How to Auto-Resolve "Where Is My Order?" Inquiries With AI

To put a number on it: an industry survey cited by the Fini WISMO platform guide pegs WISMO handling at roughly $1M a year in support spend for a brand doing ~50,000 orders annually, and the same class of order-tracking automation is reported to deflect 60–70% of those contacts. Fin's team puts the per-ticket economics at $4.50 down to under $1 once an agent handles the lookup (Fin AI); strategic write-ups go further, citing 70–90% reductions in WISMO volume (Thunai). Whatever your exact mix, this is the single highest-ROI ticket type to automate.

This is also the single best candidate for end-to-end AI automation in your whole helpdesk, because the work is lookup-and-relay — not judgment. This tutorial shows you how to build a Macha agent that does exactly that: it watches for order-status tickets, pulls the live order from Shopify, replies on the ticket with the carrier and tracking, marks it solved, and — critically — knows when to back off and hand a human a clean ticket instead of guessing.

New to WISMO automation? If you want the strategy-level why and a Zendesk-AI-agent overview first, start with our companion piece, Automate WISMO responses with AI in Zendesk. This post is the hands-on build — the exact trigger, tools, and escalation rules, step by step.

A note on what Macha is, because it changes the design: Macha is an AI agent layer that sits on top of the helpdesk you already run — Zendesk, Freshdesk, Gorgias, or Front — connected to your commerce stack. It is not a chat widget bolted onto your storefront, and it is not a helpdesk replacement. That distinction matters here. Most "WISMO automation" tools only deflect questions asked in a live-chat bubble. The agent below resolves the email and helpdesk tickets your customers actually open, on the same ticket, in your brand voice — then closes it.

What this agent does, end to end

Here is the flow we're building, taken straight from the order-status use case:

1. Trigger — A new ticket arrives in your helpdesk and looks like an order-status question.
2. Get Ticket — The agent reads the ticket: the message, the requester's email, any order number they gave.
3. Get Order (Shopify) — It looks up the order in Shopify by number or by the customer's email, and reads fulfillment and tracking.
4. Add Comment — It writes a reply on the ticket with the carrier, tracking number, and ETA, in your tone.
5. Update Ticket — If it had everything it needed, it sets the ticket to Solved. If it didn't, it escalates instead.

A real exchange, end to end, looks like this:

Customer: "Hi, I placed an order 3 days ago and haven't received any updates. Order #1847." (Macha looks up order #1847 in Shopify → order shipped via FedEx, tracking #7829401, estimated delivery tomorrow.) Macha: "Hi! Your order #1847 shipped on March 2nd via FedEx. Your tracking number is 7829401 and it's estimated to arrive tomorrow. Let me know if you need anything else!" (Ticket → Solved.)

No agent touched it. That's one of your highest-volume ticket types handled in seconds, around the clock, in any timezone your store sells into.

The building blocks

You need two connectors and one agent.

We'll use Zendesk + Shopify as the worked example because that's the pairing in the source use case, but the build is identical on Freshdesk, and on Gorgias — which is Shopify-native and a natural fit for commerce teams — the agent and tools map the same way. If you sell on Shopify and run any of these helpdesks, you can build this today.

Connect Shopify and your helpdesk

In Macha, connect both: your helpdesk (so the agent can read tickets and post replies) and Shopify (so it can look up orders). The Shopify Get Order action looks up an order by ID or order number and returns full line items, fulfillment, and payment status — and from that payload the agent reads exactly the four fields a WISMO reply is built on: fulfillment status (is it shipped?), carrier (FedEx, UPS, USPS…), tracking number, and estimated delivery / ETA. That's the whole reply, and nothing the agent doesn't need. See the Shopify integration and Zendesk integration pages for the full tool lists, and the docs for connector setup.

Building the agent, step by step

Step 1 — Create the agent and give it a tight job

Create a new agent and name it something honest like "Order Status (WISMO)." The narrower an agent's job, the better it performs — a single-purpose order-status agent will beat a do-everything mega-agent every time, because its instructions, tools, and escalation rules are all pointed at one outcome.

Write its instructions in plain language. The job description does most of the work:

You are the order-status agent for [Store]. When a customer asks where their order is, find their order in Shopify — by order number if they gave one, otherwise by their email — and reply with the fulfillment status, carrier, tracking number, and estimated delivery. Be warm and brief. If you cannot confidently identify the order, or the order has shipping trouble, do not guess — escalate to a human with a short note. Never invent a tracking number or a date.

