Original ecommerce chatbot statistics from 16,908 real AI chat conversations on 11 Baltic online stores: what shoppers ask, when they ask, where the chatbot opens, and what AI resolves on its own.
Most shoppers who open an ecommerce chatbot are not complaining. They are about to buy and need one answer first. That is the clearest finding from 16,908 real conversations handled by Parnidia AI-powered chatbots on 11 Baltic online stores between late January and October 2026.
Most chatbot statistics online recycle the same surveys and market forecasts. These ecommerce chatbot statistics come straight from chatbot usage logs: what shoppers actually asked, when, where on the site, and whether the AI solved it.
The stores cover 11 verticals: consumer electronics, furniture, power tools, decking and building materials, home and beauty, sportswear, outdoor apparel, orthopedic goods, pharmacy, agricultural supplies and party costumes. Together their shoppers sent 44,948 messages and received 75,191 AI replies.
Source: Parnidia AI chat data, 11 Baltic online stores, Jan to Oct 2026, 16,908 conversations
74.5% of conversations are about a purchase that has not happened yet. Only 23.0% touch an existing order, and 17.5% are purely post-purchase. The common belief that an ecommerce chatbot is mainly an order tracking tool for "where is my order" questions covers less than a quarter of real chatbot usage.
3 in 4 chatbot conversations come before the purchase
Share of 15,749 topic-tagged conversations. One chat can cover several topics.
The biggest single bucket is product details: dimensions, materials, compatibility and how to use the item. Add discovery questions like "what do you recommend for oily skin?" and you have the core job of AI chatbots in e-commerce: acting as a knowledgeable salesperson, not only a customer support desk.
46.2% of conversations start outside Monday to Friday, 08:00 to 18:00. A customer support team that answers only during office hours leaves almost every second question waiting until the next morning, or the next week. Real time support matters even more for stores selling into other time zones.
Shoppers keep asking after your team logs off
Conversations started per hour of day, Vilnius time, all days.
The evening is not a quiet tail. At 21:00 shoppers still start 65% as many chats as at the 11:00 peak, and 15.8% of all conversations begin between 20:00 and midnight. Weekends add another 21.1%, with Saturday and Sunday running about a third below a typical weekday.
Chats peak early in the week and dip on Friday
Share of conversations by day of the week, Vilnius time.
Tuesday is the busiest day, with 17.2% of all chats, closely followed by Monday at 16.6%. Volume eases through the week and drops noticeably on Friday (13.5%). Sunday is slightly busier than Saturday, and it is the most evening-heavy day of the week: 32.5% of Sunday chats start after 18:00, compared with 22% to 28% on weekdays. Friday evening is the quietest slot of the working week.
After-hours shoppers are also more likely to be mid-purchase: 77.0% of their chats are about buying decisions, versus 72.3% during business hours. These are exactly the questions that cost a sale when they wait overnight.
55.3% of conversations with a known starting page begin on a product or category page. Shoppers are looking at a specific item and need one detail before they add it to the cart.
| Page where the chat started | Share of chats |
|---|---|
| Product or category page | 55.3% |
| Cart, checkout or order pages | 21.1% |
| Homepage | 11.8% |
| Delivery, returns, contact or FAQ pages | 9.4% |
| Search results | 2.5% |
Based on 12,997 conversations where the page URL was recorded.
One in five chats starts in the cart, at checkout or in the "my orders" area. That is the most expensive moment to leave a question unanswered, because the shopper has already decided to buy.
Suggested buttons matter too. 16.1% of shoppers opened with a one-tap button instead of typing. The top three: order status (37.9% of button starts), delivery times (14.1%) and size help (12.4%). For placement tactics that lift usage, see 4 real examples of AI chat agents increasing e-commerce engagement.
The median opening message is 43 characters long, and only 3.0% of shoppers open with a bare greeting. People skip the small talk. But how they phrase that first message varies a lot, and it predicts how well the AI will do.
Half of shoppers ask a full question, one in eight type a search query
How shoppers phrase their first typed message, share of 16,033 conversations.
Rule-based bots and keyword search boxes break on this kind of input. Conversational AI built on natural language processing understands meaning, not exact words, and can ask a clarifying question when a shopper types only "tiles".
The AI chatbots fully resolved 53.7% of conversations and at least partly resolved 80.7%. 9.5% ended as a clear failure, and only 4.0% were flagged as needing a human agent. Each conversation was scored by an automated evaluator that reads the whole transcript and judges whether the shopper got what they came for.
AI closes logistics questions best, product specs worst
Outcome per topic, % of conversations, scored by an automated AI evaluator. The number shows fully plus partly resolved.
Logistics questions are the easy win for chatbots: delivery, returns and checkout are fully or partly resolved in about 9 out of 10 chats, and order status fails in only 2.0%. These answers live in policies and order systems that an AI can read.
