Systems and records
Retail delivery promise dataset: evidence on inventory, cutoffs and location
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
A retail delivery promise dataset should capture each promise as the customer saw it, the sellable inventory at the shipping location at that instant, the routing rule that picked the location, the cutoff, handling and transit settings with time zones and holidays, and the destination. Shipment scans, delivery scans and delay notices then show whether the promise held.
Checklist
Retail delivery promise
Key takeaways
- A delivery promise rests on three systems: sellable inventory, a routing decision and shipping settings; store each one, not only the date.
- Use the available quantity at the chosen location, read at the promise instant, never total stock on hand.
- Cutoffs, calendars and transit ranges are per location and change over time; version them.
- Grade the agent on a recomputed promise, then confirm the outcome with carrier scans and delay notices.
- Pair tasks that differ in a single inventory or routing fact so an agent must read the records.
What evidence should a delivery promise task include about inventory, cutoffs and location?
It should include the promise exactly as the customer saw it, the inventory state at the location that would ship the order at that instant, the routing rule that picked that location, the cutoff, handling and transit settings in force with their time zones and holiday calendars, and the destination. Then it needs the outcome: the shipment scan, the delivery scan, and any delay notice, so the task can show whether the promise held.
A delivery promise is a multi-source claim. A product page can say arrives by Thursday because three separate systems agreed: the inventory system said a unit was sellable, the routing logic chose a node that could ship it, and the shipping settings turned the order time into a date. If a task records only the date that was shown, an agent can learn to produce plausible dates without checking any of the facts behind them.
This page sets out a multi-source promise verification task packet that retail and ecommerce teams can fill from their own systems, with a worked case, a counterexample and the checks that prove the outcome rather than a status field.
How is a delivery date promise actually calculated?
Delivery estimates commonly add handling time and transit time, counted in business days, to the moment the order starts processing, which depends on whether it arrived before the cutoff. Google Merchant Center describes its estimate as maximum handling days plus maximum transit days, based on business days, taking the order cutoff into account, and says it adds a day when handling or transit overlaps a holiday.
Schema.org gives the same parts public definitions. Its cutoffTime property is the time after which orders received that day are no longer processed, with one day added to the estimate for later orders. Its handlingTime is the typical delay between receipt of the order and the goods leaving the warehouse, by convention in business days, and its transitTime is the typical delay between dispatch and the goods reaching the customer. Using these terms in the packet keeps fields unambiguous across teams.
Each component has its own record, and each can be wrong in its own way, so the packet stores them separately rather than storing only the computed date.
| Component | What it means | Record to capture | Typical failure in history |
|---|---|---|---|
| Order time | The instant the order was placed, with offset | Checkout event | Stored in UTC without the store's local offset |
| Cutoff | Time after which processing moves to the next business day | Shipping settings version per location | Location-specific cutoffs collapsed into one store value |
| Handling time | Business days from order to goods leaving the location | Shipping settings, product overrides | Product-level overrides missing from exports |
| Transit time | Business days from dispatch to delivery | Carrier service and zone table | Range stored as a single value |
| Business day calendar | Which days count for handling and transit | Warehouse and carrier calendars | Holidays added after the fact |
| Promise shown | The date or range the customer actually saw | Page or checkout render log | Only the order confirmation email survives |
Which inventory quantity can honestly back a promise?
Only the quantity that was sellable at the promising location at the instant of the promise. Total stock on hand is not the right number, because some of it is already spoken for or cannot be sold.
Platforms make this distinction explicit. Shopify's help center defines on hand as the sum of committed, unavailable and available inventory, describes committed units as set aside for things like unfulfilled orders, and says incoming inventory is not available to sell until it has been received. Adobe Commerce's inventory documentation describes a salable quantity that aggregates the quantities of a stock's sources and then deducts reservations, which hold inventory from order placement until the order is shipped or canceled. WooCommerce exposes a product stock status that can be in stock, out of stock or on backorder, and a backorders setting that controls whether orders are accepted without stock.
For the packet, store the state the promise engine used, the timestamp of that reading and the location it belongs to. A promise made against stock that was on backorder is a different task from one made against available stock, and the correct agent behavior differs: the first may call for a later date or a disclosure, the second for the standard calculation.
How does the fulfilling location change the promise?
The location decides the cutoff, the handling calendar and the transit zone, so the same item can carry different promises depending on which node ships it. A task has to record which locations were candidates and why one was chosen.
