AI Customer Support by the Numbers - What 3,769 Real Customer Messages Show (2026)

Most statistics about AI in customer support come from vendor surveys or global benchmarks. This page is different: it reports what actually happened in 3,769 customer messages that an AI assistant answered for Slovak companies between April and October 2026. The method, the sample and its limits are stated below, so you can judge how far the numbers carry.

Key findings

46% of customer messages arrived outside business hours (Monday to Friday, 08:00-17:00). 17% came at the weekend and 25% between 20:00 and 08:00.
67% of web chat visitors never needed a human. The other 33% were handed over to the support team at least once.
71% of messages were knowledge-base questions - prices, conditions, opening hours, procedures. Bookings were 13%, complaints 4%.
Messages came in 14 languages. English (51%) outnumbered Slovak (35%), and about 13% were in neither of them nor Czech - mostly Arabic (7.2%) and Hebrew (3.2%).
Knowledge-base questions were handed over in 11% of cases; bookings in 29%; complaints in 99% (complaints with negative sentiment are routed to a human by rule).
4% of messages had negative sentiment, 9% positive, the rest neutral.

Method


Data source	Production message logs of ELIA Assistant
Period	1 April - 15 October 2026 (Europe/Bratislava time)
Published	16 October 2026
Sample	3,769 inbound customer messages from 1,213 distinct senders
Companies	Slovak companies using ELIA Assistant in production
Channels	Web chat (70%) and email (30%)
Excluded	Test messages, test accounts and ELIA's own website chat

Definitions. A message is one inbound customer message together with the AI's reply. A sender is a web chat visitor session or an email address; we use it as a proxy for one customer. Language, intent (topic) and sentiment are assigned to every message by the language model that answers it; a message can have more than one intent, so intent shares add up to more than 100%. Handover means the AI flagged the conversation for the human support team and opened a support case. Business hours are Monday to Friday, 08:00-16:59 local time.

Handover figures cover web chat only, from 16 June 2026 (2,295 messages from 509 visitors). Handover was not recorded before 15 June, and the email channel has no handover step.

Limitations - please read before citing.

The sample is weighted toward travel and tourism, where many customers are foreign visitors. That raises the share of foreign-language and off-hours messages compared with a business serving only local customers.
Almost no traffic is Czech (16 messages), so these figures describe Slovakia, not the Czech market.
Classification is done by a language model and was not hand-checked message by message.
"No handover" means the customer was not passed to a human. It does not prove the customer was satisfied.

All figures were produced by one set of read-only SQL queries over the message log; per-company data is not published.

When customers write

Time window	Share of messages
Monday-Friday, 08:00-17:00	54%
Outside those hours	46%
- of which weekend	17%
- of which 20:00-08:00	25%

The busiest hours were 13:00-15:00, followed by 10:00. Even so, close to half of all demand landed when a typical support team is offline.

This is the most practical number on the page: if your support is staffed 8 to 5, roughly every second customer question waits until the next working day - unless something answers it.

Which languages customers use

Language	Share of messages
English	50.8%
Slovak	35.2%
Arabic	7.2%
Hebrew	3.2%
Spanish	1.9%
Czech	0.4%
French	0.4%
7 other languages or unclassified	0.9%

Arabic messages came from 60 different senders and Hebrew from 36, so this is not one persistent customer. Businesses serving travellers receive a real share of questions in languages nobody on the team speaks. An AI assistant answers in the customer's language without extra configuration.

What customers ask about

Topic (intent)	Share of messages
Knowledge-base question (prices, conditions, how-to)	71.2%
Booking	13.2%
Change of details	9.8%
Price quotation	5.9%
Complaint	3.8%
Cancellation	1.0%
Feedback	0.5%
Other	14.5%

Seven out of ten messages ask something the company has already written down somewhere. That is the part of support an AI assistant can take over safely, because the answer comes from the company's own knowledge base.

How often a human was needed

On web chat, from 16 June to 15 October 2026:

21.5% of chat messages triggered a handover to the support team.
33% of chat visitors were handed over at least once; 67% were served by the AI alone.

Handover rate by topic:

Topic	Messages	Handed over
Knowledge-base question	1,403	11%
Booking	358	29%
Change of details	86	34%
Complaint	116	99%
Other	496	39%

Why the AI handed over:

Reason	Share of handovers
Complaint with negative sentiment	18%
Customer repeated the question without a satisfying answer	18%
Answer not in the knowledge base	11%
Customer asked for a human	9%
Other	44%

Two practical readings. First, knowledge-base questions are where AI works best - 89% were answered without a human. Second, only 9% of handovers happened because the customer asked for a person; far more came from the AI recognising it could not finish the job itself. That is the behaviour you want from a support bot: answer what it can, pass on what it cannot.

Customer sentiment

Sentiment	Share of messages
Neutral	86.8%
Positive	8.8%
Negative	4.4%

Most support traffic is plain and transactional. Negative messages are a small minority - and they are exactly the ones that should reach a person, which is how the handover rules are set.

What this means for a business in Slovakia or Czechia

Plan for the evening and the weekend. Nearly half of the questions arrive outside office hours. An AI first line closes that gap without night shifts.
Start from the knowledge base. Seven in ten questions are covered by documents you already have - price lists, conditions, FAQ. The better those are, the lower the handover rate.
Keep humans for complaints and actions. Complaints, changes and bookings are where people still add the most value. Good handover rules send them there with the conversation attached.
Expect languages you did not plan for. If you serve tourists, expats or cross-border customers, a meaningful share will not write in Slovak or Czech.

How to cite this data

ELIA Assistant, "AI Customer Support by the Numbers - What 3,769 Real Customer Messages Show", 16 October 2026, https://elia-assistant.com/blog/ai-customer-support-by-the-numbers-2026. Please mention the sample size and period when quoting individual figures.

We plan to update these figures as more companies go live. If you want to see how the same metrics would look for your own support, try ELIA Assistant free for 7 days.