Viber Queue

Queue Field Details

Open more settings using the Extended button in the top right corner.

AI Topics Best Practices

Cheat Sheet Summary:

• Be clear, concise, and specific.

• Focus on one task at a time.

• Use English with proper grammar.

• Provide examples for guidance.

• Account for common errors.

• Specify relevant sections of content.

• Iterate and test for improvement.

• Use contextual details wisely.

Tip #1: Be Clear and Specific

Motivation: Ensuring your prompt is direct and unambiguous helps the AI produce accurate, focused results.

Do’s:

• Use concise language and simple sentence structures.

• Break down complex requests into smaller, logical parts.

Don’ts:

• Don’t mix multiple concepts or questions into one prompt.

• Don’t include irrelevant details or filler words.

Example:

• Yes: “Summarize this text in two sentences.”

• No: “I’d like a summary, but make it short and also explain why it’s important.”

Tip #2: Ask for One Thing at a Time

Motivation: Limiting your request to a single objective prevents confusion and improves the AI’s precision.

Do’s:

• Separate multiple requests into individual prompts.

• Keep each prompt focused on one clear outcome.

Don’ts:

• Don’t cram multiple, unrelated tasks into one prompt.

• Don’t expect the AI to infer which part of the request to prioritize.

Example:

• Yes: “Translate this sentence into French.”

• No: “Translate this sentence and explain the grammar rules.”

Tip #3: Use English for Clarity

Motivation: English is commonly used in training data, increasing the likelihood of well-understood prompts.

Do’s:

• Write in standard English with proper grammar.

• Use clear vocabulary and avoid idiomatic expressions.

Don’ts:

• Don’t use foreign languages in the prompt unless necessary.

• Don’t rely on slang or highly localized terms.

Example:

• Yes: “Summarize this paragraph.”

• No: “Can u do a lil recap of this, like quickly?”

Tip #4: Provide Examples

Motivation: Examples guide the AI by showing what you want and what you don’t.

Do’s:

• Include a “Yes” example demonstrating the desired output.

• Include a “No” example showing what to avoid.

Don’ts:

• Don’t assume the AI automatically knows the correct style.

• Don’t leave examples too vague or contradictory.

Example:

• Yes: “Correct this sentence: ‘She don’t like apples.’ Output: ‘She doesn’t like apples.’”

• No: “Fix the grammar issues here.”

Tip #5: Account for Speech-to-Text Errors

Motivation: Transcripts may contain mistakes, so being flexible with wording helps capture the right meanings.

Do’s:

• Mention approximate or similar words you’d accept, combined with examples.

• Consider phonetic variations or common transcription errors (e.g., “their” vs. “they’re”).

Don’ts:

• Don’t rely solely on exact keyword matches.

• Don’t ignore the possibility that words might be misspelled.

Example:

• Yes: “Look for phrases like ‘in summary’ or similar wording.”

• No: “Only consider the exact phrase ‘to sum up.’”

Tip #6: Focus on Specific Segments of the Conversation

Motivation: Pinpointing a part of the dialogue (e.g., the greeting) helps the AI concentrate on the most relevant content.

Do’s:

• Indicate a precise section (e.g., “Look at the first 30 seconds”).

• Highlight a particular event or phase (e.g., “After identity verification”).

Don’ts:

• Don’t ask the AI to review the entire call if only a snippet matters.

• Don’t expect the AI to guess which part you care about.

Example:

• Yes: “Summarize the customer’s issue mentioned after the agent’s greeting.”

• No: “Summarize the conversation.”

Tip #7: Iterate and Refine

Motivation: Prompt quality improves over time as you learn what works and what doesn’t.

Do’s:

• Adjust phrasing based on previous results.

• Keep track of changes and their effects on output quality.

Don’ts:

• Don’t give up after the first attempt.

• Don’t ignore patterns in the feedback you receive.

Example:

• Yes: “Simplify this sentence: ‘The phenomenon can be explained in the following manner.’”

• No: “Make this easier to read.”

Tip #8: Don’t Be Afraid to A/B Test

Motivation: Experimenting with different prompt variations can reveal which format yields the best results.

