
AI in e-commerce: customer service is the #1 use for 61% of companies
In online retail, artificial intelligence has found its first area of use: customer relations. According to a OnePoll survey for Storyblok conducted in the United States in May 2025, 61% of e-commerce companies use AI for customer service.
It is the most cited use, just ahead of marketing analysis. But the figure raises a question few companies are asking yet: if AI increasingly answers customers, who analyzes what customers say back?
The ranking of uses
Customer service comes first with 61% of surveyed companies, followed very closely by marketing analysis (60%). Far behind are the automation of administrative tasks (42%), translation services (41%) and content creation (40%).
The top two uses therefore account for most of the adoption. They share one thing: both concern the relationship with the customer, one to answer them, the other to understand and target them better.
The figure is all the more notable because content creation, often presented as the flagship use of generative AI, only comes fifth. Companies first looked for a measurable gain on a major cost item: handling customer requests.
Adoption is still uneven
These figures do not mean AI is mastered everywhere. The same Statista dossier on AI in US e-commerce shows that in 2024, 47% of consumer goods and retail companies were still in pilot mode with generative AI, experimenting without using it in production.
Only 5% said they used it fully across the organization. Between announcing a use and rolling it out, there is still a long way to go. Customer service is often the first project because it is visible and volumes are high.
Another 20% were in early production, developing usable insights and planning broader use cases. In other words, roughly three quarters of these companies were somewhere between experimenting and the first real deployments in 2024.
AI answers. Who listens?
Chatbots, suggested replies for agents, automatic routing of requests: most AI uses in customer service are about answering faster. That is useful, but it only covers half of the conversation.
The other half is customers' own messages: their questions, their complaints, their comments after talking to an automated assistant. These messages contain the real reasons for satisfaction and dissatisfaction. They also flag the cases where automation failed and the customer left annoyed.
A company that automates its answers without analyzing that feedback is flying blind. It saves time, but it does not know whether it is losing customers along the way.
Take a simple example. An automated assistant handles a refund request. Technically, the conversation is closed. But the customer then writes a review: "automated answer unrelated to my question, I had to call twice". Without analyzing that feedback, the dashboard shows a success; the real experience says the opposite.
Closing the loop
Automatic language analysis closes that loop. It reads incoming messages, reviews posted after a purchase and call transcripts, then surfaces recurring topics, dominant emotions and signs that customers may leave.
The result serves customer service, which knows where to focus its efforts, and marketing, the second AI use in the survey, which gets a clearer view of what customers expect. Retailers can find examples on our retail and e-commerce page.
The analysis relies on models such as Mistral, BERT, Llama, GPT or Claude depending on the case. Your customers' data is hosted in France and never used to train models.
Questions worth asking
Before adding yet another AI tool, three questions deserve an answer. What share of our customer conversations is actually read? Do we know which topics cause the most frustration this month? Do we measure the effect of our automated answers on satisfaction?
If the answer is no, the most profitable investment may not be answering faster, but listening better.
These questions apply to every company, whatever its size. A small online store with a few hundred messages a month gains as much from understanding its customers as a large retailer, and often acts faster on what it learns.
Listening better also makes automation better: knowing where customers get stuck tells you which automated answers to rewrite first.
Sources
- OnePoll, Storyblok, via Statista (2026). Leading uses of artificial intelligence (AI) in e-commerce in the United States in 2025. https://www.statista.com/statistics/1617307/united-states-ai-in-e-commerce/
- Statista (2026). Artificial intelligence (AI) in U.S. e-commerce — dossier.