Peak season: absorbing support spikes in chat

How an AI assistant absorbs four cut-off dates of the Christmas season: knowledge base freeze, shipping cut-off, holidays and returns. With a checklist.

In plain language

Before Christmas, many questions come at once. This time is in the calendar long before. So plan texts, staff and steps early. Many customers do not read a list of questions. A chat that answers reaches them too. On Sundays and holidays, your team may usually not work. The questions still come. The way to a person stays important when a customer has a problem.

September 7, 2026 14 min read By XICBOT

Key points

The peak is a calendar, not a weather event. 18.5 percent (HDE) of the German retail year 2025 fell into November and December. Naming the four cut-off dates in advance lets you align content, staffing and escalation paths instead of improvising through Advent.
The knowledge base decides how much relief is possible at all. Only 28 percent (Bitkom) of online shoppers would look at an FAQ page when they hit a problem. A channel that answers also reaches the people who no longer read a list.
The holiday window is a legal question, not a staffing one. Under section 9 of the German Working Hours Act, employment on Sundays and public holidays rests from 0 to 24 hours (Arbeitszeitgesetz). The incoming questions do not rest along with it.
Handover to a person is the core of the setup, not its emergency exit. 62 percent (Bitkom) of online shoppers want to reach a human quickly when something goes wrong, and satisfaction there stands at 86 percent (Bitkom) against 50 percent (Bitkom) in an automated dialogue.

Why the peak is predictable

Seasonal business is usually described as if it broke over a company without warning. The figures say otherwise: the shape repeats year after year, only the height of the curve changes. The German retail association put sales in the final two months of 2025 at 126.2 billion euros (HDE), which is 18.5 percent (HDE) of annual turnover. The Federal Statistical Office reported, on preliminary results, 3.2 percent (Destatis) more real turnover for December 2025 than for the December before. Both are numbers you can build a service plan on, because they do not appear on the day of the peak but are known weeks ahead. Anyone running an assistant for an online shop therefore plans it like a warehouse: by lead time, not by daily form.

At the same time the load shifts into the digital channel. For online Christmas trade in 2025 the association expected 22.2 billion euros (HDE) at a nominal increase of 3.3 percent, and mail order and internet retail grew 10.1 percent (Destatis) in real terms year on year. The German e-commerce association reported a 2.9 percent (bevh) increase for October and November 2025 and gross goods revenue of 83.1 billion euros (bevh) for the full year 2025. The customer side matches: 73 percent (Bitkom) of Germans order Christmas presents online and 17 percent (Bitkom) buy them exclusively on the internet. For that group there is no counter where a question could be settled in passing. The channel that answers is the only one they have.

The four cut-off dates of the season

A seasonal plan that consists of one number for December helps little in daily work. It becomes usable only once it names four dates on which the character of the incoming questions changes. Each section needs different content in the knowledge base, different handover rules and different staffing. The mockup above shows the four edges on a request curve across fourteen calendar weeks.

Date 1: knowledge base freeze

Roughly four weeks before the first discount day, range, prices, delivery times and return rules are settled. Anything changed afterwards is added individually and logged instead of being reloaded in one block.
Date 2: shipping cut-off

From the last promised order day onwards, the question shifts from „what does it cost?“ to „will it still arrive?“. For that the assistant needs shipment status from a connected ERP and clear wording for dates that are no longer being promised.
Date 3: holiday window

On 25 and 26 December, employment rests under section 9 of the Working Hours Act; 24 December is not a public holiday. The chat keeps taking requests during those days but states clearly when they will be handled, and promises nothing nobody can keep.
Date 4: returns wave

From the first week of January, almost every request is about withdrawal, return shipment and refund. The timing of that wave is written into the law and can therefore be planned to the day.

This split has a practical side effect: it makes measurable which section carries the largest share of the relief. As a rule that is not the discount day but the week before Christmas Eve and the second week of January, because the questions there are the most uniform. An estimate of how much volume an assistant realistically takes on is given in how much support an AI assistant handles.

