Virtual Agent

Traditionally I understand Knowledge Bases have been optimized for human ingestion and SEO, with regards to AI ingestion in Zoom Virtual Agents(ZVA), would there be any guidance on optimizing the following:1. CRM KB Records (Direct Integration)- to add a hidden field on record for only the ZVA to review and provide answers that is not visible to public (ZVA specific resolution guidance) to improve output. 2. Curated YouTube Playlist (Web crawl) - Adding Transcripts to videos, will this enhance the resources that the ZVA uses3. Word Documents (Upload to KB with manually adding tags and category) - with specific guidance however these would be for internal use and the ZVA cannot cite these as they are not publicly available.This is to improve the knowledge of the ZVA when dealing with support queries that require specific step by step guidance for resolution. Typically when syncing a KB there are column that will display Title, Category and Tag, is there another dimension of data that ca

Setup: Zoom Virtual Agent (AI Studio) connected to a Contact Center channel. We built a custom transfer flow that, before routing a visitor to a live agent, collects the visitor's name and email and sets the correct queue — so the specialist receives the visitor's details and lands in the right queue.Goal: For any live-agent / support handoff, collect name + email (and set the queue) BEFORE the transfer to Contact Center happens.The problem: Whenever a visitor expresses intent to reach support or a live agent (e.g. "agent", "connect with support"), a built-in curated flow named "transfer to support channel" fires automatically and immediately completes the transfer — bypassing our custom skill and flow entirely. The visitor is transferred with no name or email collected, and in some cases no queue set.In Agent tracing we can see the event: "Curated flow triggered: transfer to support channel" — it fires within ~90ms, before our own flow runs. On the live site the transfer completes (th

Request to enable RTMS auto-start for a General AppOur General App has RTMS scopes and an event subscription configured, and is installed on our account, but it does not appear under Account Settings → Zoom Apps → auto-start for RTMS apps, and meeting.rtms_started never fires.App ID: djb4BOjBRoWnNprKKIHQ9g Client ID: BbDXEZ3wTfCs3fbcLrsokA Account ID: gJC5uvydTJmIksm2RevpXg App type: General App, User-managed, Meetings product selected Scopes: rtms:read:rtms_started, rtms:read:rtms_stopped, rtms:read:rtms_interrupted, rtms:read:rtms_concurrency_limited, rtms:read:rtms_concurrency_near_limit Event subscription: webhook to https://portal.trainagent.co.za/api/zoom-rtms, validated successfully Developer Pack: activeThe app has no in-client surface by design, so there is no manual start path. Auto-start is the only way a stream can begin. Please enable RTMS for this app.The endpoint is live and verified: it answers the URL validation challenge with the correct HMAC and accepts signed events

This is my first time asking a question here: At our last zoom meeting for the first time I noticed a tiny ai note taker request that had not come up before.. I asked the host if it was his, he replied it was not his, as his work computer zoom will not allow AI note taker. When I asked a participant whether it was his, he reluctantly replied that he had a default AI that came on for all his zoom meetings. Is this possible and if so why had I not seen the AI note taker request before? I am not tech savvy so I’d appreciate an answer for novices. Many thanks in advance.

There has been a whole host of updates to the Zoom Virtual Agent capabilities, so much so that this video already feels slightly out of date!Quite a few things covered that will make your life easier as an admin, check it out!  

If you have deployed, or are thinking about deploying Zoom Virtual Agent then it is well worth enabling Quality Management for ZVA if you have a contact centre license.This brings your ZVA agents into scope for quality management & also includes auto scoring! Check out this short demo & think about the possibilities for evolving your virtual agent deployments. We are already using Auto QM ourselves to track how our virtual agents are performing & where customers are interacting with them in ways we weren’t anticipating. 

Hello Support community team; I escalated ticket #TS2221540. But this ticket is only Trouble shooting. From Support agent, my request should ask to this community.Because this is ZVA’s development or make call center by using ZVA’s voice agent. I think the following 5 step's Scenario.(Now Widget menu is Blank) 1. Customer call test help desk#(050-xxxx-xxxx).2.If customer wants to Japanese, Push 1.If customer wants to in English , Push 2.3.Hearing customer situation.4.If need, ZVA's Voice agent ask to customer for catch up detail.5.If ZVA's agent understands, he or she suggests accurate solution.6.If above's suggestion's does not fixed, Voice agent ask to customer name and mobile number.7.Can we receive from step 3 to 6 conversation's story by e-mail?8. Form improvement of speech accuracy, I'd like to manage knowledge base, add too, and Intent function.(But I'm not sure their steps)Would you mind sharing above step by step in detail by April 2nd?(I'm beginner ZVA)I appreciate your dedic

