AI customer support tool for e-commerce
An artificial intelligence agent is only useful if it knows what it doesn't know. The difference between a good tool and a frustrating one lies exactly there.
The promise of an AI customer support tool is simple to describe and difficult to deliver well: responding quickly, with correct information, and knowing how to recognize when a human needs to intervene. Many tools fail in one of these three points — usually the last one, responding with confidence to questions they actually cannot resolve correctly.
How the WhatSMS AI agent works
The artificial intelligence agent consults real sources before responding — the store's product catalog, order status, FAQ documents uploaded by the team, or the content of the site itself. This means that the answer to "is this product available in size M?" reflects actual stock, not a generic guess.
Support for PDFs, catalogs, and the store URL
Configuring the agent's knowledge does not require rewriting existing information — you can directly upload a PDF with return policies, the product catalog, or simply provide the website URL so the agent learns from the already published content.
Seamless handoff to the team
When a question falls outside what the agent can answer with certainty — a complaint, an ambiguous situation, a specific request not covered by the knowledge base — the conversation is passed to a human, with an automatic summary of everything already discussed, so the team doesn't have to reread the entire conversation from scratch.
Transparency with the customer
The customer is always informed that they are interacting with an AI agent — there is never an attempt to pass off automation as a person, which avoids the frustration of discovering later that "Ana," who responded, was actually an automatic system.
How to test this in your store
What happens when information is incomplete
If the uploaded knowledge base does not cover a particular question, the agent recognizes this gap and escalates to the team, instead of trying to fill the void with an unreliable answer. This is perhaps the most important feature of an AI customer support tool: knowing how to distinguish between what it knows with confidence and what requires human confirmation. This is also why many stores manage to reduce the volume of repetitive tickets — the agent handles the predictable, the team handles what requires judgment.
Updating the knowledge base regularly
An outdated knowledge base is worse than none at all — if the store's policies change, it is worth ensuring that the information used by the agent always reflects the most recent version.
The initial setup usually takes a few hours, depending on the amount of information to be uploaded. You can create a free account and test how the agent responds to your customers' real questions, adjusting the knowledge sources as you identify gaps.
Frequently asked questions
Does the AI agent completely replace the customer support team?
No. The agent handles repetitive questions supported by real sources — catalog, orders, FAQ — and escalates to a human, with an automatic summary, whenever the situation requires judgment or falls outside the uploaded knowledge base.
Which artificial intelligence models does the agent support?
WhatSMS supports GPT-4, Claude, or internally hosted (self-hosted) Ollama models, depending on the store's preference for cost, performance, or data privacy.
Do I need to rewrite the store's catalog for the agent to understand it?
No. You can upload the catalog PDF directly, existing FAQ documents, or simply provide the website URL, and the agent learns from that already published content.
Does the customer know they are talking to an automatic agent?
Yes, always. Transparency is mandatory — the agent never pretends to be a person, which prevents the frustration of the customer finding out later that they were talking to an automatic system.
How long does it take to set up the agent's knowledge base?
Usually a few hours, depending on the amount of material to upload — policy PDFs, product catalog, existing FAQs, or the website URL.