Do’s:

• Test multiple prompt structures and compare outcomes.

• Record results to identify winning formats.

• Test both small adjustments (e.g., phrasing) and drastically different structures.

Don’ts:

• Don’t assume the first prompt is optimal.

• Don’t forget to measure performance over several trials.

Example:

• Test 1: “Explain why the sky is blue in simple terms.”

• Test 2: “Why does the sky appear blue? Use plain language.”

Tip #9: Use Contextual Details Wisely

Motivation: Including relevant background info can help the AI better understand the scenario.

Do’s:

• Provide necessary context (e.g., role of speakers, call purpose).

• Mention pertinent details that guide interpretation.

Don’ts:

• Don’t overload the prompt with irrelevant background.

• Don’t omit crucial information that the AI needs to perform well.

Example:

• Yes: “Rewrite this sentence in formal tone: ‘Hey, can you help me out with this issue?’ Context: Email to a manager.”

• No: “Rewrite this sentence in a formal tone.”

Additional Tips:

• Tone Matching: The tone of your prompt (formal, casual, instructive) can influence the output. Be explicit about the desired tone when relevant.

• Example: “Explain this to a 10-year-old” vs. “Explain this to a professional audience.”

Distribution Strategy

The distribution strategy you select determines how activities that use the queue will be allocated to your agents.

Each strategy takes into account agents' priority (also know as penalty) in the queue depending on the rights they use (0 – highest priority, 10 – lowest priority). You can set up the priorities in each queue by clicking Change in the Agents column.

Examples below illustrate how each distribution strategy distributes activities to agents. We will work with 6 agents and their priorities in the queue:

• Rachel and Paul – priority 0 – the most competent to deal with activities in the queue

• Tom and Sophie – priority 5 – can help out if Rachel and Paul are busy

• Julia and Terry – priority 9 – should only answer activities in the queue if there is no other option

Everyone

1. This strategy first distributes the activity to all the users with the highest priority at once. If they are available, it will ring there until Wait time runs out or all agents reject the activity.

2. If there are no agents with the highest priority available or if no one in the group answers the activity, it will be distributed to all users of the next priority group.

3. If there are no agents available or if the activity is rejected or not answered before Wait time runs out, it will distributed to all agents

   1. with higher priority if they become available while it is ringing.

   2. with lower priority.

Example 1:

1. All agents are available.

2. An activity comes in via the queue and starts ringing for Rachel and Paul. Neither of them answer and Wait time runs out.

3. The activity starts ringing for Tom and Sophie as they have the next priority. They both reject the activity.

4. The activity goes back to Rachel and Paul as they are available and have the highest priority.

Example 2:

1. Everyone but Julia is chatting to customers.

2. An activity comes in via the queue and starts ringing for Julia. As it's ringing, Tom's and Sophie's activities end. Julia doesn't answer and Wait time runs out.

3. The activity starts ringing for Tom and Sophie. Meanwhile, Rachel's activity ends. Tom rejects the activity, Sophie doesn't answer and Wait time runs out.

4. The activity starts ringing for Rachel as she has the highest priority.

Everyone in Priority Order

1. This strategy first distributes the activity to all the users with the highest priority at once. If they are available, it will ring there until Wait time runs out or all agents reject the activity.

2. If there are no agents with the highest priority available or if no one in the group answers the activity, it will be distributed to all users of the next priority group.

3. If there are no agents with the next priority available or if no one in the group answers the activity, it will be distributed to all users of the next priority group in line, regardless of any agents that have become available in higher priorities.

4. If the activity is not answered even in the group with the lowest priority, distribution will start again from step 1.

Example 1:

1. All agents are available.

2. An activity comes in via the queue and starts ringing for Rachel and Paul. Neither of them answer and Wait time runs out.

