Ecommerce support teams handle large volumes of order, returns, refund, and shipping queries — often with tight SLAs and direct revenue implications when AI responses are inaccurate.
FortiVault and FortiAgent give ecommerce support teams live order and fulfilment data in every AI response, per-category governance gating that separates high-risk refund automation from low-risk shipping FAQ automation, and a full audit trail for every AI decision.
The Ecommerce Support Problem
Ecommerce AI support is often measured by ticket deflection — how many queries the AI handles without a human. At high volume, even a small percentage of inaccurate responses translates to a large number of customers receiving wrong information about their orders, refunds, or returns.
The risk is not uniform across query types. An AI that is highly accurate on order status queries may perform significantly worse on refund policy interpretation or returns eligibility. Without category-level accuracy measurement, you cannot tell which categories are ready to automate and which are not.
FortiVault's per-category Trust Score and Automation Gating mean order status can be automated independently of refund queries. Each category earns the right to automate based on its own measured accuracy — not a blended overall resolution rate.
Ecommerce Query Types
FortiVault's per-category gating means each query type has its own automation policy. Low-risk, high-volume queries can be automated earlier. High-stakes refund and billing queries are gated independently at a higher accuracy threshold.
Order status queries are high volume and relatively low risk. FortiAgent retrieves live tracking data from Shopify, DHL, and FedEx and responds with current fulfilment status. This category can be automated earlier in the governance progression as accuracy is typically high on factual order data.
Returns queries require accurate policy retrieval and sometimes connector calls to initiate the returns process. FortiVault gates automation based on measured accuracy in this category — ensuring FortiAgent's policy interpretations are correct before running without human oversight.
Refund decisions have direct financial consequences. FortiVault gates refund automation at a higher accuracy threshold and logs every refund-related AI decision with the full data trace — which Stripe data was retrieved, what FortiAgent proposed, what was approved.
Product queries are informational and low-risk. FortiAgent responds from your configured product knowledge base. Live inventory data can be retrieved via Shopify connector for availability queries. This category typically reaches automation threshold earliest.
Delivery failure, damaged goods, and address correction queries require live carrier data. FortiAgent retrieves current status from DHL, FedEx, or your logistics connector and provides grounded responses based on actual fulfilment state — not assumptions from training data.
Live Connector Data
FortiAgent retrieves live data from your connected systems before every response. Order status comes from Shopify — not a cached knowledge base entry. Tracking information comes from DHL or FedEx — not an estimate from training data. Refund eligibility comes from your configured Stripe rules.
Every connector call is logged as part of the audit trail. For any AI response, you can see which system was queried, what data was returned, and how that data shaped the response — before and after any human review.
See all FortiAgent connectorsShopify
Live order status, fulfilment, and inventory data
Stripe
Payment status, refund eligibility, billing history
DHL
Live shipment tracking and delivery status
FedEx
Carrier tracking and fulfilment updates
Royal Mail
UK domestic fulfilment tracking
Custom REST APIs
Any internal system via registered API
The Governance Layer
FortiAgent handles execution — retrieving live data, generating grounded responses, initiating write-back actions like returns or refund requests. FortiVault governs what gets sent to the customer.
Every FortiAgent response passes through FortiVault's evaluation before reaching the customer. The Trust Score for that category determines whether the response is sent automatically or held for human review. Every decision — auto-sent or reviewed — is logged.
Governance architecture deep diveSeparate gating per query type
Order status automation can be enabled before refund automation. Each category has its own Trust Score and threshold — they are not coupled.
Automatic review when accuracy drops
If FortiAgent's accuracy in a category drops below the threshold — due to stale knowledge, connector data changes, or new query patterns — FortiVault automatically requires human review. No manual intervention needed.
Write-back with governance
FortiAgent can initiate returns, process refund requests, or update order records through connected systems. Every write-back action is logged and can be gated by the same governance controls as responses.
Human review queue
When a response requires review, FortiAgent drafts it and holds it for agent approval. The agent approves, edits, or rejects. The outcome is recorded either way.
How It Works
Every step is grounded in live data and evaluated by FortiVault before the customer sees anything.
Order status, return request, refund question, or shipping issue — via the FortiAgent widget or connected channel.
Calls Shopify for order and fulfilment status, Stripe for payment and refund eligibility, DHL or FedEx for tracking — whichever connectors the query requires.
Response is built from live connector data and your configured knowledge base. No training data used — only what your systems returned.
Trust Score checked for this query category. If accuracy meets the threshold, response is sent. If not, it enters the human review queue.
Order status may auto-send; refund decisions may require agent approval. Each category has its own gate — they are not coupled.
Every connector call, knowledge source, rule applied, and outcome recorded in the audit trail. Traceable per response.
Common Questions
Yes — this is exactly what per-category gating is designed for. Order status queries typically reach a reliable accuracy threshold earlier than refund eligibility queries. FortiVault's Automation Gating operates independently per category, so order status automation can be enabled while refund queries remain in human review. They are not coupled.
FortiAgent calls the Shopify API directly before responding — retrieving the current fulfilment status, tracking number, and carrier update for that specific order. It does not answer from knowledge base content or training data for live data queries. The Shopify connector call is logged in the FortiVault audit trail for every response.
The FortiVault audit trail logs every refund-related AI decision at the decision level: which Stripe data was retrieved, what refund eligibility check was run, what FortiAgent proposed, and whether a human approved or overrode it before sending. This gives you a complete traceable record of what the AI decided and why.
Yes — FortiAgent supports write-back procedures for returns initiation in Shopify and refund initiation in Stripe. These actions are gated by the same governance controls as any other response: they can only be taken if the category's accuracy threshold is met, and every write-back action is logged in the audit trail.
Live Shopify, Stripe, and fulfilment data. Per-category automation gating. Full audit trail from query to resolution.