A checkout reply policy template for DTC teams

By Rene Siller. Sources checked 2026-10-31.

A downloadable reply policy with filled subscription rows, four reply actions, a source conflict rule, ownership fields and failure tests.

## Quick answer

Write one row for each recurring question. Give it a precise source with an effective date, one reply action (answer, answer with a condition, ask or look up, or hand off), the words the reply must not use, a named owner and the order check before any further contact. Then test the row and record expected and actual behavior separately. Leave an observation blank until someone has run and recorded it.

Shopify documentation provides order-state context. The worksheet makes your own source, ownership and observed behavior explicit.

Read the primary source (https://help.shopify.com/en/manual/promoting-marketing/create-marketing/abandoned-checkouts) for the documented context and its scope.

## See four replies, then the row behind them

A shopper types: is this a one-time order or a subscription? The fluent reply is easy to write and easy to get wrong, because it can sound certain about things nobody has read. So before writing a reply, the policy has to say what is known, what is missing and who decides the rest.

Take a store selling a product called Monthly Refill, offered as one-time or subscription. Say its approved terms, effective 1 October, say two things:

The subscription option ships every 30 days.

The one-time option is a single order.

Those terms say nothing about cancelling, refunds or delivery dates. Here is what that allows.

Table: Four reply choices, with the facts that justify them
Situation | Say | Avoid, and why
Answer · “Is this one-time or a subscription?” The cart record shows subscription selected | “You have the subscription option selected, so this one is recurring. It ships every 30 days.” | “It’s a subscription, and you can cancel anytime!” · The option is right, but the terms contain no cancellation statement
Ask or look up · The same question, with no cart record | “It comes both ways, so it depends on what you picked. Did you choose one-time or subscription?” | “Yes, Monthly Refill renews every 30 days.” · It infers the selected option from the product name
Answer with a condition · “Will this ship again next month?” No cart record, and each option has a sourced answer | “If you choose the subscription, it ships every 30 days. If you choose one-time, it’s a single order.” | “Yes, your next box arrives on the 14th.” · It assumes the option and promises a date the terms do not contain
Hand off · “I was charged after cancelling, can I have a refund?” The terms are silent. The case was submitted to the billing queue, and the queue record confirms it was accepted | “I can’t confirm a refund from here. Your case is with our billing team for review.” | “No problem, I’ve refunded that for you.” · It commits to an outcome nobody has decided or sourced

Every tempting reply there sounds more helpful than the one beside it, and that is the danger. Reading well is not evidence. Notice too that the second and third situations both lack a cart record, yet they get different actions. The third can be answered because each branch has a sourced answer and the shopper can see which branch applies. The second asks about this cart itself, so it needs the fact first.

The fourth reply carries a different trap: it reports an action. “Your case is now with our billing team” is true only because the queue record shows the case was accepted. The wording cannot stand in for the action. If no owner can receive the case, say so and state the next step you can honestly give.

The other actions have stopping points too. Stop answering if the source has changed since its effective date, stop a conditional answer if you cannot confirm the condition, and stop asking if the record is missing and the shopper cannot supply the fact.

The second situation sits on one completed row in the policy. It records the reasoning as well as the reply, so a teammate can apply it to a question the row never anticipated.

Table: A worked subscription input. Replace the source and owner with your store’s actual records.
Field | Worked subscription example
Shopper’s question | Is this a one-time order or a recurring subscription?
Facts needed first | Which purchase option is selected in this cart
Source to check | The selected purchase option and the current subscription terms, effective 1 October
Reply action | Ask or look up until the option is known, then answer
Why this action | The product comes both ways, so its name cannot settle which option this shopper chose
Answer boundary | State only what the selected option and approved terms support
Must not say | “Cancel anytime” without the terms, a renewal date you have not read, or a refund outcome
If sources disagree | Stop answering and check the current authoritative record with its owner
Owner | Named subscription-policy owner in your team
Before the next contact | Recheck the current order and case status
Pass condition | Correct option, sourced terms, unresolved detail assigned, no duplicate recovery prompt

The same row and its blank copies are in the [reply policy sheet (CSV)](downloads/companion.csv).

Use the reply policy sheet to record the source, owner, expected reply and actual observation separately. Replace the example source and owner with your own records, then leave actual observations blank until tested.

## Fill the source before polishing the reply

Copy the shopper’s original wording into the worksheet. Add the precise source location and its effective date where available. Record the answer that source supports and the question it leaves open. Only then write the reply. Keep each row to one question. Intercom's guidance for its Fin agent gives the same advice for a different reason, saying each instruction should address a single objective and warning against complex and/or conditions.

Suppose the product page offers both one-time and subscription purchase options. A generic line about easy cancellation does not establish which option the shopper selected. The policy should ask for or retrieve the relevant record rather than infer it from the product name.

