We are excited to launch AI agents and automated resolutions!

Hello @jeremy12 

Looks like we've got an open ticket with you right now and my team is engaged with our Customer Advocate helping you out there. We'll work with you on the specific details there.

Thanks!

There are so many “what if” scenarios here that it seems absurd to roll out this pricing before those scenarios are 100% settled.

“We have tools for false positives and this will continuously be improved”

Then charge us when it IS improved, not when we know beyond any shadow of a doubt that Zendesk customers are going to be charged by resolution for things that are not truly resolved. Whether it is 1 false positive or 1,000, we should not be charged for any of them and we should not have to go through the effort of creating support tickets to get the false positive charges removed. It would probably cost me $100 worth of time to get $10 of false charges removed. 

I was so excited for advanced AI and was going to build so many cool things around it, but the pricing update took ALL the wind out of my sails and I don't want to go through the effort of building these tools if I just need to cancel the AI add-on when my contract is up. What a truly anti-customer move from a customer service company.

Hello @adam14 - 

Thank you for reaching out and thank you for your feedback. 

First, we agree that you should not be charged in instances where AI agents don’t resolve a customer issue. That is exactly why we have gone to such lengths to ensure industry leading accuracy rates and why we commit to continually improving. 

Second, our Advanced AI SKU is agent productivity focused and, by and large, does not generate automated resolutions. You should feel comfortable using Advanced AI and not need to worry about usage-based costs. The one within Advanced AI that can generate automated resolutions is Intelligent Triage. If you would like to ensure that no additional costs are incurred, you have the ability to control usage through your Admin Center.

Hi.

If you want to avoid incurring additional costs, you can control usage from the Management Center, but when I checked that page, it was as follows.

- Maintain functionality and allow usage overage
Impact on service

• AI agents, including bots, will continue to respond to customers, saving agent time and improving the customer experience

Cost

• Higher rates will apply for usage overage

- Suspend the feature and do not allow overage usage
Impact on service

• AI agents, including bots, will not be able to respond to customers until auto-resolution is added
• The number of requests forwarded to live agents will increase, resulting in increased first reply time and backlog
• Auto-reply with article and auto-reply for intelligent triage will not be available for triggers
• Web form recommended article suggestions will not be available

Cost

• No additional cost

If you choose “Suspend functionality and do not allow overage”, won't that mean that the bot won't be able to respond to customers?

The increase in agent transfers is probably because, since the bot's automated responses are not available, customers have no choice but to ask an agent, and that's why there is an increase.
 

I also don't understand why the recommended articles for web forms can no longer be used. They were originally free, so why tie them in with the bot? It's a feature that's almost completely unrelated.

I don't understand this pricing structure at all.

Hello 小林正左子 - 

We aim to make our in-product experiences as simple as possible and we continue to iterate on these descriptors. Thank you for letting us know areas where we need to be more clear.

This management center page allows you to control what happens when you reach the current entitlement of automated resolutions (both included and paid). This gives you the flexibility to decide if you would like the use of AI Agents to continue beyond your current entitlement and be subject to Pay-As-You-Go pricing. 

To directly answer your question, choosing the “Suspend functionality” option means that instead of exceeding your current entitlement and being subject to additional costs, inquiries will be routed to an agent. This only applies when you exceed your current included and paid amount of Automated Resolutions and will not have any impact prior to reaching your limit. At any time, if a customer speaks, either by customer request or through the “suspend functionality” option, it will not be counted as an automated resolution. 

Article recommendations in Web form and in email are features that deliver automation to customers on channels other than our Messaging. AI Agents solve customer questions before an agent needs to get involved whether that’s via Bots in Messaging, or article recommendations in the web form. They’re all powered by AI technology that responds to customer questions and issues. These features have gone through pricing strategy shifts over the years and article recommendations in the web form, for example, haven’t always been free. Outcome-based pricing and automated resolutions aims to unify all of these AI automation capabilities under a single pricing structure. 

Hi, Jeff Curran

Thank you for your response.

I have a few questions for you.
1. You explained that "choosing the 'Suspend functionality' option means that inquiries will be routed to an agent instead of incurring additional costs beyond your current entitlement." However, isn't "incurring additional costs beyond your current entitlement" incorrect? Shouldn't it be "no additional costs will be incurred beyond your current entitlement" instead?
2. You mentioned that "article recommendations in the web form have not always been free." Could you tell me where this explanation was provided to us customers? Also, is my understanding correct that these article recommendations refer to auto-replies?