3. The activity starts ringing for Tom and Sophie as they have the next priority. They reject the activity.

4. The activity starts ringing for Julia and Terry as they have the next priority. They reject the activity.

5. The activity goes back to Rachel and Paul – the distribution starts again from the beginning.

Example 2:

1. Everyone but Julia is chatting to customers.

2. An activity comes in via the queue and starts ringing for Julia. As it's ringing, Tom's and Sophie's activities end. Julia doesn't answer and Wait time runs out.

3. Because Julia is in the lowest priority, the activity goes back to the beginning, but Rachel and Paul are busy, so it starts ringing for Tom and Sophie. They don't answer and Wait time runs out.

4. The activity goes back to Julia as she is the only agent available in the next priority group.

Least Recent

Least recent call distribution: the time of the last call is calculated based on activities in the last 2 hours.

By default, the Least recent distribution strategy takes into account All activity types except Custom.

If you want it to only take into account Only same activity type as this queue or All activity types, go to Global Settings, and change the Least recent – check by activity types field.

1. This strategy first distributes the activity to an available agent with the highest priority.

2. If there are several agents that have the highest available priority, the one with the longest time since closing their last activity will be selected. If that is the same as well, the agent that has been logged in to the queue longest will be selected.

3. If there are no agents available or if the activity is rejected or not answered before Wait time runs out, it will go to another agent

   1. with higher priority if they become available while the activity is ringing.

   2. with the same priority.

   3. with lower priority if there are no agents with the same or higher priority available.

If an agent is the last available one with the highest priority and they reject or don't answer an activity, it will be distributed to them again straight away (unless Automatic pause is turned on).

Example 1:

1. All agents are available.

2. An activity comes in via the queue and starts ringing for Rachel. She doesn't answer and Wait time runs out.

3. The activity goes to Paul. He doesn't answer and Wait time runs out.

4. The activity goes back to Rachel. She doesn't answer and Wait time runs out. 

5. The activity goes back to Paul.

6. Unless both Paul and Rachel Go Unready or start a pause, the activity will never go to a lower priority.

Example 2:

1. Everyone but Julia is chatting to customers. Automatic pause is turned on in the queue, meaning that if an agent doesn't answer an activity, they will be put on an unpaid pause.

2. An activity comes in via the queue and starts ringing for Julia. As it's ringing, Tom's, Sophie's and Rachel's activities end (in that order). Julia doesn't answer and Wait time runs out – she is put on an unpaid pause.

3. The activity goes to Rachel because she has the highest priority. She doesn't answer and Wait time runs out – she is put on an unpaid pause.

4. Paul is still busy, so the activity proceeds to the next priority.

5. The activity goes to Tom as he has been inactive longer than Sophie.

Random in Priority Order

1. This strategy first distributes the activity to a random available agent with the highest priority to whom the activity has not yet been distributed.

2. If the activity is rejected or not answered before Wait time runs out, it will go to another agent to whom the activity has not yet been distributed

   1. with the same priority.

   2. with lower priority if there are no agents with the same priority available.

3. Once the activity has been distributed to all agents in all priority groups and no one has answered, the process begins again from step 1. 

Example 1:

1. All agents are available.

2. An activity comes in via the queue and starts ringing for Rachel. She doesn't answer and Wait time runs out.

3. The activity goes to Paul. He doesn't answer and Wait time runs out. 

4. The activity goes to the next priority group, to a random agent –Tom. He doesn't answer and Wait time runs out. 

5. The activity goes to Sophie. She rejects it.

6. The activity goes to the next priority group, to a random agent – Julia. She doesn't answer and Wait time runs out.

7. The activity goes to Terry. He rejects it.

8. The process starts again from step 1.

Example 2:

1. Everyone but Julia is chatting to customers. 

2. An activity comes in via the queue and starts ringing for Julia. As it's ringing, Tom's, Sophie's and Rachel's activities end. Julia doesn't answer and Wait time runs out.

3. Julia was the only available agent in the lowest priority – the activity will go to Rachel (Paul is still busy). She doesn't answer and Wait time runs out.

4. The activity goes to the next priority – first to Tom, then Sophie. They both don't answer.

5. The activity goes back to Julia.

Notes

Least recent activity distribution: the time of the last activity is calculated based on activities in the last 2 hours.