VoraPeople. Work. Intelligence.
Access early

Welcome to your
next working environment.

Your agency.
One Vora.

Find clients. Build AI agents. Run your agency.
AI chatbot for customer service: the exception it cannot grant

AI chatbot for customer service: the exception it cannot grant

A website chatbot can explain a rule but not waive it. See which questions to answer, which to hand off and how to check the handoff arrived.

Which questions a chatbot may answer and which it must hand off

A customer service chatbot should answer questions that have a single approved answer and hand off questions that need a decision. Settle which questions the chatbot may answer by itself, which facts have to be looked up live and when a person has to step in.

Do this for chat on its own. A website chatbot has no voice to soften a refusal, so the wording of every boundary matters more than it does on the phone.

Start with a list of the twenty questions your team answers most often. Mark each one as answer, look up or hand off. The marked list becomes both the chatbot brief and the first test set.

Write the refusal wording for each handoff group, too. "I can't waive that, but I'll pass your request to the team" keeps the customer informed without promising an outcome the chatbot cannot deliver.

A hypothetical exception: the deposit waiver

Hypothetical example, not a customer result. A boat-hire company requires a deposit. A customer writes, 'I'm a regular, can you waive the deposit?' The chatbot finds the deposit policy in its approved sources.

Locating a policy gives the chatbot no authority to make an exception; it should explain the rule and send the request to whoever is responsible. So the chatbot states the deposit rule, says a team member decides on waivers, and passes the request along with the customer's name, the booking and the question asked.

The wording matters as much as the routing. A reply that says the deposit is fixed and then goes silent leaves the customer guessing, while one that names the next step keeps them informed. Write that reply into the chatbot brief and check it in the test set.

Handoff contents and checking receipt

A handoff should name the right client and carry the customer's question, the details gathered so far, the decision still open and the next owner. In the deposit example that means the booking, the waiver request and the manager who decides.

Do not stop at 'the bot sent it'. A sent alert, an arrived case and a finished task are three different observations. Open the handoff as the receiving person and confirm they can act without asking the customer to repeat anything.

Then track how long unresolved handoffs wait. A handoff that arrives correctly but sits unread for two days is still a failed answer from the customer's side.

A test set before launch

Before launch, run a test set that includes a deposit-style request, which belongs in the restricted group. The full test set is described on our AI chatbot for your website guide.

Record what the chatbot said, what it handed off and what the receiving person saw, then repeat the set after any change to a policy or a boundary. A chatbot that passed last month can fail after one policy edit, and the deposit-style request is usually the first to show it.

Running this for clients?

For a customer-service pilot, Vora's native inbox provides a place to inspect website-chat work handed to staff. Rehearse the deposit-exception conversation and check whether the receiving employee sees exactly which decision remains theirs.

AI chatbot for customer support questions

What can an AI chatbot for customer service answer on its own?

Questions with one approved, current answer from a source that has an owner, such as hours, service descriptions or published rules.

Why can't the chatbot just grant the exception?

Locating a policy gives the chatbot no authority to make an exception. It should state the rule and send the request to whoever is responsible.

What should the handoff include?

The right client, the customer's question, the details gathered so far, the open decision and the next owner.

How do I test an AI chatbot for customer support?

Run the same approved question set, include a deposit-style request, and check the handoff from the receiving person's side. The full test set is in our AI chatbot for your website guide.

How often should I rerun the chatbot test set?

After every change to a policy, a boundary or an approved answer, and on a regular schedule even without changes. Compare each run with the last one.

Made for the people behind the work.Take a look around

Find your next stop.

Pages, products and the journal. All in one place.

Make yourself at home.

A different view, whenever you feel like it.

A small discovery.