Thanks.

Hello and thank you for following-up. 

We’ve included automated resolutions in your Support or Zendesk Suite plan so that you can try AI Agents easily and without additional cost. However, if your usage exceeds your current entitlement, you would be subject to usage-based billing for any overage. The ‘Suspend functionality’ option will prevent AI Agents from fully automating a customer inquiry by routing to an agent, either synchronously if agents are online or asynchronously through tickets, when you reach your current automated resolution entitlement. By doing this, it prevents you from incurring any unplanned costs associated with automated resolutions. 

To your second question, there was a cohort of our customer base that was paying per resolution for auto-replies in the web form. You can see more information about how that worked in this article. As we moved to the Suite, we began including a set of free usage for that product in the Suite. We’ve continued to do that with our free automated resolution entitlements. 

I have the same concerns as most people posting here - inaccurate automated resolutions.  The problem for us is that customers have no idea their ticket is solved just because they Viewed an article.  We have an automated email trigger that goes out 1 day after resolution to tell them their ticket is solved and offering a CSAT survey.  They are surprised and UPSET by this which is causing unnecessary negative CSAT scores.  We have since implemented a new trigger telling them that ZenDesk thinks they just solved their ticket by viewing an article and did they really want that - this happens when the ticket moves to Solved.  This may help with the CSAT scores, but this is all unnecessary.  I think all of us would be a lot happier if an automated resolution happened ONLY when the customer marked it as solved.  There should be no interpretation by AI for resolution.  The customer most vote that their issue is resolved.

Hey @rich11 simply looking at a sujjected article won't mark the ticket as solved. When the user clicks on the link they are presented with a popup near the bottom of the screen saying that they can mark this as solved.

@rich11 absolutely agree ont this! We tested first in September 2024 - quit the topic as it was not working as it should. Just did another round of testing, while the status seems to not go to solved anymore when it shouldn't - the tag “ai_agent_automated_resolution” is still added when it shouldn't - added my feedback in another article . Hoping to see proper fixes here soon, as the feature itself would be really exciting if it would work 100% 

https://support.zendesk.com/hc/en-us/articles/4408833721498-Using-autoreplies-to-recommend-articles-in-email-notifications 

@rich12 We have faced a similar experience. A significant number of customers do not realize that by marking an article as useful, they close a ticket. We have also implemented a trigger - that if a client first marked an issue as solved, but then disliked it, the ticket is reopened.

@zach16  you'd think, but that is NOT what we are seeing in some cases. I know that customers can click on a Problem Solved button if they open a suggested article.  I spoke with a customer whose ticket showed she looked at each of the three articles recommended.  She explained that she did not look at one of them.  She did not click on on and of the links at all .  She sent a ticket and that was it.  I have other tickets that show only 1 one the three articles was viewed, but a different article was listed as having solved the ticket.  Shouldn't the one that solved the issue show as Viewed in the ticket???  Something is really wrong.  I have a work around in place, but the customer experience is poor.  I created a trigger such that an “are you sure” email is sent to any Ticket that is solved by AI or Automation.  This helps, but still not a great experience.  I am seeing customers reply “I never solved the ticket!”.  I strongly believe AI (or ZenDesk rules created by ZenDesk) should not be trying to determine if these cases are closed.  

I ran into the exact same issue as  Viachaslau. Customers would mark an article as useful, the ticket would solve without them or us knowing, and they would leave us negative feedback. That is frustrating enough, but being charged as an automated resolution for that is extra bad icing on the cake.  I implemented the exact same trigger to bring back tickets with negative feedback that were automatically solved, which at least lets us help the customer, but unless I am missing something, we would still be charged for a false positive automated resolution. I am on the verge of disabling the feature outright; it has been far more problematic than useful so far, and I don't like paying for problems.

Hey @rich11 - nice to see you here!  This is not expected behavior as far as I know.  I would recommend having Zendesk Advocacy look at the ticket(s) in question to figure out what happened.  @retourenstation the automated_resolution tag should also strip, so you might want to have Advocacy look at that as well.  Hope this helps!

Hello,

I am experiencing the same issue like some of you recently, where tickets are solved as if the end-user clicked “Yes, close my request” without them knowingly doing so.

I am suspecting this is due to link expanders, similar to this issue where safe link or url expander service automatically opens the link in the email to close the issue. My suspension is because some of the reported cases are solved immediately when it would not have been possible for the customer to go through the article. I am not sure how to verify my suspicion about this yet.

