Every retail operations lead knows the shape of the November-to-January wall before they hit it. Black Friday opens the floodgates. Cyber Monday pours more in. The first week of December becomes the busiest support week of the year. Then January arrives with the returns wave, and the phone does not stop.
Here is the part that hurts. The surge lands on a support team that is often half temporary. Seasonal hires learn the returns policy on the fly, in the middle of the busiest six weeks of the year, while the queue keeps growing.
The result is a stretch where your phone rings more than ever and answers less well than ever.
The peak, defined
The holiday contact surge runs from late November into the first weeks of January, and it has a predictable shape.
Start with volume. Black Friday contact volumes run 2 to 3 times normal, and at the sharpest moments up to 3 to 5 times normal (Concentrix, 2026, vendor platform data). Holiday support sessions rise about 22% across the period (Typewise, 2025, vendor data drawn from 10 million interactions). The first week of December alone produces a 79% ticket spike (Shopify, vendor platform data).
Then look at composition. Most of those contacts are not complicated. They are order-status questions, the kind the industry calls WISMO, short for "where is my order." WISMO makes up roughly 21% of retail contact volume on average and climbs to about 36% in December and January (DigitalGenius, vendor platform data). Gorgias puts it near 18% of support tickets (vendor platform data). Either way, roughly one in five contacts, and worse at peak, is a customer asking a question about a parcel.
Then returns. US returns reached $890 billion in 2024, or 16.9% of retail sales (NRF, industry body, a strong figure). UPS reports returns peaking around 2 January. So just as the order-status wave subsides, the returns wave replaces it, carrying its own set of policy questions.
That is the peak: two overlapping waves, a status wave and a returns wave, across six to eight weeks.
Why WISMO is the wrong use of a human
Do the arithmetic and the problem stops looking like a staffing inconvenience.
ContactBabel's 2025 US Contact Center Decision-Makers' Guide puts the average cost of an inbound call at $7.16 (industry body, widely cited). If WISMO is about 21% of retail contact volume on average, and higher in peak, then a meaningful slice of every support dollar goes to a task with almost no judgement in it: telling a customer where a parcel is.
Take a modest example. A retailer handling 2,000 calls in a peak week at $7.16 per call spends about $14,300 on calls that week. If roughly a fifth are WISMO, that is about 400 calls and about $2,900 spent answering "where is my order." Across six to eight weeks of surge, that repetitive, low-judgement contact consumes a large share of the team's hours. It also consumes exactly the hours when your trained people are most needed for the calls that require judgement: a damaged delivery, a return dispute, a customer who has already called twice.
Human time is the scarce resource in a peak season. Spending it on order status is the wrong allocation.
The training problem nobody counts
Seasonal hiring is how retail survives the surge. It is also how policy errors happen.
A new hire who has been on the phones for two weeks does not know the difference between the standard return window and the extended holiday one. They do not know which shipping cut-off applies to which region, or whether an opened item qualifies for exchange. So they guess, or they interrupt a colleague who is equally slammed, or they put the caller on hold.
That cost rarely appears on a spreadsheet. It shows up as inconsistent answers, as callbacks from customers who were told the wrong thing, as returns processed outside policy, and as escalations a confident answer would have prevented. Multiply one wrong answer by the volume of a peak week and it becomes real money.
Consistency is the quiet requirement of a holiday support team. When the answers come from many temporary voices, consistency is the first thing to go.
What an AI receptionist can and cannot do
This is the section to read carefully, because the honest version matters more than the flattering one.
What it can do:
- Answer every call during the surge, including the ones that arrive while every human is already on a call.
- Answer policy and FAQ questions from a knowledge base you provide: shipping windows, return windows, store hours, exchange rules, how to start a return.
- Capture the caller's details and the order reference, so any follow-up starts with information instead of a blank page.
- Send the caller an SMS with a link, a return label, or next steps.
- Route or escalate the call to a human when the question needs one.
- Work outside support hours, when the surge does not pause just because the office closed.
