We are excited to launch AI agents and automated resolutions!

Hey @AA 

Thank you for flagging the note about your account executive. We have raised these questions to your account team again to prioritize a response. They will follow up as soon as they are able!

As far as just saying hi to a bot and being charged, that’s not quite correct. A customer must do one of the following things within the course of a conversation:

• Receive a Generative Reply as a response to a question sent by the user
• Click on a recommended article link and read the article as a response to a question sent by the user
• Navigate to the the last step of an answer flow

Once they’ve done at least one of the previous actions and don’t escalate further to a human agent within 72 hours, Zendesk would consider the conversation as resolved. If the customer were to say Hi and then leave the conversation, this would not be considered an automated resolution and there would be no charge.

Hello @viachaslau11,

We appreciate the feedback. We’ve been taking feedback from customers during the trial period and are actively working to address cases where our AI agent is not providing meaningful resolutions for customer inquiries. In particular, we are working to correct for conversations that end with customers replying that responses are not helpful. We are also rolling out additional checks and validation post-conversation to correct for other scenarios where resolutions are counted for unhelpful conversations. We expect to have these changes in place before customers are asked to pay for resolutions. 

We will also continue to improve our bot capabilities to introduce conversational context as well. We are working on timing for this and will communicate further when that is available.

When it comes to which features are free and which are paid, it’s always a balance. Firstly, to your point about building flows that use intent. When we launched AI Agents and outcome-based pricing at Relate we made our intent models available to all customers for use in Bots. This no longer requires the Advanced AI add-on, however the Advanced AI add-on still grants you capabilities to assist your agents and build more intelligent workflows. Additionally, if most of your flows are about capturing customer data and then handing off to an agent, these will not be tracked as an automated resolution. If instead this captured data is used to resolve a problem for the customer through an answer flow, then we would consider that to be an automated resolution.

Most bot products on the market charge based on overall conversation volume and not based on the outcome of those conversations and do not take into account the particular features a user might interact with. Our goal is to provide a value-aligned price when we solve a problem. As mentioned above, we are working diligently on addressing any misaligned scenarios. 

@ryan48 I appreciate the additional context, but I'm afraid I wasn't clear enough in my original post.

It's not that I'm unable to understand what those metrics indicate, but that the way that Zendesk is defining value is not aligned with value created for my customer or company.

Instead, the system as it is set up uses metrics that strongly favor delivering value to Zendesk with little regard to provable value for the customers. The metrics I listed were simply illustrating that point, which you have underscored in your response.

I think that we can all agree that a support interaction that doesn't deliver value to the requester or the company supporting that requester isn't worth the effort or expense.

However it's done, whether using a metric that's more aligned with actual resolutions, reducing the overall cost of these features, or some other approach, refocusing on delivering value to the customer will always be a winning strategy, especially in an extremely competitive marketplace like AI-enhanced support.

@winslow 

I appreciate the feedback and as mentioned we're continuing to tune the performance of automated resolutions as we go through this trial phase with customers. Our goal is to provide value and only charge when that is delivered. We will keep working toward that in partnership with our customers. 

We agree that this market is competitive and we believe that outcome-based pricing is how we can differentiate our product. As mentioned in previous posts, as we looked at other players on the market most bot products are monetized based on things like the volume of conversations regardless of outcome, or even pass through the cost of tokens sent to an LLM to customers. Additionally, very few of these product are self-service and often require services engagements or ongoing developer resources to build, deploy and maintain. We will continue to work with you and the rest of our customers to align pricing and value as we believe that's the best path forward for Zendesk and our customers.

Thanks for your answers @ryan48. I always appreciate when Zendesk product managers are actively involved in discussing their products with customers.   

We’ve been taking feedback from customers during the trial period

Ideally, these issues should be addressed before the product becomes public. Here it looks a bit like putting out fires 😇.

We will also continue to improve our bot capabilities to introduce conversational context as well. We are working on timing for this and will communicate further when that is available.

Hopefully, this improvement will be introduced soon. Many questions remain essentially unanswered just because the bot doesn't keep the context of a conversation. And by the way, they are now counted by you as resolved.

When we launched AI Agents and outcome-based pricing at Relate we made our intent models available to all customers for use in Bots. This no longer requires the Advanced AI add-on, however the Advanced AI add-on still grants you capabilities to assist your agents and build more intelligent workflows. Additionally, if most of your flows are about capturing customer data and then handing off to an agent, these will not be tracked as an automated resolution. If instead this captured data is used to resolve a problem for the customer through an answer flow, then we would consider that to be an automated resolution.