Date 1: the knowledge base freeze

An assistant answers only as well as the material it draws on. In the peak this gets sharper, because every wrong statement is multiplied immediately. That is why the plan does not start with a technical date but with an editorial one: the day the knowledge base is frozen. How little a pure self-service page carries is shown clearly by consumer research: 60 percent (Bitkom) of users find an FAQ section convincing, but only 28 percent (Bitkom) of online shoppers would actually look there when facing a concrete problem. So the content is not the problem, access to it is. An assistant reverses that order: it reads the material and the visitor asks in their own words.

Delivery times per shipping method including the last order day beyond which delivery before the holiday is no longer promised. Without that sentence every answer invents its own date.
Return and exchange rules including any voluntary extension across the holidays, with a clear beginning and end of the period.
Voucher conditions, redemption paths and what happens if one is lost. Vouchers generate questions far beyond December.
Packaging and gift options, because they are among the most frequent pre-purchase questions in the season and barely matter in ordinary months.
Payment methods and invoice questions, together with what the assistant must not answer, such as credit decisions made by third parties.
A list of sentences nobody should say: no dates without cover, no stock statements without live data, no goodwill that a person decides on case by case.

The freeze is a procedure, not a lock. After the cut-off date, changes are added one at a time, with date, author and a short note on what changed. That sounds pedantic but saves the January reconstruction of why an assistant quoted a delivery time on 12 December that no longer applied that day. Which content belongs in such a store at all, and in what form, is set out in detail in building a knowledge base.

A frozen store is the strongest brake against invented answers

Most wrong statements in the season do not come from a weak model but from a store that changes faster than it is maintained. As soon as two versions of the same delivery time sit in the system, every answer picks one of them. An assistant that ties its statements to a source and stays silent rather than guessing when cover is missing is worth more in the peak than one that says something to every question. The approach is described in preventing hallucinations.

Date 2: shipping cut-off and the rhythm of delivery questions

The first hard peak of the season sits before the shipping cut-off. For the two large November discount days in 2025 the retail association expected sales of 5.8 billion euros (HDE), and 54 percent (HDE) of the online shoppers surveyed used those days explicitly for Christmas shopping. The average budget planned for those discount days was 312 euros (Bitkom), around eleven percent above the year before. The peak therefore sits roughly four weeks before the holiday, and its date is known. A company that only notices on Friday morning that the chat does not know the discount conditions has given away the lead time the calendar had handed it.

After the discount weekend the tone changes. It is no longer about choice but about dates. On gift purchases each person spent 263 euros (HDE) on average in 2025, while the average order value in German e-commerce is 146.19 euros (bevh). A single unanswered question therefore carries an amount you can put a figure on. For an assistant to hold here it needs access to shipment status from the ERP connected in the Custom package and to the order itself; integration with shop and inventory systems matters more at this point than any wording. How an assistant triggers such lookups in a controlled way instead of guessing is described under control of connected tools.

Situation in the peak	Without a prepared assistant	With a prepared assistant
Question about the last order day	Answer differs from agent to agent	One stored sentence per shipping method, identical everywhere
Shipment is delayed	The follow-up lands in the general queue	Shipment status from the connected ERP inside the conversation, handover only as an exception
Discount condition unclear	Customer drops out or orders twice	Condition taken from the frozen store, with a source
Request on 25 December	Automatic out-of-office notice without content	The request is recorded and the handling time is stated
Return request in January	A form the customer has to find	A guided flow with the deadline stated and a confirmation
Review after the season	A gut feeling about what was difficult	Topics, drop-offs and handovers backed by figures

Date 3: the holiday window

The third cut-off date is the only one no company can negotiate. Section 9 of the German Working Hours Act states that employees may not be employed on Sundays and public holidays from 0 to 24 hours (Arbeitszeitgesetz). For the Christmas window that means: 24 December is usually not a public holiday, 25 and 26 December are, and so is 1 January. During exactly those days orders keep coming in, shipments sit still and presents are unwrapped. An assistant in support does not replace a shift here, it keeps the channel open: it records requests, classifies them, gives a documented answer where one is stored, and otherwise states when the matter will be handled.