Product questions are harder. 14.0% of product-detail chats fail, the highest rate of any topic. Across the 5,798 individual questions the AI could not answer, the gaps were consistent:
None of these are AI problems. They are data problems. An AI agent can only be as good as the product feed and knowledge base behind it. Here is how to prevent AI chatbot hallucinations and wrong answers when data is missing.
You might expect long conversations to signal trouble. The data says otherwise: the longer a chatbot conversation runs, the more likely the shopper is browsing products with the AI.
| Shopper messages | Chats | Got product cards | Positive tone | Negative tone |
|---|---|---|---|---|
| 1 | 6,590 | 6.0% | 0.8% | 0.6% |
| 2 | 3,978 | 10.2% | 5.3% | 2.2% |
| 3 to 4 | 3,589 | 15.9% | 10.4% | 5.4% |
| 5 to 7 | 1,425 | 22.7% | 15.2% | 12.2% |
| 8 or more | 578 | 32.7% | 19.2% | 17.8% |
Based on 16,160 evaluated conversations.
In chats with 8 or more messages, one in three included product recommendation cards, and positive tone rose from under 1% to 19.2%. Negative tone rises too, so long chats are where an AI agent wins or loses the shopper.
Most conversations are short, though. 40.3% end after a single question and answer, and the median chat takes 2 shopper messages and about 1.5 minutes. Median chatbot response time is 6 seconds to the first reply, and 90% of first replies arrive within 16 seconds. Tone, a rough proxy for customer satisfaction, is calm overall: 90.3% of chats are neutral, 6.0% positive and 3.7% negative.
Tone is set by the outcome, not by the opening line. Negative chats are rare, and they follow a clear pattern:
Shoppers in Latvia and Estonia ask about order status almost twice as often as shoppers in Lithuania: 23% of chats versus 14%. To compare like with like, we looked at one footwear and apparel group that runs the same brands, the same online stores and the same AI chatbot in all three countries: 5,026 chats in Lithuania, 2,036 in Latvia and 893 in Estonia.
Latvian and Estonian shoppers ask about orders, Lithuanians about products
Share of chats by topic, same retail group, same brands and AI in all three countries.
Lithuanian shoppers use the chatbot to decide what to buy. They ask more about sizes and fit (22.1%), stock (18.0%), product recommendations (14.8%) and physical stores (13.1%). Latvian and Estonian shoppers lean on it for after-sales customer support: order status, delivery and, in Latvia, returns and defects (30.1%). One likely reason is cross-border shipping, which means longer delivery times and more "where is my parcel?" questions, but the data alone cannot prove the cause.
Some things do not change with the border. Questions about promotions and discount codes take the same share everywhere, 14% to 15% of chats. How well the AI chatbot copes does vary:
| Outcome | Lithuania | Latvia | Estonia |
|---|---|---|---|
| Fully resolved | 69.1% | 71.0% | 63.7% |
| Failed | 6.9% | 7.1% | 8.6% |
| At least one unanswered question | 21.5% | 19.6% | 24.6% |
| Negative tone | 1.8% | 3.2% | 2.5% |
Based on 7,955 evaluated chats from one Baltic footwear and apparel group, Apr to Oct 2026.
Estonia has the lowest resolution rate and the most unanswered questions, a sign that the Estonian knowledge base needs the most work. The lesson for stores selling across the Baltics: do not copy one chatbot setup into every country. Each market needs its own delivery, order tracking and return answers in its own language.
Want to see what these chatbot statistics would look like on your store? Get a demo of the Parnidia AI Chat Agent. Still comparing tools? Read our review of the best AI chat agents for e-commerce.
Mostly product questions. In 16,908 conversations on 11 Baltic online stores, 38.5% of topic-tagged chats were about product details such as specs, dimensions and usage, and 23.2% were product recommendations. Order status came third at 16.4%.
All day and well into the evening. 46.2% of conversations started outside Monday to Friday, 08:00 to 18:00, and 15.8% started between 20:00 and midnight. Chat volume peaks around 11:00 and 14:00.
In this dataset the AI fully resolved 53.7% of conversations and fully or partly resolved 80.7%. Only 4.0% were flagged as needing a human agent. Delivery, returns and checkout questions were resolved or partly resolved in about 9 out of 10 chats.
Mostly on product pages. 55.3% of chats with a known starting page began on a product or category page, 21.1% in the cart, checkout or order pages, and 11.8% on the homepage.
Yes. In one retail group running the same stores and AI chatbot in all three countries, order status questions made up 23% of chats in Latvia and Estonia versus 14% in Lithuania. Lithuanian shoppers asked more about sizes, stock and product recommendations. Questions about promotions took the same share, about 14% to 15%, in every country.
Usually because the data is missing, not because of the AI. The most common unanswered questions were about specs and fit (19.5%), delivery timing for a specific item (15.2%), specific orders (13.7%), prices and discounts (11.7%) and restocks (9.1%). Complete product attributes improve answers more than prompt changes.
Leo Mazzura
Co-founder & CEO @ Parnidia
Book a demo. We’ll show examples from similar clients and walk you through the software.
