount>)", the order facts in one line, "Recommend:" one line, then "Reply:" the customer reply in the store's voice under 80 words that never promises a refund not yet approved. Set Band. Leave Status Requested for a person in the band. Never refund, never send. Plain text. Output: The band, the recommendation and the drafted reply, plain text.

When a dispute is logged it reads the payment and the dispute from the payment tool, the order and messages from the storefront and the tracking from the shipping tool, and writes the evidence pack with a recommendation to contest or accept. A person submits it in the payment tool.

1. 01Dispute {{dispute}} in {{payment_tool}} on order {{order_number}} for {{amount}}, reason code {{reason_code}}, deadline {{deadline}}. Read the dispute and the payment, the order with its addresses and messages, and the tracking with the delivery scan. Check the Refund Requests table for an earlier refund on the same order. Output: The facts gathered, one line each, plain text.
2. 02Write the Evidence Pack in the playbook's order: order and amount, address match, tracking and delivery scan, messages with the customer, the policy shown at checkout, the listing at purchase. End with "Recommend: contest" or "Recommend: accept" and why. Set Status Packaged. Never submit. Plain text. Output: The evidence pack with the recommendation, plain text.

Every morning it reads on-hand quantities from the inventory tool and 30 days of sales from the storefront, computes days of cover per SKU, and for anything under the reorder point drafts the purchase order from the playbook's rules. A person approves and orders.

1. 01If a SKU is given ({{sku}}) plan that one only; otherwise every SKU. Read on hand, supplier and lead time from the inventory tool and 30 days of units sold from the storefront. Compute Sold Per Day and Days Of Cover. Update the Stock rows. Output: Each SKU with on hand, sold per day and days of cover, plain text lines.
2. 02For each SKU under cover per the playbook (days of cover below lead time plus 7, or on hand below the reorder point), set Status "PO drafted" and write the Draft PO: why, units (30 days of sales rounded to the pack size), supplier, last unit cost, total, last order reference. Everything else with enough cover: Status OK. Never place the order. Plain text. Output: The purchase orders drafted, one per SKU, plain text.

Every morning it reads new reviews from the reviews tool, drafts a reply per the playbook's voice for 3 stars and above, and flags 1 and 2 stars for a person with a suggested reply. A person posts.

1. 01If a review is given ({{review}}, {{product}}, {{customer}}, {{rating}} stars: {{text}}) work on that one only. Otherwise read new reviews since yesterday from the reviews tool and create a Reviews row for each. Output: The reviews to answer with rating and product, plain text.
2. 02For each review of 3 stars or more, draft the reply in the playbook's voice under 50 words and set Status Drafted. For 1 or 2 stars write "NEEDS A PERSON: <reason>" then "Suggested:" a reply, and set Status "Needs a person". Never post. Plain text. Output: The drafted replies and the flags, plain text.

Every Monday it reads customers and their order history from the storefront, finds anyone past 1.5 times their usual interval, picks the offer from the playbook, skips anyone with an open refund, dispute or complaint, and queues the approved list to the email tool behind approval.

1. 01Read customers and their order history from the storefront for the last {{days}} days (default 180). For each customer with 2 or more orders compute the usual interval and the days since the last order. List those past 1.5 times their interval. Skip anyone with an open Refund Request, Dispute or a review marked Needs a person. Output: The customers due a win-back with orders, interval, days since and the offer per the playbook, plain text.
2. 02Create or update a Win-backs row per customer (Status Suggested) with the Note: the numbers, the offer and why, any reason to hold. For rows already set to Approved by a person, add the customer to the win-back segment in the email tool and set Status Queued. Never send. Plain text. Output: The rows written and the customers queued, plain text.

At 17:30 it counts the day's orders, the late deliveries, the refunds waiting by band, the disputes due this week and the SKUs under cover, names the three things to fix tomorrow, writes the digest and emails it to the store owner.

1. 01Day {{day}} (default today). Count today's orders from the storefront. From the tables: Orders Watch rows with an exception and Status Watching or Update drafted; Refund Requests with Status Requested, by Band; Disputes with Status New or Packaged and a Deadline within 7 days; Stock rows with Status Under cover or PO drafted; Reviews marked Needs a person. Output: The numbers as plain text lines.
2. 02Write the digest: the numbers as short lines, then "Fix tomorrow:" three items each naming the row and the person or band. Create the Daily Digest row. Email it with Turtle Notify. Under 180 words, plain text. Output: The digest as sent, plain text.