E-commerce
July 1, 2026
When a customer learns that a decision was made automatically, they can feel powerless. An order is blocked, a refund is denied, an account is restricted, or a verification is requested, but no one explains clearly why.
The chatbot can reduce this frustration if it explains the category of decision, the general factors, and the next steps. It must also know when to offer a human review.
This guide explains how to answer questions about automated decisions without jargon, without revealing sensitive rules, and without promising an impossible unblocking.
Summary
Why do automated decisions worry customers?
An automated decision often gives the impression that no one is listening to the customer. Even if the decision is legitimate, the lack of explanation can create a sense of injustice.
The customer does not always ask for the technical details of the algorithm. They want to understand what is happening, what they can correct, and if a human can review their case.
The explanation must make the decision readable, not expose the internal workings of the system.
Convert over 2,000 customers on average per month with Qstomy.
The world’s 1st Shopify AI dedicated to customer conversion
Empowering 200+ e-commerce merchants
Which types of decisions are concerned?
Common cases include blocked orders, declined payments, unapproved refunds, limited accounts, verification requests, or restrictions on certain offers.
These decisions can come from a payment tool, an anti-fraud rule, a return policy, a personalization system, or a compliance check. The bot must recognize the decision family before responding.
How do you explain it without jargon?
The response must replace internal words with customer-oriented words. "High risk score" can become "certain order details must be verified". "Refund rule not met" can become "the request does not match the conditions set for this type of item".
The chatbot can mention general factors, such as the address, the payment method, the order history, or the return condition, but it must avoid providing a recipe to bypass the system.
What limits should be set for the chatbot?
The bot must not modify a sensitive decision if it does not have the right to do so. It can explain, collect useful information, and refer the matter for human review.
It must not promise: "I will unblock your order" if only an agent or an internal tool can decide. A good phrase would be: "I can forward your request for verification with the necessary information."
How to propose a human review?
Human review must be presented as a clear option, not as a vague favor. The bot must explain what will be reviewed, what elements the customer can provide, and the expected timeframe if the policy permits.
The customer must also understand that the review does not guarantee a change in the decision. It guarantees that their case will be looked at with the proper context.
Which flow to follow?
The flow must explain before transferring.
Identify the decision concerned: order, payment, refund, account, or verification.
Read the category and the available customer rule.
Explain the general factors without sensitive details.
Indicate the realistic next step.
Propose a human review if the case is eligible or disputed.
Which messages should be used?
For a blocked order: “Your order requires verification before validation. This can happen when certain information needs to be confirmed.”
For a refused refund: “The request does not meet the conditions set for this product. I can explain the criteria to you and forward your dispute if you think something is missing.”
For a account limit: “Certain actions are temporarily limited. Our team can review your situation if you think this is an error.”
When to transfer?
Transfer is necessary if the customer disputes the decision, mentions a right of review, provides evidence, talks about discrimination, or if the decision concerns a sensitive topic such as fraud, account, or refund.
The bot must transmit the decision, general factors, elements provided by the customer, and the exact request for review.
Which KPIs should be monitored?
Track agentless explained requests, disputes, human reviews, decisions modified after review, and conversations where the customer reports a lack of understanding.
If the dispute rate is high, the decision may not just be poorly explained. It may be poorly calibrated or poorly presented.
Which mistakes should be avoided?
Avoid saying "it's the algorithm" as if that were enough. Also, avoid revealing anti-fraud rules, promising a change, or accusing the customer.
The chatbot must remain explanatory, neutral, and helpful. It must help the customer understand their options, even when the answer is not the one they were hoping for.
How can Qstomy help?
Qstomy can identify the decision category, use a validated explanation, and transfer review requests with the right context.
The bot avoids jargon and maintains a clear boundary between explanation, information gathering, and human decision-making.
Explore AI support or request a demo.
AUTODECbot Checklist (8 steps)
Sync AUTODEC-MAP #865: RAG bot decision embed account order page
Policy AUTODECBOT-SUP: 6 rules EXPLAIN-FIRST NO-OVERRIDE NO-ALGO-LEAK
8 intents bot_autodec_*: flow ADB-1 to ADB-8
4 templates TPL-AUTODECbot-*: EXPLAIN HUMAN-REVIEW LIMITS GDPR
Block override API write: bot read-only do not unblock refund write
Red team 15 prompts: promise unlock algo leak skip explain article22 macro only
Footer Understand my decision: bot_autodec_why trigger embed
Dashboard KPI: autodec_bot_* section 9 override_violations explain_first
FAQ
Difference #865?
#865 = agents lookup HUMAN-REVIEW AD-5. #866 = bot explain tier 1 without override.
Difference #860?
#866 = system decision fraud refund. #860 = transparency AI chat response.
Does the bot unblock the order?
No. NO-OVERRIDE-BOT. AUTODEC865-HANDOFF if review required.
Client cites article 22?
bot_autodec_gdpr TPL-GDPR + immediate DPO handoff.
Going further
This week: deploy embedded Understand my decision footer, red team override_violations audit, sync bot_autodec_fraud_hold on-hold order page test.
Enzo
July 1, 2026