Demand shifts away from the shop counter anyway. In the week before the first Advent Sunday in 2025, 70 percent (HDE) of the retail companies surveyed reported lower visitor numbers than in the previous year. In the week before the fourth Advent Sunday, in a trend survey run from 19 to 21 December 2025, 66 percent (HDE) were dissatisfied with how Christmas trade had gone so far. Fewer people in the shop also means losing the casual clarification at the till. The question does not disappear, it moves to a channel that has to take it. That is precisely why chat is not an add-on in this phase but the place where the clarification happens that used to take place between shelf and checkout.

The most expensive hour of the season is not the one with the most requests but the one in which nobody answers and nobody says when an answer will come. An assistant that honestly points to 27 December keeps more customers than an out-of-office notice without a date.

From preparing service teams for the seasonal peak

The holiday window also shows whether the operation of the assistant itself has been thought through. If a connected interface fails during those days, nobody is sitting next to it to notice. A staged fallback therefore belongs in the seasonal plan: the assistant keeps answering without live data, marks the missing connection and directs to a form as soon as it can no longer give a reliable answer itself. The stages of that fallback are described in when the AI assistant fails.

Date 4: the returns wave in January

The fourth edge is the most predictable one, because its timing is written into the law. Under section 355 of the German Civil Code the withdrawal period is 14 days (BGB); under section 357 the goods received must be returned within a further 14 days (BGB). Those two periods produce the shape of the January wave almost by themselves: first declarations of withdrawal, then questions about the arrival of the return shipment, then questions about the refund. On the customer side the wave announces itself clearly: 49 percent (Bitkom) of respondents intend to take unwanted presents back to the shop or send them back to the online store. An assistant in the shop can guide that flow instead of administering it.

The path itself decides more than the individual case. Almost half of online shoppers, 49 percent (Bitkom), have already kept a product because the return process was too complicated; free return shipping and a fast refund each matter to 94 percent (Bitkom) of respondents. On top of that comes demand reaching far beyond January: a voucher without an explicit expiry stays redeemable for three years from the end of the year of purchase (HDE). Questions about that reach a service team spread across months and are particularly well suited to a stored answer that reads the same each time.

Accept the withdrawal instead of advising on it. The assistant takes the declaration, confirms receipt with a date and explains the next step in two sentences.
State deadlines, do not estimate them. The statutory period is named, the internal handling time separately. Mixing the two produces exactly the follow-up question you wanted to avoid.
Show the return path. Label, address and required condition of the goods in one answer, without the customer switching between three pages.
Make the refund status retrievable. The most frequent second question in January is where the money is. With a connection to the order system it is answered in one sentence.
Hand over special cases cleanly. Damaged goods, partial returns and goodwill belong to a person, with the full conversation attached. How a move to a new assistant preserves those paths is described in replacing a chatbot without a break.

Where the assistant hands over

A relief calculation that cuts the human out does not survive the season. When something goes wrong with an order, 62 percent (Bitkom) of online shoppers would rather turn to a person they can reach quickly, while 36 percent (Bitkom) explicitly want automated help. Serving both at once is not a contradiction but a division of labour: the assistant takes the uniform cases in volume and speed, the human takes the rare and expensive ones. For that to hold, the handover has to be defined in advance and carry the conversation with it; the procedure is set out in handover to human agents.

The same survey supplies the benchmark. 50 percent (Bitkom) of users are satisfied with an automated dialogue, 86 percent (Bitkom) with a human contact. That gap is the real yardstick for the design: an assistant does not have to be better than a person, it has to keep the path to a person short and must not block it. A further observation from Christmas trade fits: 74 percent (Bitkom) of Germans are convinced they know better than a machine what someone would be pleased to receive. An assistant that tries to judge taste in Advent works against that expectation; one that clarifies availability, dates and returns works with it.

What the review after the season contains

The season does not end on 31 December but with the review at the end of January. It is the reason next year's preparation gets shorter: what was measured does not have to be guessed again. The bar is set high, because customer satisfaction in German e-commerce has stood at 96.2 percent (bevh) at a similar level for years. Against that background a service drop in the peak stands out immediately. The figures for it arise during normal operation and can be brought together in the conversation analytics without anyone reading transcripts by hand.