What it cannot do, and this is the part that matters:
It cannot pull live order status, tracking numbers, or delivery dates out of an e-commerce platform. That requires an order-system integration, and that integration is not built. Custom integrations are Enterprise-only, so unless one is built for you, the assistant has no live order lookup.
Do not read a capability into the assistant that is not there. The honest description is this: it takes the call, captures the order reference and the caller's details, answers the policy question, sends the customer a link, and routes or messages them. It does not replace a live order-lookup integration. It does not know where the parcel is. It knows what your policy says, and it makes sure the customer hears it and the right person gets the reference.
That distinction is the whole point. For a large share of WISMO calls, the customer is asking two things at once: a policy question (when should it arrive, what happens if it does not) and a status question (where is it right now). The assistant can own the first completely. The second belongs to an order system you either have integrated or you do not.
Where an order lookup does exist, the assistant can route the call to it or to the human who can run it. Where it does not, the assistant still removes the policy and capture work from your team, and the customer still gets a faster answer than a queue would give them. Hear a live assistant handle a call →
The limits, stated plainly
An honest article about this technology has to name what it does badly.
Accuracy has a ceiling. VAmoS Bench, an academic evaluation published in July 2026 covering 11 voice stacks and 3,300 simulated calls, found overall task completion of 63.1%. Simple tasks completed 65.8% of the time; complex tasks, 52.2%. The best stack reached 71%, the worst 43%. Read that as it is: roughly two in three tasks done end to end, worse on anything complicated. It is why routing and escalation matter more than autonomy.
Consumers are resistant. ServiceForge and OnePoll surveyed 6,000 adults across the US, Canada and the UK in October 2025. 33% said they would hang up if connected to an AI. 53% said they trust a business less when AI is used. In a separate survey of 1,011 US consumers by Kinsta and Propeller Insights, 49.6% said they would consider cancelling over AI-only service. Gartner's fieldwork with 5,728 customers found 64% would prefer companies did not use AI in customer service at all.
The cost of getting it wrong is measurable. COPC's 2025 benchmarking found satisfaction at 74% for the most recent AI interaction, rising above 90% when the interaction was resolved, but NPS at minus 70 points when it was not. Human handoff is the most frequent point of failure. An unresolved AI interaction does not just fail; it damages the relationship.
Holiday shoppers raise the bar further. They are stressed, time-pressured, and often spending money they have already committed. A bot that cannot help them is worse in December than in July, because the stakes feel higher and patience is thinner.
None of this makes the technology useless. It means you aim it at the calls it can genuinely finish, with an easy path to a human for the rest. One finding is worth holding onto: COPC found satisfaction 34% higher when callers were told they were speaking to a machine. Say so up front.
A note on the numbers you will see quoted
A quick word on the statistics floating around this topic.
You will see claims that 85% of callers never call back after a missed call, that 80% of people hang up on voicemail, and that a small business loses $126,000 a year to missed calls. None of these can be traced to a primary source. They are repeated so often that they read as fact, but the trail ends at marketing pages, not studies. This article does not use them.
The figures here come from named sources with stated methods or published reports: ContactBabel for cost per call, Concentrix, Typewise and Shopify for holiday volume, DigitalGenius and Gorgias for WISMO share, NRF and UPS for returns, VAmoS Bench for accuracy, ServiceForge, Gartner and Kinsta for consumer resistance, and COPC for outcomes. Some are vendor platform data, which means they reflect one vendor's customer base, and they are labelled that way. Some are industry-body or academic, which is stronger. Where a number is weak, this article says so. Hold any vendor to that standard, including this one.
The practical next step
Before you decide anything, hear one work. The demo hub at /voiceagent/demo/ lets you talk to a live assistant in your browser, no signup. Ask it a policy question, then ask it where your order is and listen to how it handles not knowing.
If you want the cost picture first, the breakdown is at /resources/how-much-does-an-ai-receptionist-cost/. OrdrNow is $199/month plus a one-time $199 setup fee, including 300 minutes.
The decision is not whether AI can run your holiday support desk. It cannot. The decision is whether it should take the repetitive policy and capture calls, so your trained people can handle the ones that need a person.