 ⚫️ Process to test AIC 3.0 and transition your current agent to AIC 3.0:Duplicate your existing agent Ensure the ZVA 3.0 toggle is enabled on the cloned agent Run test scenarios in preview to confirm expected behavior (Optional) Once validated, go to your production agent and scroll down to the bottom of the page to find the traffic splitting setting (Optional) Start by routing 10% of traffic to the cloned agent with ZVA 3.0 enabled, 90% of traffic to your existing agent. (Optional) Monitor containment rate, resolution rate, and surveys to identify any adjustments needed in the ZVA 3.0 agent. Make adjustments and publish the AIC 3.0 agent. Gradually increase the traffic routing until you are ready to enable ZVA 3.0 on your production agent. Once you are at 100% of traffic on the AIC 3.0 agent, then you can go back to the AIC 2.0 original version, turn on AIC 3.0 and 100% of traffic on the slider. Delete the test agent.  ⚫️ Why did we launch AIC 3.0 or what is different? Read this: htt

If your IT team wants to understand the Zoom Marketplace connector for Sharepoint and Zoom Virtual Agent, below you will find common questions and best practices. Link the to connector is here. 1. How Authentication and Permissions WorkWhen you connect SharePoint as a knowledge base source in Zoom Virtual Agent, the integration uses OAuth authentication. This means: The SharePoint account used during authentication authorizes the connection between Zoom and Microsoft 365. The Virtual Agent does not gain elevated or independent permissions — it can only access content that the authenticated account has permission to view. In other words, the Virtual Agent “inherits” the same access rights as the SharePoint account used to establish the connection. This design ensures that SharePoint’s native access controls remain fully enforced. If a document or site is restricted for that account, the Virtual Agent will not be able to retrieve or surface it. 2. Restricting Access to Specific Sites

The Arch Linux tarball file downloads. However the build or just tar xJf fails with an “unexpected EOF in archive”.Here is example output of the build and the use of tar command.Package (1) New Version Net Changezoom 7.0.0-1 935.63 MiBTotal Installed Size: 935.63 MiB:: Proceed with installation? [Y/n] y(1/1) checking keys in keyring [------------------------------------] 100%(1/1) checking package integrity [------------------------------------] 100%(1/1) loading package files [------------------------------------] 100%(1/1) checking for file conflicts [------------------------------------] 100%error: failed to commit transaction (conflicting files)zoom: /usr/bin/zoom exists in filesystemzoom: /usr/share/applications/Zoom.desktop exists in filesystemzoom: /usr/share/mime/packages/zoom.xml exists in filesystemzoom: /usr/share/pixma

Ever tried to prompt and get the LLM to gather details from the customer and then use that to make decisions? Has it ever failed? That’s because your architecture for how variables and LLMs interact might need a refresh. Best practice 1: Always use a tool to set the variable and then actually map it to a real global custom variable. Yes, it takes longer in the beginning but it’s worth it for the consistency in experience. Best practice 2: Create a set of 10 “backup” custom variables for you to use as placeholders while you build. You can then come back and rename them to the correct value they represent or exchange them with a new variable. This helps keep you in the habit of actually mapping to a variable field, saves time while you build, and let’s you be more creative with your UX design. Keep a list of what you used the “backup” variables for and always go back and rename them or make new variables to switch out. Best practice 3: Don’t just prompt a reference to a variable. Actuall

Hi everyone,We’ve been using Zoom ZVA’s built-in transfer_to_live_agent tool to route customers into our ZCC flow. Overall it works well (around 90–95% success rate), but we’re running into an unavoidable issue where the transfer fails because the required variables aren’t being set by the agent properly before execution.  These variables are critical on our end since they determine which queue the customer gets routed to.Has anyone else experienced something similar with variable handling during transfer? If so, how did you resolve it or work around it?Any insights would be amazing!Thanks,Arpit 

Has anyone been able to actually reach a human for support? The virtual agent gave me a ‘supervisor’ which sounds like a fake title given out and then that person gave me an email that bounces back email. I can see why they are loosing customers.

You’ve seen ​@ShawnFerrell’s awesome instructions on how to build AI Receptionist (concierge) for Zoom Phone…..Now see it in action as I challenged myself to build one in under 10 minutes!Zoom have packed a huge amount into AI Receptionist & made it really easy to deploy. Check out the video on YouTube & leave a comment on the channel for what you would want to build!  

Important Note: This guide is tailored for the ZVA 3.0 experience. If you’re using 1.0 or 2.0, please let me know in the comments and I’ll happily do a write-up on how to build an AI Receptionist for your ZVA version!Why Build an AI Auto-Receptionist?If you manage Zoom Phone, you already know the Auto Receptionist. It’s the front door of your phone system — callers dial in, listen to menu options, press a number, and get routed. It works. It’s reliable. It tells you almost nothing about why people are actually calling. Zoom Virtual Agent (ZVA) changes that. Instead of routing calls based on key presses alone, a virtual agent listens to what callers say, understands their intent, and routes them accordingly — while capturing data on every interaction. It even assists with deflection, by being empowered to handle FAQs that bog down your frontline teams. The result is the same routing your team already manages, plus analytics that reveal what your callers actually need, enabling you to ta