The word cancel changes the reply action because it can refer to different records. The selected option is what the shopper chose in the cart. Shopify Subscriptions creates a subscription contract after a subscription purchase. An individual order and that continuing subscription need their own checks. So when a shopper writes “I want to cancel,” a good first reply is: “Do you mean the next delivery, the subscription, or an order you’ve already placed?” Then use the current documentation and record for that object.

Add one rule to every row: if a reply and a page disagree, do not assume either is right. Stop answering, check the current authoritative record with its owner and correct whichever is wrong. Suppose the current terms say every 30 days and an older help page says something else. The owner confirms the terms govern, corrects the page and checks any reply that repeated it. Now reverse it: the terms are correct and a draft reply promises easy cancellation. The terms stand, and the reply or its rule gets fixed. If your checkout shows a subscription notice, list it among the row's sources so the reply is checked against what the shopper saw.

## Test the reply and its owner

Table: Test each distinct reply and ownership state
Case | Expected | Fail when
Source covers the question | Use the approved answer accurately. | The reply adds a term absent from the source.
Source covers only part | Answer the known part; preserve the unresolved detail. | A guess fills the gap.
Correct source, wrong reply | Keep the terms; fix the reply or rule with its owner. | Correct policy is rewritten to match the error.
Stale page, current record | Owner corrects the stale page and affected reply. | The conflict is ignored or authority goes unchecked.
Valid owner off shift | Keep the case in its agreed queue; give accurate next steps. | The case disappears or an unsourced deadline is promised.
Routing target missing | Use an honest fallback; record a failed or inconclusive handoff. | Human contact is promised as if the ordinary queue worked.
Order completed before next contact | Reconsider the next recovery contact. | A recovery prompt arrives after the purchase.

An off-shift team and a missing routing target are different cases. The fourth reply above only holds if the queue really accepted the case, so prove the destination exists for the after-hours check, then test the missing-target exception separately. Intercom documents different behavior when no human routing target is configured. A working ordinary queue cannot clear a missing route. For every case, record the expected action before the check and keep the actual message and destination beside it.

Run the owner check on your actual setup, since tools handle this differently. Intercom documents that with no human routing target, Fin does not offer escalation and the conversation ends with a get more help step, and it suggests telling customers your availability or next steps. If you use another tool, find its equivalent and see what the shopper is told. Gorgias also warns that a rule's auto-reply and its AI Agent can both respond to the same ticket, so read what actually arrived.

Record expected and actual behavior in separate columns. “Configured” is not a test result. Read the message and inspect where the case arrived. Keep the failed row open until someone fixes the condition and repeats the check.

## Copy the rows into your own policy

Use the reply policy sheet to record the governing input, expected action and actual observation separately. The worked rows establish the decision rule; fill the owner, dates and observed result from your own setup. The worked rows repeat the four situations above with your replacements still to make. The rows for your store carry no observations until someone runs them.

[Reply policy sheet (CSV)](downloads/companion.csv)

## What a completed worksheet looks like

A teammate who did not write the policy should be able to locate the source, explain the boundary and identify who owns the open issue. If that person needs the author to interpret a cell, make the cell more specific. Set a retest trigger rather than relying on a calendar alone: rerun a row whenever its source, its owner or the reply tool changes.

The reply policy CSV keeps the source, reply boundary, expected action and actual observation in separate fields. Use the worked rows to set the rule, then record your actual output, observed date, owner and next action. Keep only necessary customer context and use your approved access controls.

## Bring the decision to Zobi

Zobi’s homepage describes an EQ-model approach trained on commercial outcomes. For a DTC team, a useful evaluation starts with a real buying question, the approved information that should answer it and the point at which a person should take over.

Bring your completed decision sheet to the discussion. Inspect the proposed reply, information access and handoff against the conditions in this guide. Agree what would be tested in your store before drawing a revenue conclusion.

Explore Zobi for DTC (https://zobi.com/dtc/) with the shopper question, approved source and exception you want the reply route to handle.

## Sources and method

The named platform facts come from the linked documentation, checked on 31 October 2026, and apply to those documented workflows. Fit judgments, ownership rules and the proposed tests are Zobi editorial analysis. Worked inputs illustrate the decisions; record your store’s observed results separately in the working sheet.

Sources: Shopify subscription contracts (https://help.shopify.com/en/manual/products/purchase-options/subscriptions/shopify-subscriptions/manage-subscriptions/manage-contracts), Intercom guidance (https://www.intercom.com/help/en/articles/10560969-fin-guidance-best-practices), Intercom routing behavior (https://www.intercom.com/help/en/articles/15234261-configure-fin-s-behavior-when-no-human-agents-are-available), Gorgias rules and AI Agent (https://docs.gorgias.com/en-US/use-rules-with-ai-agent-5415482). Brand positioning: Zobi (https://zobi.com/).

## Publisher disclosure

Published by Zobi, which offers checkout-conversation services.

Zobi · zobi.com