Hand the chat to a person before the bot loses the customer

Fields

• Chat so far, latest message last

Questions

1. Does the customer ask to speak to a person, a manager or someone by phone? Yes / No

   Read every customer message. Count a request, in any words, for a person, an agent, a manager, a callback or a phone call instead of the assistant, an automated reply or more messages, and a request to stop talking to a bot. Mentioning a person they spoke to before is not a request. Yes: The customer asks, in any words, to deal with a person. No: The customer does not ask for a person, including when they are only frustrated.

2. How well is the assistant handling the customer's request so far? Choice

   Read the whole chat. Judge the assistant's answers against what the customer is trying to get done. If more than one option fits, choose the one lowest in the list.

   • Too early to tell The assistant has answered at most once, so there is not yet a pattern to judge.
   • On track Each answer moves the customer toward what they want, and the customer follows it.
   • Misunderstood The assistant answers a different question from the one the customer asked.
   • Cannot do it The request needs an action, exception or decision the policy does not let the assistant take.
   • Looping The assistant repeats an answer or step the customer has already tried or rejected, or asks again for a detail already given.

3. How frustrated is the customer in their latest messages? Scale

   Read the customer's last two or three messages. Rate what they write about the assistant, the service or the wait. Capitals and punctuation count only together with the words.

   • None Calm and matter of fact.
   • Mild Impatient wording, such as "still not working" or "I already said".
   • Clear Complains directly about the assistant, the answers, the wait or the service.
   • Severe Anger, shouting in capitals, insults, or saying the service is the worst they have had.

4. Does the customer express dissatisfaction that the policy counts as a complaint? Yes / No

   Read every customer message and the policy in the context. The policy counts any expression of dissatisfaction about the service, the staff or how a problem has been handled as a complaint, whether or not the word complaint is used. Yes: The customer expresses dissatisfaction of that kind at least once. No: The customer asks questions or reports a fault without expressing dissatisfaction with the provider. A plain fault report, such as "my internet is down", is not a complaint on its own.

5. Does the customer mention circumstances that may mean they need extra support? Yes / No

   Use the circumstances listed in the context. Count only what the customer says about themselves or someone in their household. Do not infer anything from tone, writing style or spelling. Yes: The customer mentions health problems, disability, bereavement, old age, money trouble, a language or literacy barrier, difficulty using the chat, or similar circumstances. No: The customer mentions none of these.

6. Does the customer dispute a charge, report a billing error, ask for their personal data, or threaten legal action? Yes / No

   Read every customer message. Yes: The customer disputes a charge, reports an error on a bill or payment, asks for a copy of their personal data or for it to be deleted, or says they will take legal action. No: The customer does none of these things. Asking what a charge is for, without disputing it, is No.

7. Does the customer report a safety incident with equipment? Yes / No

   Read every customer message. Yes: The customer reports smoke, a burning smell, sparks, overheating or an electric shock from equipment. No: The customer reports no such incident. A fault light or a device that will not turn on is No.

8. Does the chat raise a topic on the policy's list of reserved topics for a person? Yes / No

   Read the whole chat and the list of reserved topics in the context: compensation, a credit or money off for lost service; any refund; moving home; a death on the account; a claim that someone used the account without permission. Yes: The customer raises at least one topic on that list. No: The customer raises none of them. Complaints, disclosures, disputes and safety incidents are asked about separately and do not count here.

9. Does the customer say they will cancel, leave for another provider, or post about it publicly? Yes / No

   Read every customer message. Yes: The customer says they will cancel, leave, switch provider or post about their experience in public, including as a condition ("if this is not fixed today"). No: The customer says none of these things.

10. Has the customer confirmed that their problem is solved? Yes / No

    Read the customer's messages, especially the last ones. Yes: The customer says the problem is fixed or they have what they need, such as "that worked, thanks". No: The customer has not confirmed it. The assistant saying the problem is solved does not count.

11. What should happen next in the chat, under the hand-off policy? Choice

    Read the whole chat and the policy in the context. Choose what should happen next. If more than one option fits, choose the one lowest in the list.

    • Close the chat The customer has confirmed the problem is solved and raises nothing the policy sends to a person.
    • Let the assistant continue The assistant is making progress and nothing in the chat is for a person.
    • Offer a person The assistant is struggling or the customer is frustrated, but nothing in the chat requires a person yet. Ask the customer if they would like one.
    • Hand over to a person now The customer asked for a person, raised a matter the policy sends to a person, or needs something the assistant cannot do. Transfer with the reason.

Lens columns

asks_for_person, asks_for_person_probability, assistant_progress, assistant_progress_probability, frustration, frustration_average, complaint_expressed, complaint_expressed_probability, vulnerability_signal, vulnerability_signal_probability, dispute_or_right_raised, dispute_or_right_raised_probability, safety_incident, safety_incident_probability, reserved_topic, reserved_topic_probability, leave_threat, leave_threat_probability, resolved, resolved_probability, next_step, next_step_probability