Prompt templates

HumanFirst provides pre-written prompt templates to help you get started in your NLG workflows. These templates are built from some common workflows we've seen over the years.

Feel free to copy and adapt these to your use case.

Call summarization

Below is an online chat conversation between a "User" and a customer service "Agent".

/// START CONVERSATION

{{ conversation }}

/// END CONVERSATION

What is the key issue causing the client to contact the fashion retailer?

Please give your answer as a single sentence without a bullet or a number.

Topic modeling

Below are descriptions of key issues as reported by customers to a customer support agent.

/// START DESCRIPTIONS

{{ text }}

/// END DESCRIPTIONS

Respect the following rules:

1. Chatbot intents are action based for example - "Renew subscription"

2. Chatbot intents cover only a single action

3. Chatbot intents are action focused

4. The provided summary might contain more than one intent. For example "I need to check on my shipping and renew subscription" should be separated to "check shipping" and "renew subscription"

5. Each intent should be represented only once, look for duplicates and remove them

Thinking step by step, review these summaries and provide me with a list of chatbot intents according to the rules above that that would cover these key issues.

Call rating

Below is a chatbot transcript from a call-center providing support to existing and potential customers.

The conversation is between a "User" and an "Agent" who is helping resolve the issue.

/// START TRANSCRIPt

{{ conversation }}

/// END TRANSCRIPT

Please act as an impartial judge and evaluate the quality of the conversation. Think through the problem step by step and evaluate your reasoning

Below is your grading rubric:

- Criteria: Asking for information previously provided

- Score 0: Agent repeatedly asked for the same information. Examples include asking for the order number multiple times without addressing the actual concern.

- Score 1: Agent does not ask for information that has already been provided.

- Criteria: Acknowledging customer issue

- Score 0: Agent fails to understand or acknowledge the customer issue. Examples include failing to repeat back to the customer the issue.

- Score 1: Agent clearly acknowledged the customer's issue in their own words. Examples include summarizing the problem to confirm understanding before proceeding.

- Criteria: Resolution Efficiency

- Score 0: Agent did not resolve the issue within the call or provided a vague timeline for resolution.

- Score 1: Agent resolved the issue during the call or provided a clear and reasonable timeline for resolution. Examples include immediate issue resolution or stating a specific time frame for follow-up.

- Criteria: Empathy and Understanding

- Score 0: Agent showed no empathy towards the customer's situation or was dismissive of their concerns. Examples include ignoring customer frustrations or responding rudely.

- Score 1: Agent demonstrated empathy and understanding of the customer's situation. Examples include acknowledging the inconvenience caused and expressing a willingness to help.

- Criteria: Follow-Up Actions

- Score 0: The agent does not mention or offer any follow-up actions when needed, leaving the customer unsure of the next steps. Example: Failing to confirm that a return process has been initiated.

- Score 1: The agent clearly outlines any necessary follow-up actions or confirms that they will be taken, providing the customer with reassurance about what to expect next. Example: Confirming an email will be sent with return shipping labels.

Total the scores

Do not write down your reasoning, only provide your scores for each criteria and the total score.