Hey Folks!Depressingly this video is already out of date as 3.0 very recently launched, but hopefully it will still help you familiarise yourself with ZVA build & moving away from the classic chatbot 1.0 concepts.https://youtu.be/3PjVg2hfq6k

If you are asking a series of questions to collect information but if the customer’s answer then determines the next question (an “if-then-else” prompt) this is the best practice with ZVA Agents: 1. Turn on AIC 3.02. Make an instruction skill 3. Limit “if-then-else” prompts to five scenearios using an instruction skill. If you use  six or more “if-then-else” statements in your prompt, then the experience will degrade because the context of the conversation is not best matched for one agentic conversation. If you have more than five “if-then-else” statements in a single prompt, use these questions to continue following best practices: » Why do we need these questions? » Can we include the intent of these questions inside of questions #1-5? » Will my customer be fatigued at this point and give false or incomplete answers in order to “just get it done?” » Have you prepared the caller before starting the “if-then-else” statements that this process take more than 5 minutes to complete? To c

I’ve been playing with ZCA quite a lot recently & just added the latest video to the #apieceofCX library with a look at ZVA Analytics.Very impressive & super accessible.Watch on YouTube here!

 Check out the features from January that we recommend you consider for your ZVA management practice! All features listed here: https://support.zoom.com/hc/en/article?id=zm_kb&sysparm_article=KB0067812

Hi everyone. I'm trying to setup the integration between AVZ and Manage Engine ServiceDesk Plus.Based on the KB https://support.zoom.com/hc/en/article?id=zm_kb&sysparm_article=KB0057946 I insert "itdesk" and select my region (EU). I'm redirected to the auth page on Zoho but after the authentication process I land on an error page that says:  Invalid ClientClient ID passed does not exist Any idea?Is the ServiceDeskPlus supported or do I miss something?"itdesk" refers to other ManageEngine products? RegardsLuca  

Hello, I need your helps. I got 050 phone number by eKYC and paid “Charge Name:Zoom ContactCenter Monthly Usage -”. But I could’t find voice agent menu(I showed chat agent and classical chat bot menu only).I’d like to make the simple IVR(Japanese or English and Loss the urgent matter(mobile , PC) or Admission Application) using Zoom Virtual Agent simply.Would you mind sharing the step by step the flow guide immediately?Also I’m very happy to the guide with Japanese directly. I appreciate your dedicated support for Japan in advance, Hiro.  

I would like to be able to filter matched queries that I have already coached and trained. I would expect coached queries to drop from the list of matched queries or there should be a filter to hide "coached queries". Do they never disappear? If not, how is that scalable? If they should be dropping from the list, what step am I missing? My steps are:1. Navigate to Intent Management.2. Scroll to an Intent group.3. Select an intent within the group.4. Click on the Coach tab.5. Click the check box next to queries to be coached.6. Add to training phrases.7. Train.8. Publish. The use case for why it is problematic is sorting through multiple identical or similar queries to find the queries that I have not coached. Below is a real-world example of one of my intents:1. I have an intent for Financial Input.2. I have 195 queries for "Financial review," 4 queries for "Financial input," and 1 query for "Can I get financial input, please?"When I have 5000 queries, how will I filter out "Finan

We have a complex contact center flow that is working well but we cannot stop the transfer tool from saying its inner chain of thought inspite of various guardrails.Any ideas? None of our prompts provide instructions on reading content back to the caller.eg. AI Agent Hello. How can I help you today?Consumer How can I set up a Saturday morning detention?AI Agent Okay. Let me transfer you to one of our support agents who can help with this. Can you please tell me any more details we may need to know before I transfer you?Consumer Uh, our period structure is just a Monday to Friday.AI Agent Right. Thank you for that detail. Is there anything else you'd like to add before I transfer you?Consumer No.AI Agent Setting up a Saturday morning detention. The period structure is currently Monday to Friday only. No further details provided.AI Agent Here's what I've captured: Wants to set up a Saturday morning detention but period structure is Monday to Friday only. Connecting you to an agent now.AI

Make sure that you are optimizing your agent for the job they were “hired” for. Do not prompt-bomb the guidance and write a Da Vinci Code length guidance prompt. It’s hard for a human to read and hard for AI to understand. So, what is the ideal length of your guidance? First, factor in that ZVA is not only digesting your guidance prompt but also, Length of the conversation - which we are optimizing using summary. Number of skills and their complexity. How tool responses are being returned. How guidance is written as well (the architecture and formatting matter).  Best Practice:Target for one agent’s guidance prompt: 2,000–4,000 characters(~500–1,000 tokens)Why this range works bestEnough room for:role definitiongoalstoneguardrailstool-calling rulesShort enough that the model consistently attends to itEasier to debug when behavior goes sideways Upper limit: 8,000 characters. You won’t be stopped, but you also won’t see the results you nor your users want. Once it goes past this, you’re