Peak season without night shifts: preparing Black Friday as a system

Black Friday rarely fails on the offer. It fails because inventory, support and shipping communication are not prepared.

I played eight years of handball in the German first division. That teaches you one thing: you win a game in the week before it, long before the second half. Peak season works the same way. We run three of our own brands, nano, mate and MUSTAX, with two people. We have no reserve bench that pulls night shifts in November. So peak has to be a system.

Here is the timeline, counted backwards from the sale day. Plus the hard line: what you should stop starting two weeks out.

Why the timeline runs backwards

Most people plan forwards. "We start in October" sounds like enough time and is still too late, because nobody knows what comes first.

Backwards is more reliable. You set the sale day as day 0 and work back from there. Every task gets a last sensible date. Whatever no longer fits, you cut on purpose instead of attempting it in a rush during peak week.

Roughly it looks like this:

The rest of this article fills those rows with what actually happens on our side.

10 to 8 weeks out: inventory and replenishment

The most expensive mistake in peak is a bestseller that sells out on Friday morning while the ads keep running.

Three things you settle now:

• Which products go into the sale. Five clear ones beat thirty unclear ones. At nano we learned that an offer everybody understands beats a discount matrix.
• How much stock you allocate per product. Take the last peak as your base and add the factor your growth supports. Stay conservative there.
• What happens when stock tips over. Automatic thresholds that warn you and pull the item out of the sale if needed.

The third point is the one almost everyone skips. On our side an alert lands in Slack when a sale product drops below a defined threshold. There is no dashboard anyone would have to open. The message comes to us.

If you sell on several channels, that is not a nice-to-have. Two channels selling the same stock produce cancellations in series during peak. How to set that up cleanly is in multichannel inventory management.

8 to 6 weeks out: build what has to carry the load

This is the window where you can still set up real automations. Not later.

The order that has worked for us:

1. Catch WISMO. Around 40 percent of all support requests are "where is my order". During peak the share goes up, because more parcels are in transit and transit times get longer. The build is here: automating WISMO requests.

2. Answer standard questions automatically. Shipping cost, delivery time, discount code not working, sizing advice. Our AI answers around 65 percent of requests automatically. The rest comes to us as a draft or an escalation.

3. Prepare returns. Yes, already. The wave is certain, and it arrives at a moment when you are tired. A self-service portal takes most of it off your plate: self-service returns portal.

The reason for this order is simple. Every point reduces requests instead of answering them faster. Answering faster scales linearly. Fewer requests does not scale with volume at all.

6 to 4 weeks out: preparing the support wave

Support during peak is a preparation problem. Capacity comes second. You cannot prevent the volume. You can only make sure it arrives in an orderly way.

Three building blocks:

Templates that are current. Go through all standard replies and update them to peak reality. Delivery times, sale conditions, return window. A template that still promises "delivery in 2 days" in November creates more work than it saves.

Auto-replies with substance. An acknowledgement that only says "we will get back to you" is a wasted message. Ours contains the order status, the tracking link and the three most common answers of peak week. A noticeable share of requests resolves itself before a human sees it.

One escalation rule everyone knows. What goes out automatically, what goes to a human as a draft, what gets escalated immediately. On our side emotional cases, payment problems and anything above a certain order value always escalate.

If you are starting support from scratch, the guide to customer service automation is a better entry point than this section.

And measure along the way. Without numbers you will not know after peak whether it worked or whether you just survived. Which metrics are worth it is in customer service KPIs.

Shipping communication: the underrated lever

Most requests during peak come up because the customer does not know whether something is going wrong. The things that actually go wrong are a smaller share.

So during peak we follow one rule: better to inform one time too often than one time too little.

Concretely, our customers get this in November:

• An order confirmation with a realistic delivery time, not the one from summer
• A shipping confirmation with a tracking link as soon as the label is created
• A message when a parcel sits still longer than usual
• A delivery confirmation that also triggers the review request

The third point is the most important one. A parcel stuck in the same status for three days will produce a ticket without a message. With a proactive message it usually does not. That is the difference between "these people are on it" and "nobody is getting back to me".

And one more uncomfortable item: communicate the cut-off for Christmas delivery early and visibly. Hiding it buys you a second wave of requests in December.

4 to 2 weeks out: the technical load test

Peak brings load spikes in places that never show up in daily business. The store itself is rarely the problem. The connections in between are.

What we check:

• Rate limits. How many requests per minute does your fulfilment API allow, your email tool, your ERP? During peak you run into them, in daily business never.
• Queues and retries. What happens to an order when the fulfiller stops responding for a moment? It has to go into a queue and be retried later instead of getting lost.
• Webhooks under load. Do all events arrive, even when 300 orders come in within an hour?
• Alerts. If something stalls at night, somebody has to be woken up. An error nobody sees is an error that runs until Monday.

The test asks what happens when a part fails, which goes well past the question of whether it works at all. Pull the plug on one system as an experiment and see whether your flow waits cleanly or quietly loses data.

The line: what you stop starting two weeks out

This is the part nobody likes to hear, and it saves you the most.

Two weeks before the sale day, new construction ends. From here you touch nothing load-bearing. Specifically, stop starting:

• A store relaunch or a new theme. Every layout change is a new conversion risk with no time left to measure it.
• A switch of fulfiller or ERP. Migrations need at least one quiet month to surface the errors that only appear in operation.
• A new helpdesk system. The team learns it during the worst week of the year. Templates and rules are missing exactly when they count.
• An AI that answers on its own without an approval phase. Automatic replies need a few weeks in draft mode before you let them go. Otherwise you send nonsense to a thousand customers during peak.
• New payment methods. Sounds like a switch, and it is a new flow in bookkeeping, returns and refunds.

What you can still do two weeks out: change copy, update templates, book in stock, adjust thresholds, sharpen alerts. Anything reversible that does not rebuild a flow.

If you now notice that half your list falls under "too late": write it down and do it in January. The next peak is certain, and January gives you the quiet to do it.

After the sale: the returns wave

Peak does not end on the Monday after. It ends when the returns are through, and on our side that is four to six weeks.

What counts now:

• Self-service instead of email threads. The customer files the return, the label is created automatically, the status is visible.
• Refunds without chasing. Goods receipt triggers the refund. Every day of delay is a ticket.
• Stock back on sale immediately. Checked goods have to be sellable again within a day, not after the next stocktake.

At mate returns take around 5 minutes a day in normal operation. In January it is more, and it stays a task instead of a week.

And then the most important part: put a date in mid-January in your calendar and write up the peak. What held, what broke, which request came in most often. That is your template for next year, and in two months you will no longer be able to reconstruct it.

Conclusion

Black Friday is an operations event with marketing in front of it. Treating it as a marketing event with an operations appendix is how people end up in night shifts.

The backwards timeline sorts that out for you. Inventory eight weeks out, automation six weeks out, templates and load test in the final month, then quiet. Plus a plan for the returns wave before it arrives.

We do this because there are two of us and we have no buffer. That is exactly why it also works for you when your team is small.

If you want to know which of these building blocks gives you the biggest lever before peak: talk to us at Flowhouse. We are happy to show you how it runs at our own brands.