When artificial intelligence answers and when it calls a person

Photo: Avila Ops Tecnologia
Artificial intelligence answers when the answer already exists in writing and is the same for everyone: deadlines, shipping, tracking, exchange policies, hours, coupon use, kit contents. It calls a person when the answer requires a decision, exception, or emotion: complaints, wrong orders, angry customers, discounts outside policy, anything involving refunds. Write this rule before turning on AI, not after the first problem.
On Black Friday, AI covers questions repeated 100 times a day, leaving people the ten that decide whether customers return.
What can AI answer alone?

Photo: Avila Ops Tecnologia
Anything documented that does not vary by customer. It reads the exchange policy, delivery table, and product list, responding immediately at any hour using the texts you supplied. Unlike a standard automatic reply, an AI agent understands questions phrased in different ways, including typos and transcribed audio, and answers in your store’s language rather than offering a menu.
It can also check orders when connected to the system: “Order 4821 was paid yesterday and ships today; tracking will arrive here.” This resolves November’s most common question without opening a dashboard. The WhatsApp AI agent guide shows when volume justifies it.
When must it call a person?
Whenever the right answer is not written. A starting list:
| Situation | Who answers |
|---|---|
| Deadlines, shipping, tracking, exchanges, hours, coupons | AI |
| Order status, with system connected | AI |
| Product suggestions within the catalog | AI, with limits |
| Wrong or broken product arrived | Person |
| Cancellation or refund request | Person |
| Customer mentions Procon, complaints, lawyer | Person, the same day |
| Discount or terms outside published rules | Person |
| Question repeated twice without resolution | Person |
| Customer requests a person | Person, immediately |
When handing off, AI says: “I’ll transfer you to Carla. She replies within two hours during business hours.” It then stops responding in that conversation until the person takes over.
How can I configure the rule without technical skills?
Write three lists on one page: what AI answers, what it never answers, and words requiring a person. Gather its sources: exchange policy, delivery table, FAQs, product descriptions. If a text is missing, AI will invent. The automatic message guide shows the tone these texts need.
Define its identity: the store’s automated assistant, not a person. Define data handling: it checks orders only after the customer provides the order number and matching name or CPF, and keeps conversations no longer than needed. The AI and LGPD guide covers this.
A clothing store turns on AI in October with the three lists. In week one, it answers 70 conversations about deadlines and exchanges and hands off eight: two wrong products, three refund requests, three customers requesting people. The owner reads all eight, corrects two wrong size answers, and by November AI handles most questions without anyone staying awake.
What to do now
Write the three lists today and check that source texts exist. Test for two weeks in October, reading every forwarded conversation, before leaving AI alone in November. We configure AI for repetitive answers and call you for decisions: avilaops.com/automacao-de-atendimento or WhatsApp: wa.me/5517997811471.
Frequently asked questions
Can WhatsApp AI give discounts?
Only published ones, such as a campaign coupon. Negotiated discounts, exceptions, or special terms require a person. Include this in its never-answer list.
Must customers know they are speaking to AI?
Yes. Beyond being right, it prevents frustration: people who know it is an automated assistant request a person when needed instead of fighting the machine.
How many daily conversations justify service AI?
When repetitive questions take more than an hour of someone’s day, or arrive after hours and remain unanswered until the next day. Below that, WhatsApp Business quick replies suffice.
Turn what you learned into your next step.
Tell us about your company and what you want to improve. We can explore the right tools, integrations or custom project together.
