How Copperline Outfitters ran peak season with zero seasonal hires

Every September, Copperline Outfitters' support team used to start hiring. The DTC outdoor retailer sells tents, sleeping bags, technical shells and trail shoes, and a large share of its revenue lands between Black Friday and Christmas. For most of the last decade, peak season meant about 110 temporary agents: six weeks of onboarding, a two-week ramp, and a CSAT dip that everyone had come to expect.

In 2025 Copperline hired no temps. Conversation volume peaked at 3.4 times its October baseline. Pazau Agent resolved 71% of conversations end-to-end, and CSAT went up.

I spent January with Copperline's CX operations team going through the data. This is what they did, in order, including the parts that didn't work the first time.

The shape of the problem

Copperline's contact volume isn't just bigger in December. The mix changes too.

Intent	Share, October	Share, peak (Nov 24 – Dec 24)
Where is my order (WISMO)	22%	38%
Returns	14%	11%
Exchanges (size / color)	9%	14%
Sizing and fit questions	11%	13%
Product questions	18%	9%
Promo and price adjustments	6%	8%
Everything else	20%	7%

During peak, four intents make up 76% of volume. They are WISMO, returns, exchanges and sizing. All four can be answered or completed from data Copperline already has. The order is in Shopify, the shipment is in the carrier's tracking API, and inventory is in the warehouse management system.

That was the thesis: if the agent can read and write those systems within the right limits, most of the December spike never has to reach a person.

What they built

Copperline went live on Pazau in April 2025, which gave the team seven months before the test that mattered. Sam Duarte, Copperline's Head of CX Ops, ran the project with two operations analysts and one engineer who worked on it part-time.

"We didn't want a bot that was nice about telling people to wait. We wanted it to do the exchange. Our bar was: if a trained agent with the right permissions would just handle it, Pazau should just handle it."

— Sam Duarte, Head of CX Ops, Copperline Outfitters

Procedures

The team wrote 23 Procedures, in plain language, mostly by adapting their existing agent training docs. The four peak intents got the most attention:

• WISMO checks the order in Shopify and pulls live carrier tracking. If a package hasn't had a scan in 72 hours, the Procedure opens a carrier trace and tells the customer the specific next step and date. It doesn't just paste a tracking link.
• Returns checks eligibility against the order date and product category. Holiday purchases have an extended window to January 31. It then generates a prepaid label and emails it.
• Exchanges checks live inventory in the warehouse system for the requested size or color. It reserves the unit, creates the exchange order, and issues the return label, all in one conversation. If the item is out of stock, it offers the nearest in-stock alternative or a refund.
• Sizing uses Copperline's own fit data per product line, including notes like "runs small in the shoulders; size up if layering". It answers from that data, not from generic size charts.

Actions

The Procedures call 14 Actions into three systems:

shopify.get_order            shopify.create_exchange_order
shopify.create_return        shopify.issue_refund          (limit: $250)
shopify.apply_price_adjust   (limit: $40, once per order)
wms.check_inventory          wms.reserve_unit              (hold: 30 min)
carrier.get_tracking         carrier.open_trace
labels.generate_return       email.send_template
loyalty.get_member           loyalty.add_points            (limit: 500)

Every Action with money attached has a limit. Refunds over $250 go to a human approval queue in Pazau Inbox with the full context already assembled, so the agent only has to review and click. During peak, 2.1% of refund requests hit that queue, and the median time to approval was 11 minutes.

What broke, and what they changed

Peak didn't go perfectly. Three problems stood out, and how the team handled them is the most useful part of this story.

1. A carrier outage on December 2

A regional carrier's tracking API returned stale data for about nine hours on the Tuesday after Cyber Monday. Pazau Agent kept telling customers their package was "in transit, arriving Thursday" when the carrier had stopped scanning.

Insights caught it first. The WISMO repeat-contact rate for that carrier went from 4% to 19% within three hours, and the anomaly alert paged the on-call analyst. The fix was a one-line Procedure change: "If the last scan from this carrier is older than the carrier's status page incident start time, tell the customer we're aware of a carrier delay and we'll email them when tracking updates." The team published it at 2:40 p.m. Repeat contacts returned to baseline by evening.

2. The exchange inventory race

On the first Saturday of December, the warehouse system reported inventory that was accurate to the minute. But 40 customers could ask for the last medium in a jacket at the same time. Early in the weekend, 63 exchanges were promised against units that were already gone.

The team added the wms.reserve_unit Action with a 30-minute hold and changed the Procedure to reserve first, then confirm. Oversold exchanges fell to zero for the rest of the season. Every affected customer received a proactive email and a $20 credit before they wrote back.

3. Gift receipts

Gift recipients returning items don't have the order number, and they don't know the buyer's email. The first version of the returns Procedure couldn't help them, so all of those conversations went to people.

The analysts added a gift-return path. It looks up the order from the gift receipt code, issues store credit instead of a refund to the original payment method, and never reveals the buyer's name or order details. The change was simulated against 400 historical gift-return conversations before it was published. Gift returns went from 0% to 64% AI-resolved within a week.

The numbers

Here's peak 2025 compared with peak 2024, which ran on Copperline's previous helpdesk with a scripted chatbot and 110 temps.

Metric	Peak 2024	Peak 2025
Conversations (Nov 24 – Dec 24)	412,000	486,000
Peak daily volume vs. October baseline	3.1×	3.4×
Resolved end-to-end by AI	9% (chatbot "containment")	71%
Seasonal agents hired	110	0
Median first response, chat	4 min 20 s	2.1 s
Median full resolution time, email	19 h	6 min
CSAT	4.1 / 5	4.6 / 5
Cost per contact	$5.80	$2.35

A note on the 71%: it's a strict definition. A conversation counts as resolved only if no human touched it, the customer didn't come back about the same issue within 7 days, and the customer didn't abandon it or ask for a person. The 9% "containment" figure for 2024 used the old chatbot vendor's definition, which counted any conversation where the customer closed the window. Measured the 2025 way, the 2024 number would have been about 4%.

The permanent team of 64 agents didn't shrink. It changed what it works on. During peak, those agents handled the 29% of conversations that needed judgment: damaged gear with safety implications, warranty disputes, and the customer whose wedding tent didn't arrive. Average handle time on human conversations went up 22%. The team considers that a good sign, because it means the easy conversations weren't reaching them.

What other retailers can take from this

After a month with the data, here's what I'd tell any retail CX leader planning for next November:

1. Start in the spring. Copperline's peak worked because Procedures had seven months of real traffic to be tuned against. A September go-live is a risk you don't need to take.
2. Give the agent write access, with limits. Read-only AI can explain your return policy. The volume drops only when the agent can process the return.
3. Put a hard limit on every Action with money attached. Limits make approvals rare, and approvals make limits safe.
4. Watch repeat contact, not containment. Copperline found the carrier outage because the metric they watched was customers coming back, not customers going quiet.
5. Make your team the escalation path, not the overflow valve. People are better at the hard 29% when they aren't also handling the easy 71%.

Copperline is already planning peak 2026. Their target is 78% AI resolution, and they plan to add voice for order status calls, which make up about 15% of December volume and still go to people today.

Read the full case study at /customers/copperline-outfitters, or see how Pazau works for retail and e-commerce. If you want to see what your own peak could look like, bring 1,000 of your conversations to a demo and we'll run them through Pazau Agent.