WISMO stands for where is my order. It is one of the most common contact types in ecommerce and logistics support, and most teams try to automate it first because the question looks simple: a customer wants a status update, not a conversation. Giving a correct answer is harder than it looks. It takes a live lookup across an order management system, a carrier's tracking feed, and sometimes a CRM, every single time. Deflection rate will tell you how many of these contacts an AI agent handled without a human. It will not tell you how many of those answers were actually right, and that second number is what decides whether the automation is saving money or quietly creating more work.
What WISMO Means and Why It Floods Support Queues
Every order generates the same handful of events: placed, shipped, out for delivery, delayed, delivered. Customers check status compulsively once a delivery window opens, which is why WISMO volume spikes on exactly the days a team can least afford it, right before and during peak shipping periods. The contact looks low complexity because the question never changes. The answer does, because it depends on live data from systems that do not always agree with each other: a carrier feed that can lag behind the package, a warehouse system with its own cutoffs, a CRM that may or may not have the latest address change. An agent reading from a stale cache will eventually hand a customer a confident, wrong answer, which is worse for trust than no answer at all.
Treat WISMO as its own automation initiative, with its own order tracking and returns assistant and its own metrics, rather than a line item buried inside a general support bot's resolution numbers. That separation is what makes the next problem visible instead of hidden.
Why Deflection Rate Alone Can Hide a Failing WISMO Automation
Deflection rate counts how many contacts an AI agent handled without routing to a human. For WISMO, it is tempting to treat deflection as the whole scoreboard: the bot answered, the ticket never reached an agent, the automation worked. Deflection counts a fast wrong answer exactly the same way it counts a fast right one.
Picture a customer asking where an order is. The agent pulls a tracking number, reports "out for delivery" from a feed that synced six hours earlier, and closes the contact. The package was actually marked delayed an hour after that sync. The customer does not reopen the same ticket. They open a new one, now annoyed at being told something false, and that second contact usually lands with a human because the customer no longer trusts the bot. Deflection rate logged one success. The real outcome was one wrong answer plus one escalated, harder conversation, which costs more than the single human contact it was meant to avoid.
This is the case for grading resolution instead of deflection: a WISMO automation that only reports how many contacts it took off a queue is reporting activity, not value. The fix is not to stop automating WISMO. It is to stop measuring it with one number.
The Three Metrics That Show Whether a WISMO Agent Is Actually Working
Three numbers, read together, show whether a WISMO automation is paying off or quietly costing more than it saves.
Deflection rate on its own only tells you the bot is busy. Pair it with resolution accuracy and you know whether the busy bot is right. Add re-contact rate and you know whether being right the first time is actually cutting total volume, rather than just moving the same volume later in the week. A support lead piloting WISMO automation should set a baseline for all three before expanding past a small test volume, so a drop in re-contact rate, not just a rise in deflection, is what justifies scaling it.
How AI Agents Resolve WISMO Contacts Across Chat and Voice
Resolution accuracy is a measurement problem, but it starts as a design problem. An agent that answers from a canned response breaks resolution accuracy before anyone ever tests it. An agent built to look up the real order first at least has a chance of being right.
Real Order Lookups Beat Canned Tracking Replies
A WISMO agent needs a way to check an order directly in the order management or ecommerce platform at the moment of the conversation, not a message written once and reused for every status. That lookup works the same whether the contact arrives in chat or over voice: the agent calls the tool, reads back the current status, and states the next expected milestone instead of a vague "it's on its way." The build cost of this step is small next to what a wrong canned answer costs in re-contacts later.
Route the Exceptions to a Human, Not a Loop
Some WISMO contacts are not status questions at all: a package marked delivered that never arrived, an address that changed after the order shipped, a refund the customer wants instead of a reship. These need a person, and the agent should recognize that quickly instead of repeating itself. A handoff step that passes the full transcript and the order details the agent already pulled means the human picks up mid-problem instead of re-asking for an order number. See which help desks support native handoff. When the handoff cannot start, a fail path keeps the conversation moving instead of leaving the customer stuck on a loop.
Piloting WISMO Automation: What to Measure Before You Expand
Rolling a WISMO agent out to every channel on day one makes it hard to tell whether a problem sits in the agent or in the order data it's reading. A phased pilot keeps the two separable.
Trilogy started with one core agent, then scaled it across 90 products, going from a 35% AI resolution rate in its first week to 70% today. The same discipline applies to WISMO: expand by order type and channel, not by raising a deflection target.
Before the pilot starts, decide which systems the agent needs to query, what the fail path does when a lookup fails, and who owns the weekly review of the three metrics. A free account (see current plans) is enough to build and test the order lookup against a sandbox store. When it's time to scope the live connections to your order management system and contact-center telephony, a working session can map that out before the pilot's first ticket arrives.