That last line is not optional. The fastest way to destroy trust is a confident wrong tracking number, so the instruction to never fabricate and to escalate on doubt is the backbone of a safe agent.

Step 2 — Give it the tools it needs (and only those)

Add exactly the four actions the use case calls for, in this order of use:

1. Zendesk → Get Ticket — read the incoming ticket and requester.
2. Shopify → Get Order — look up the order by number or customer email, with fulfillment and tracking.
3. Zendesk → Add Comment — post the reply on the ticket.
4. Zendesk → Update Ticket — set status to Solved (or escalate).

Two of these — Add Comment (a public reply) and Update Ticket (status change) — are high-impact tools that actually write back to the customer and the queue. Macha surfaces an activation confirmation and a persistent advisory card on those, so you enable them deliberately. Keep the toolset this tight: an agent with only the tools its job requires can't wander off and do something you didn't intend.

Step 3 — Wire the trigger

The trigger is what makes this autonomous instead of a button you press. In the agent's Triggers tab, add a trigger, pick Zendesk → Ticket Created as the event, and then fill the condition field with plain English: "Customer asks about order status." This isn't a rules engine of subject CONTAINS "order" clauses — the condition is a natural-language description of when this agent should wake up, and Macha evaluates each new ticket against it semantically. So "where's my package," "hasn't arrived yet," and "any update on #1847?" all match, while a refund request or a product question does not. You can tune it later by tightening the wording (e.g. "Customer is asking about the shipping or delivery status of an existing order, not requesting a return or cancellation") to narrow what fires.

The result: every new ticket that reads like a WISMO question wakes the agent automatically — no human triage in front of it.

Two trigger settings are worth setting on purpose:

• Debounce. Customers often fire off a ticket and then immediately add "oh, order #1847" in a second message. A short debounce (a few seconds) lets the agent wait for the dust to settle and read the complete thread, instead of replying to half a question. Zendesk and Freshdesk ticket events both support a debounce delay.
• Scope it to new tickets only. You want this on freshly created order-status tickets, not retriggering on every internal update to an existing one.

Step 4 — Test before you let it loose

Don't point a brand-new agent at your live queue on faith. Run it against a handful of real past order-status tickets first and read its replies the way a customer would. Check three things: did it find the right order, is the tracking number actually correct, and did it escalate the messy ones instead of bluffing? When the replies are ones you'd happily send yourself, turn the trigger on. You can pause an agent at any time — toggling it off sets its triggers inactive rather than tearing anything down, so you can flip it back on the moment you're comfortable.

The part competitors skip: when NOT to auto-resolve

Every WISMO automation vendor will show you the happy path — order shipped, tracking exists, customer delighted. The real world has a tail of cases where auto-resolving is the wrong move, and a good agent is defined by how it handles them. Build these escalation rules in explicitly:

• The order isn't found. Wrong order number, guest checkout under a different email, or a marketplace order that isn't in Shopify. The agent should ask one clarifying question or hand it to a human — never assert "we have no record of your order," which reads as an accusation.
• No tracking yet (pre-fulfillment). The order is paid but not shipped. The honest reply is "it's confirmed and being prepared, you'll get tracking when it ships," not a fabricated date. If a customer is anxious about a pre-fulfillment delay, that's often a human conversation.
• Shipping trouble. Tracking shows delivered but the customer says it never arrived, or the parcel is stuck/returned-to-sender. These are lost-package and address-correction tickets wearing a WISMO costume. The agent should recognize the signal and escalate with the order context attached, not loop the customer through tracking links.
• It's not really about the order. "Where's my order, and also I want to cancel it" or "...and the discount didn't apply." A blended ticket needs a human or a different agent. Keep this agent's lane narrow and let it hand off.

The design principle: the agent's job is to resolve the easy 80% perfectly and route the hard 20% cleanly — with the order already looked up and a one-line summary attached, so the human who picks it up starts ten steps ahead instead of from scratch. That handoff quality is what separates a deflection gimmick from real support automation.

A second honest caveat: this agent is only as good as Shopify's data. If your tracking is unreliable, your carrier feeds lag, or fulfillment is handled in a system Shopify can't see, automate WISMO after you've fixed the data — not before. And the durable win isn't deflecting more WISMO tickets; it's reducing them at the source with proactive shipping notifications and a clear tracking page. Industry best-in-class teams keep their WISMO rate under 4% of orders (WISMOlabs). Automation handles the ones that still slip through.