Shopify's order routing documentation shows how many conditions are involved. Routing applies a series of rules and then prioritizes locations; the ship from closest location rule uses straight-line distance, not driving distance; a location without online order fulfillment turned on is excluded; when two locations share the same address, the older one is prioritized; and if the customer's address cannot be matched to coordinates, the closest location rule is skipped and the oldest location is used. Each of these can move an order to a node with a different cutoff.
Store the routing rule set version, the candidate locations with their available quantities at the instant, and the chosen location. Without that, a task cannot show whether a missed promise came from the shipping settings or from a routing decision.
What should the multi-source promise verification task packet include?
Each task is one promise shown for one order line, with every record needed to recompute it and confirm what happened. The field groups below can be used as the packet schema as is; they draw on the order management system, inventory and warehouse systems, shipping settings and carrier tracking.
Keep the promise and the outcome in separate groups and never expose the outcome fields in the starting state given to an agent. A task that leaks the delivery scan teaches nothing about promising.
| Field group | Fields | Source record | What it proves |
|---|---|---|---|
| Task header | Task id, packet version, promise type (product page, cart, checkout, service reply), label status | Packet build log | Which promise is being tested and whether the case is real or illustrative |
| Order line | Order and line ids, item and variant, quantity, order instant with offset, delivery method | Order management or ecommerce platform | What was promised and when |
| Destination | Postal code or region, country, address match status | Order address, geocoding result | Transit zone and whether location rules could run |
| Inventory at promise | Per candidate location: available, committed, incoming, backorder flag, reading timestamp | Inventory system snapshot | Whether sellable stock backed the promise |
| Routing decision | Rule set version, candidate locations, chosen location, reason | Order routing log or allocation record | Which node's settings apply |
| Shipping settings | Cutoff, handling range, transit range, business day calendars, holiday list, versions | Shipping configuration history, carrier service | The inputs needed to recompute the date |
| Promise shown | Date or range displayed, render instant, page or channel | Render log, confirmation message | What the customer was told |
| Outcome | Ship scan, delivery scan, split shipments, delay notice and customer response, cancellation or refund | Warehouse, carrier tracking, helpdesk | Whether the promise held, and how a miss was handled |
Worked case (illustrative): the stock is in the other warehouse
This case is illustrative and does not describe a real retailer. Times are stated with time zones and durations in business days; no money amounts are involved. It follows the packet format: starting state, linked evidence, decision and verified outcome.
- Starting state: a customer in the Eastern time zone orders one item at 14:40 Eastern on a Tuesday; the product page showed a promise range.
- Inventory at promise: the east warehouse shows zero available and two committed units; the west warehouse shows four available units.
- Routing: the closest location rule would choose the east warehouse, but it has no sellable stock, so the order routes to the west warehouse.
- Shipping settings: the west warehouse cutoff is 13:00 Pacific, handling 0 to 1 business days, transit 3 to 5 business days to the destination zone.
- Decision: 14:40 Eastern is 11:40 Pacific, inside the west cutoff, so processing starts Tuesday; the latest date is six business days later using maximum handling plus maximum transit.
- Outcome check: ship scan from the west warehouse on Wednesday, delivery scan within the range, no delay notice; the promise is labeled held.
- What a weak agent does: computes the date from the east warehouse settings because it is closest, which yields a promise the inventory could not support.
- Counterexample: the same order at the same instant, but the east warehouse has one available unit. Routing chooses the east warehouse, the order is inside its 15:00 Eastern cutoff, and with that warehouse's shorter transit range to this zone (illustrative: 1 to 2 business days) the correct promise is several business days earlier. The two tasks share every field except one inventory reading, so an agent that ignores inventory by location will get one of them wrong.
How do you verify whether the promise held?
Recompute the promise from the stored inputs, compare it with what was shown, then compare both with the carrier's delivery scan. Grading an agent on the recomputed promise, not on the date it happens to output, separates a correct calculation from a lucky one.
At a glance
How do you verify whether the promise held
The law gives the promise a compliance dimension in the United States. The FTC's Mail, Internet, or Telephone Order Merchandise Rule requires a seller to have a reasonable basis for any shipping time it states, or for shipping within 30 days when it states none, and to offer the buyer the option to consent to a delay or cancel for a prompt refund when it cannot ship on time. A packet that captures delay notices and customer responses lets an agent be tested on what to do after a missed promise, not only on the date. How the rule applies to a given business is a question for counsel.
- Inputs check: the agent used the chosen location's cutoff, calendars and ranges as of the order instant.
- Inventory check: the promise was backed by available stock at that location, or the backorder case was disclosed.