This is another problem we see in Zendesk. Namely unpredictability in actions. We didn't just buy this add-on, we had multiple conversations with our account manager. And so we pay money to use intents, and in weeks the same feature becomes available to everybody, but we have to pay for resolutions that are not really resolutions because they are just our custom bot flows. We're basically paying for volume, and that's the approach of your competitors, which you challenge in the comment. 

We're on the Zendesk Suite Growth Plan, so we'll be allowed 10 automated resolutions / agent /month.  What is an agent?  We have 3 seats in our plan and 50 light agents.  Given that, how many automated resolutions am I given in a month?

• 30 (10 * 3 seats)
• 500 (10 * 50 light agents) or
• 530 

@ryan48 I wanted to ask a followup question. AFAIK, we will begin getting charged on July 10th (today - please correct me if I'm wrong on that). In preparation, I shut down the bot last week.

Despite that, “resolutions” continue to pile up. I'm not sure what else needs to happen to make them stop so I can avoid being charged. The email autoresponder is also off/deactivated.

What specific sources of these resolutions might there still be?

@lachrystal - I believe only full agent seats count towards your resolution allowance, giving you 30 automated resolutions per month out of the box.

@winslow - Not sure why your resolution # is increasing but, as I understand it, outcome / resolution based pricing will only go into effect for you on your next contract renewal date after July 10th.  I'm sure @ryan48 will be able to provide additional context here.

Hey @viachaslau11 ,

Thank you for your feedback. Direct responses from our customers influences our product roadmap and is the reason why you will start to see updates next week and with iterative improvements continuing until our launch on August 14th. We will always continue to listen to customers, learn more and adjust. 

When it comes to Advanced AI vs. AI Agents, our goal is to clearly delineate the capabilities and value that we provide to Agents and Admin and the automation provided to users. With Advanced AI, you have access to features that enable agents to perform more effectively and admins to build better workflows and reporting. With AI agents, you will be able to continue to use data capture capabilities as part of an agent conversation without incurring automated resolution costs. Automated resolutions will only be generated when an AI agent is able to resolve an issue without involving a human agent.

Hey @winslow 

Hopefully I can clear this up for you quickly. A few points, you should have seen communication from us this week that we've extended the date to August 14th.  Additionally, as @brandon12  points out, this pricing comes into effect at your next renewal after that date. In your case, that's in December 2024 and you will not incur any charge before that date.

There are multiple automation capabilities beyond Zendesk Bots that provide Automated Resolutions.  There's more information in this article but this would includes autoreplies with article recommendations to tickets via triggers or via the webform in help center or via the legacy Web Widget. 

Included in your plan you have access to a set of free automated resolutions over the course of your contract term. For Suite Enterprise customers, that 15 ARs per agent per month over the course of your contract. On August 14th, we will roll out additional functionality for managing your automated resolutions, including capping your usage at your current subscription level so you never incur overages.  You'll also get more information on the usage for different automation features mentioned above.  

Automated resolutions will only be generated when an AI agent is able to resolve an issue without involving a human agent.

@ryan48  Once again, I have to emphasize that the custom answers we create are not technically AI-agent related (we usually create them when we need real customer data and we make API calls inside bot steps). But Zendesk will charge us as for AI-resolved issues.

Basically, we have (1) Generative AI replies based on our help center, and (2) manually created replies based on real customer data. We would love to have a generative AI replies based on our CRM, but Zendesk AI bots do not have such capability. 

@brandon12  Thank you!  

@ryan48 and @brandon12 thank you for the additional information! While I did not receive any comms about the data change, I do see the updated timeline in the referenced article and appreciate the call-out.

Since I started using AI Bots in May 2024, but also updated my contract last month to add Advanced AI add-ons, please confirm that the new date of August 14, 2024 does not apply as my billing start date and that I am free to experiment without incurring usage charges until my next contract renewal data in December 2024.

I have noticed that Zendesk considers an out-of hours business flow and automated resolution. We have a bot flow which checks business hours and if it is out of hours, we post a message and link to the ticket form. This is not a resolution as there is no other out-of-hours option; the chat won't be transferred to an agent and so will be liable to be charged. This seems unfair.
 

Hey @rachel12 - 

The recommended solution here is to use the “Transfer to Agent” step, even if no agents are online.  The key is to specify in the step-language that the user is creating a support request.  The ticket will be created per usual, and when your staff logs in it will be there waiting for them.  This will also allow you to continue the conversation async via the widget and via email, and (as far as I know) will not be considered an automated-resolution by Zendesk.  Hope this helps!