What it costs to run

Macha bills in credits, charged per AI action — not per ticket "resolved" or "deflected." The default model, GPT-5.4 Mini, is 1 credit per action, ranging roughly 0.5–9 credits depending on the model you pick. A WISMO resolution is a handful of actions (read ticket, look up order, reply, update status), so it's a small, predictable number of credits per ticket — and because you only pay for actions actually taken, the messy ones that escalate to a human cost less, not more.

Compare that to a manually handled WISMO ticket: $1–$5 in direct agent labor, and $5–$8+ fully loaded (Ringly, Kustomer; the higher end aligns with the $6–$12 all-in figure WISMO platform guides cite). Multiply that by the highest-volume ticket type you have and the math is the easy part.

A back-of-envelope model, with every number to be checked against your own queue: say WISMO is 40% of a 5,000-ticket month (2,000 tickets), and a tight order-status agent resolves 60–70% of them end-to-end — the deflection range vendors like Fini and Gorgias report for narrow, well-scoped order-tracking automations (the same guides anchor it at roughly $1M/yr of WISMO support spend per ~50,000 orders, so the savings scale with volume). That's ~1,200–1,400 tickets your team never opens. At a conservative $3 of loaded cost each, that's $3,600–$4,200 a month back, before you count faster first responses and the CSAT lift of an instant answer at 2 a.m. The credits to run those resolutions are a fraction of that.

Two honest caveats on the model. First, deflection rate is not a number we can promise — it depends on how clean your Shopify data is, how clearly customers phrase tickets, and how aggressively you tune the escalation rules. Measure it on your own queue for a week before you bank any of it. Second, the durable win isn't a high deflection percentage; it's a lower WISMO volume in the first place (see below). You can try the whole thing on the 7-day free trial, no credit card required — see pricing for plan details.

Why resolve on the ticket, not in a chat widget

Most tools that "automate WISMO" are storefront chat widgets — they answer the question if the customer happens to ask it in the live-chat bubble before they email. But a huge share of WISMO arrives as email and helpdesk tickets, contact-form submissions, and post-purchase replies. Those never touch a chat widget.

Because Macha lives inside your helpdesk, it resolves WISMO wherever the customer actually raised it — on the Zendesk, Freshdesk, Gorgias, or Front ticket — and leaves a clean audit trail in the tool your team already lives in. No new inbox, no migration, no asking customers to change how they reach you. That's the whole point of an agent layer on top of your existing stack rather than another silo beside it. (If you're on Zendesk specifically, the Macha on Zendesk page walks through exactly how the agent operates inside the Agent Workspace.)

Frequently asked questions

Does this replace my helpdesk? No. Macha sits on top of Zendesk, Freshdesk, Gorgias, or Front and works your existing tickets. You keep your helpdesk, your macros, your reporting — the agent just handles the WISMO volume inside it.

What if the customer doesn't include an order number? The Get Order action can look up orders by the requester's email, so most tickets resolve without an order number. When there's genuine ambiguity — multiple recent orders, a guest checkout under a different email — the agent asks one clarifying question or escalates rather than guessing.

Will it ever invent a tracking number or delivery date? It shouldn't, and your instructions should forbid it explicitly. The agent only relays fulfillment and tracking data it actually read from Shopify; if that data isn't there (pre-fulfillment, or no tracking), it says so honestly and escalates the anxious cases instead of fabricating.

Does it work on Gorgias / Freshdesk / Front, not just Zendesk? Yes. The example uses Zendesk + Shopify, but the same four actions and trigger map onto Freshdesk, Gorgias, and Front. Gorgias is Shopify-native, so it's an especially natural pairing for commerce teams.

How is it billed — per resolved ticket? No. Credits are charged per AI action, with the default model at 1 credit per action (0.5–9 by model). A WISMO resolution is a few actions, and escalated tickets cost less because fewer actions run. See pricing.

How long does it take to set up? If Shopify and your helpdesk are already connected, building and testing this agent is an afternoon, not a project. The longest part is reading its test replies carefully before you switch the trigger on.

Ready to take WISMO off your team's plate? Connect Shopify and your helpdesk, build the order-status agent above, and try it on the 7-day free trial, no credit card required. Read the order-status use case and the docs to go deeper.