- Date check: the agent's date or range matches the recomputed one.
- Outcome check: delivery scan inside the promised range, or a delay notice issued before the promised ship date.
- Recovery check: for missed promises, the agent offers the options the business's policy and applicable law require.
Which records quietly corrupt a delivery promise packet?
Two common sources of damage are snapshots taken at the wrong time and settings exported as they are today. Each item below produces tasks whose correct answer cannot be recomputed from the packet.
Drop a task when an input cannot be reconstructed rather than filling it with a current value. A smaller packet with complete inputs measures more than a large one built on guesses.
- Inventory read after the order: the snapshot already includes the reservation the order created.
- Current settings applied to old orders: cutoffs and handling ranges change seasonally.
- Missing holiday calendars: business-day counts drift by one or two days around holidays.
- Time zone loss: order, cutoff and scan times stored without offsets.
- Manual promises: service staff promising dates outside the engine, recorded only in free text.
- Split shipments treated as one: each line can carry its own location and promise.
Delivery records, customer privacy and the right to license them
Delivery packets carry addresses, names, phone numbers and sometimes delivery photos, and the promise check needs none of them at full precision. Reduce addresses to the zone or postal region the transit table uses, tokenize customer and order identifiers consistently across systems, and remove free-text delivery instructions that often contain personal details.
Rights are a separate question. Carrier tracking data, marketplace orders and third-party logistics records may be governed by contracts that limit reuse, and de-identification alone does not grant a right to license them. Counsel reviews which sources the business may include.
Licensing delivery promise histories with SourceX
Retailers that kept inventory snapshots, routing logs, shipping settings history and carrier scans side by side hold the inputs this packet needs. Licensing selected operational records like these to AI developers is work SourceX handles end to end: sourcing and qualification, rights review, contracting, data preparation, delivery and payment.
A first conversation needs no data, only a short company assessment. Promise records leave the company only after it has signed off on the scope and terms, and from then on SourceX's handling of the de-identified data, including any sublicense or onward distribution, is whatever the contract permits. Large deliveries can stay in the company's own storage or ship on encrypted drives, since SourceX does not host multi-terabyte datasets.
Frequently asked questions
What is a retail delivery promise dataset?
It is a set of delivery promises shown to customers, each linked to the inventory, routing decision, cutoff, handling and transit settings that produced it, and to the shipment and delivery scans that show whether it held. It is used to train or evaluate agents that quote or check delivery dates.
Why is on-hand inventory the wrong number for a promise?
On-hand stock includes units already committed to other orders or set aside as unavailable. A promise should rest on the sellable quantity at the location that will ship, read at the instant of the promise.
How does an order cutoff change the delivery date?
An order placed after a location's cutoff starts processing the next business day, which pushes the estimate by at least one day. Because cutoffs are set per location and time zone, the same order can fall on either side depending on where it ships from.
Is a delivered status enough to label a promise as kept?
No. The label should compare the carrier's delivery scan with the promise the customer saw, and record split shipments and delay notices. A delivered status alone says nothing about whether delivery happened inside the promised range.
Which systems hold the records for a promise packet?
Typically the ecommerce platform or order management system for orders and promises, the inventory or warehouse system for stock by location, the shipping configuration for cutoffs and handling, carrier tracking for scans, and the helpdesk for delay notices and customer responses.
Sources
- Google Merchant Center calculates estimated delivery as maximum handling days plus maximum transit days in business days, accounting for the order cutoff, and adds a day for holidays. Source
- Schema.org defines cutoffTime as the time after which orders are not processed that day, adding one day to the estimate, and defines handlingTime and transitTime as the two delays in a delivery estimate. Source
- Shopify defines on hand inventory as committed plus unavailable plus available, and says incoming inventory is not available to sell until received. Source
- Adobe Commerce calculates salable quantity from a stock's sources less reservations, which hold inventory from order placement until shipment or cancellation. Source
- The WooCommerce REST API documents product stock status values in stock, out of stock and on backorder, and a backorders setting. Source
- Shopify order routing applies rules to prioritize locations, uses straight-line distance for the closest location rule, excludes locations without online fulfillment turned on, prioritizes the older location when two share an address, and uses the oldest location when the customer address cannot be matched. Source
- The FTC's Mail, Internet, or Telephone Order Merchandise Rule requires a reasonable basis for stated shipping times, or for shipping within 30 days if none is stated, and a delay option to cancel for a prompt refund. Source
See if your company qualifies
A short company assessment. No data uploads are needed.