Brandon

Hey Viachaslau,

AI Agents encompass automation on several channels including our Zendesk Bots capabilities. Automated resolutions will apply to all of these channels and features. All of them use AI in some way, through embedding models, our out of the box intent models, language classification, or generative AI. As you point out, Bots can rely on logical operators, branching flows, data capture and API integrations to solve customer problems as well. All bot products on the market charge either via the number of conversations engaged in or the numbers resolutions delivered regardless of the types of automations or machine learning models used and we see our pricing to be in line with that.

Can you explain more about what you're looking for in a CRM integration? What kind of data would you leverage from your CRM and what use cases would that serve for your customers? What CRM product are you using?

Thanks!

Couldn't agree more with @viachaslau11. AI agents should be just that, agents that use AI tolls to solve customers. A logic flow that admins made should not be counted as an AI agent as there were zero AI tools used to support this customer.

Hey @Raul Winslow ,

I just wanted to respond back and confirm that you are free to experiment without incurring usage charges until my next contract renewal data in December 2024.

Thanks!

Hi. 

Does anyone know where the option is to set a cap on the number of automated resolutions that can be consumed by your account each month?

I have had a look and cannot find the option anywhere for some reason

Hi @megan27, 

You can find details on automated resolutions, including how to cap usage in the link below

https://support.zendesk.com/hc/en-us/articles/5352026794010-About-automated-resolutions-for-AI-agents

Let us know if we can provide more information. 

With how frequent false positives are, I'd be surprised if the introduction of ‘Overages’ doesn't severely hurt your business, Zendesk. It's insane that you'd rollout these changes in this state, without the ability to report false positives. It renders a conversation bot virtually useless after a few customers leave the chat early unresolved, but using up your automations..

Hey Eric,

We’ve put specific controls in place to detect and exclude abandoned conversations and conversations that don’t provide a satisfactory resolution so that they are not counted as Automated Resolutions. Additionally, within the Bot conversation transcripts page you can provide feedback on the conversation that Zendesk reviews regularly. I’m going to ask our team to proactively review your recent conversations to determine if there’s anything unusual happening with your account. 

Hi Zendesk,

we recently started using the AI article suggestions in our email receipt confirmation - in theory this was super easy to implement which was great - now analyzing results, it does not work exactly as we hoped (and unfortunately leads to unreal success stories in reporting etc.) 

We figured, a request would only be considered ai-auto-resolved IF the requester actually clicked on “Yes that was helpful/Close my request” > now we find several tickets, where the recipient did get the 3 article suggestions, viewed them (in some cases even marked them as unhelpful), never clicked on “yes this was helpful/close my request” and still the tag “ai_agent_automated_resolution” was added? This makes no sense to us, and also makes creating follow up triggers here quiet impossible/much trickier. As you can imagine, if a requester says an article solved their request, the ticket goes to solved status and noone on the team will ever see it, we do want to send a confirmation email for that as well. Letting the requester know that if that happened by accident or if they have further questions, to email us again. We built that trigger based on the tag “ai_agent_automated_resolution”

Now why is that tag being added to tickets where clearly it was not auto-resolved? How can we identify only tickets, where the requester clicked on “yes close my request” ?

**** Below only two of many many many examples where tickets should have never in any way be counted as automated_resolution - if this behaviour is not corrected we will have to turn this new exciting feature off again - Update 24 hours later: had to set this workflow to inactive again, as there really seems to be no way to cleary identify tickets that were indeed and truly auto-resolved by an article suggestion. Look forward to any updates here. 

After contacting support, it looks like tickets that are deleted or marked as spam are also being counted as automated resolutions?

The support team is unable to clearly tell us where our usage is being used. The web-widget bot stores no contextual data and no report/metrics for the amount used by 'deleted tickets', which in my opinion, should not be charged in the first place. 

As others may have pointed out, automated resolution after 72 hours is too-harsh, e.g. we don't operate on weekends. That forces us to give a response in 1 business day, and that is assuming there aren't bank holidays before/after weekend. 

Hello @retourenstation 

Thanks for reaching out. The tag ab_resolved is added to the ticket when the customer marks the article as solving their problem (please note, this tag will soon be renamed to ar_marked_helpful). This should allow you to identify tickets which were closed for that reason only and use it in your triggers. Additionally, we are working on a small update for autoreplies such that clicks on unhelpful will never be marked as an automated resolution.