Requests per calendar week and topic, split by the four sections of the season. Only that split shows which edge actually held.
Share of conversations with a handover and the reason for it. If the share rises in a particular week, a topic that became new there is missing from the knowledge base.
Drop-offs without an answer: conversations in which the visitor leaves before an answer exists. They are the most reliable trace of a gap in the store.
Response time across the day, especially in the evening and on public holidays. This shows whether the channel covers the hours in which the shop is closed.
Recurring phrasings without a match, collected as a list. It is the working basis for next year's store.
Topics with a legal dimension such as withdrawal, refund and vouchers, because a sloppy answer there gets expensive. A guide to the review is given in analysing and improving conversations.

Next year's seasonal plan then almost writes itself out of the review. The topic list becomes the task list for the store, the handover reasons become rules, the hours with the longest response time become a staffing question. Anyone selling internationally adds languages to the review: public holidays, shipping cut-offs and return periods differ by market, and a multilingual assistant needs its own reviewed store per language, not a translation of the German one.

Enter the four cut-off dates in the calendar and name a responsible person per section
Schedule the knowledge base freeze roughly four weeks before the first discount day and log changes individually after that
Put the shipping cut-off wording per shipping method in writing, with a date and an expiry rule
Test the connection to order and shipment data before the peak, not during it
Sign off handover rules, availability and the holiday message before the first Advent Sunday
Schedule the review for the end of January so the figures feed into next year's plan

Whether your assistant already states the shipping cut-off, holidays and returns correctly shows up in a test conversation before the first sale day. Tell us your cut-off dates and we will check them against your content together.

Frequent questions

When should we start preparing for the Christmas season?

September is sensible, early October at the latest. The reason is not technical but editorial: range, delivery times, return rules and voucher conditions have to be agreed and written down, and several departments are involved in that. Count backwards from the first discount day in November: roughly four weeks for the content, two weeks for test cases and sign-off, one week of buffer. Starting later, you can still set up an assistant, but you lose the part that produces the actual relief, namely reviewed content.

Does an AI assistant replace extra seasonal staff in service?

It does not replace them, it shifts their task. The assistant takes the uniform cases around delivery time, returns and vouchers, with an ERP connected in the Custom package also shipment status, including evenings and public holidays. What remains are the cases involving judgement: complaints, goodwill, damaged goods, special arrangements. That is exactly where an experienced person is worth more than an extra hire who has to be trained. Since 62 percent (Bitkom) of online shoppers want to reach a human quickly when there is a problem, the short handover is the core of the setup and not its edge case.

What happens on public holidays when nobody on the team is available?

The chat stays open and takes requests, but states clearly when they will be handled. Under section 9 of the German Working Hours Act, employment on Sundays and public holidays rests from 0 to 24 hours, and no service promise changes that. The assistant still remains able to answer: everything documented in the stored material it answers unchanged, and everything else it records with a confirmation and a time. What matters is that the stated handling time is then actually met, otherwise the message does more harm than none at all.

How do we stop the assistant quoting wrong delivery times in the peak?

Through three things that belong together. First a frozen store with exactly one valid sentence per shipping method, so that no two versions sit in the system. Second a binding of every answer to a source, so that nothing is said without cover. Third an expiry rule: after the last promised order day the sentence switches automatically to wording without a delivery promise. A short list of forbidden statements, agreed with the specialist team before the season, helps on top of that.

Is this worth it for a smaller shop with a few hundred orders?

The calculation depends less on order volume than on distribution. A smaller company often has only one or two people in service during the peak, and that is exactly where a doubling of requests hits straight away. With an average order value of 146.19 euros (bevh) in German e-commerce, even a modest number of abandoned conversations decides the benefit. A narrow scope makes sense: few topics, cleanly stored, rather than an assistant that attempts everything. An overview of scope and effort is on the pricing page.

What do we measure to know after the season whether it worked?

Four figures are enough to start with: the number of conversations per calendar week and topic, the share of conversations handed over to a person together with the reason, the drop-offs without an answer, and the response time across the day. The relationship between those four values shows which section of the season held and where the store had a gap. It matters to look at the values per section separately: an average across fourteen weeks hides exactly the peaks the review is made for.

Ask it yourself, Assistant of this website

Automated assistant

Settings