Shopify and BigCommerce stores: deflect the order-status flood, survive Black Friday without temp hires

The question shapes that dominate ecommerce support

Public benchmarks across Shopify, BigCommerce, and WooCommerce stores line up on a handful of question types that show up over and over:

• Order status (“where's my order?”) runs 35-45% of all inbound. Tracking links break, carriers update late, customers shop from new addresses, all of it generates the same ticket.
• Returns and refunds (“can I return this?”, “how long does a refund take?”) is the second tier at 15-25%. The answer is in your policy page, but customers won't go find it.
• Product fit (“does this come in tall?”, “is the strap leather or vegan?”) is 10-15% for apparel and home goods. The answer is in the PDP, but the visitor wants confirmation from a human.
• Shipping (“do you ship to Australia?”, “what's express?”) is 8-12%. Live rates change, so static FAQ pages always trail reality.
• Everything else, including discount-code troubles, duplicate orders, address changes, and the true edge cases that deserve a human, is the remaining 10-15%.

The first four buckets, accounting for the bulk of inbound, are deflectable by an AI that can answer with citations and look up an order. The fifth needs a human. OpenAgent draws the line for you.

How OpenAgent handles each bucket

Three pieces of configuration cover most of the question shapes above. They take about thirty minutes to set up, total.

Knowledge base: paste, upload, done

Drop your returns page, shipping page, sizing guide, and care instructions into the knowledge document upload. OpenAgent chunks them, embeds them with your provider, and stores the vectors in pgvector. The AI now answers “can I return this?” by quoting the policy with a citation, not by hallucinating.

Order-status tool: scoped to the visitor's email

Wire the visitor's email (captured by the widget's pre-chat form) to a workflow function that calls GET /admin/api/2024-10/orders.json?email=... on the Shopify Admin API. Whitelist the tool on the agent. The AI now answers “where's my order?” with a real tracking link and the last known carrier status. No webhook setup, no inventory of orders kept on the OpenAgent side.

Escalation policy: tier-2 to a human in under 60 seconds

For the long-tail 10-15% of conversations the AI can't solve, the escalation policy hands off to whoever is online, with the conversation transcript and the order lookup result already attached. The human starts at the answer, not at the question.

Sample math: a 10,000-ticket-per-month store

These numbers are pre-loaded into the calculator below for a typical mid-market store. Override any input and the result updates live.

Monthly human-handling cost saved: 10,000 × 55% × $7 = $38,500.

Monthly LLM bill on the deflected portion: 5,500 × $0.0011 = $6.05.

Net monthly saving: $38,500 − $6 − $3 ≈ $38,491. Annual: ~$461k.

Even if you halve every assumption (5,000 conversations, 30% deflection, $5/ticket), you land at about $48k/year savednet of cost. The math holds at almost every scale.

The first 90 days

What we tell stores to expect on the way to the steady-state numbers above:

• Week 1: upload returns page + shipping page + sizing guide. Add the official theme snippet from the install guide (it also links logged-in customers to their conversations automatically). Configure the order-lookup function with your Shopify Admin API token. Containment lands at ~30% with these three docs alone.
• Week 2-4: review the conversations the AI escalated. Add the top 5 missing answers as KB documents. Containment climbs to ~45%.
• Week 5-12: tag closed conversations by reason (returns, shipping, product, technical). The auto-tag report shows which question type still leaks to humans most often. Patch the KB there. Containment plateaus at ~55%.
• Steady state (90+ days): 55% containment, 1.4x throughput per human agent, and a 3-line conversation summary attached to every escalation so the human starts at the answer.

Why this matters for Black Friday

Cyber weekend produces 3-5x the normal ticket volume on most stores. The conventional fix is to hire seasonal contractors and put them on a crash-course of your support workflows. That's expensive (~$20k for a small batch of contractors), slow (training takes longer than the season), and quality is uneven.

OpenAgent has no concurrency cap. The AI handles the spike at the same deflection rate as the rest of the year; your humans focus on the escalations that matter. Most stores skip the seasonal hire entirely after their first Black Friday on the platform.

What to watch in your own numbers

The metrics that move first when this is working:

• Containment rate (% of conversations closed without a human). Industry target: 50-65% for ecommerce.
• First-response time on escalated conversations. Should drop from minutes to seconds because the AI already gathered context before handing off.
• CSAT on AI-only conversations. Should sit within 5% of the human-handled CSAT. If it drops further, KB has gaps.
• Tickets per active customer. Self-service deflection shows up here as a quiet reduction over the first 90 days.