@ahmed11 We thought the same thing, so I reached out to a customer who experienced the issue and met with their IT who verified it is NOT a link expander.  I further verified this when an inside sales rep at OUR company forwarded a customer email to our support team, and it happened to our sales rep.  We do not use a link expander.  It is also a very random thing.  The same customer who I contacted about it sees the problem in only some tickets.  Some are perfectly fine.  I checked all the obvious stuff - browsers, mail apps, OS, etc.  Nothing obvious on the customer side.  I have a ticket open with ZD Support who can reproduce the problem.  It is currently with ZD engineering.

@brandon12 Thank you for your attention to this. I am not sure what you mean with “the automated_resolution tag should also strip” > honestly that tag beeing added to tickets, even when the requester did not mark the request as solved is our biggest issue with this feature at this point. 

Zendesk confirms, that the tag will be added also if within 72 hours of the ticket being created/the article suggestions being received/viewed by the recipient and no further public comment is then made for 72 hours (from agent or requester) that tag “ai_agent_automated_resolution” is added to the ticket. 

Let me give you a real life example: Requester submits a ticket on Thursday 06:30am > our team is in some sort of Peak Season and response times are longer than usual. The Requester received the autoreply with article suggestions right away, and views the suggested articles, but does not feel his/her request is solved, so no further action is taken. The ticket is assigned to an agent on Monday morning and he/she responds around 10am - this is more than 72 hours since the requester viewed the suggested articles. The tag “ai_agent_automated_resolution” is added, soley based on the fact that there was no further public comment within those 72 hours. However a human agent indeed has to handle the request, so this, in my opinion, should not be tagged or counted as an automated-resolution - hope this makes sense.

We would love to continue using the feature and implement it for all our brands - but in our first test with approx. 100 real life tickets, we noticed about 30% were indeed not auto-resolved. I can not imagine this not to be an issue for any other Zendesk client, as during Peak Seasons response time may very well be longer than 72 hours. Is there any chance Zendesk would consider optimizing this - and/or how can I submit this to the Zendesk Advocacy Team to look into this? 

In my opinion: only when a requester, who received article suggestions in autoreplies clicked “Yes, solve my request” it should be tagged and counted as an automated resolution. How does everyone else here feel about this, am I misunderstanding or missing something?

Hey All,
 

Thanks for providing feedback and giving us real work examples you're seeing. I will provide some context here and some guidance on next steps. 

As initially designed, you're correct that we look for situations where customers have clicked on article links and reviewed an article. If with 72 hours there are no public customer or agent replies on the ticket we currently count this as an Automated Resolution. As we looked at this, 90%+ of tickets are resolved within 72 hours. We also wanted to account for scenarios where customers did not click the “Yes this solve my request” button since we no that won't happen every time a customer gets the information they need. So we believed that the combination of these 2 factors would resolve false positive that might occur. 

We rolled out this change back in August and prior to this post we weren't fielding significant feedback from customers that it wasn't meeting their expectations. We're reviewing that now based on this thread.

This particular feature has been part of our Zendesk Suite for roughly 7 years now. Looking forward, we are planning to launch our new AI Agents Essential capabilities that power Messaging conversations to ticket channels like Email, Webform, etc. This should be a significant upgrade based on Generative AI and should have much higher automation rates. 

Myself and my team are going to review this feedback and determine what action we can take based on your feedback. Please give us some time while we do a deeper analysis. As always, we want to listen to our customers and adjust to their feedback so they feel that they're getting value from the features we provide. 

Thank you!


 

Hey All, 

I wanted to provide an update here. We have a team of people evaluating the impact of changing the definition and introducing LLM evaluation to autoreplies with article recommendations to determine how it will impact conversations. As part of this process, we're building out datasets and doing data analysis that will take a few additional week of research across our customer base. Based on that output, I will inform people on this thread of the specific actions we will take and the timeframe for those actions.

Thanks!

Hi @ryan48, @shawna11 -

Looking forward to hopefully running into you at Zendesk Relate!  Meantime, I had a question that I know you can shed clarity on.  When it comes to customers on an annual billing plan, do either plan-allowance automated resolutions or pre-purchased automated resolution packs carry over month-to-month within the annual subscription package?  We have some seasonal use cases that anticipate a surge of holiday-related Q4 ARs, and I'm wondering if it's better to over-index the slower months for budgeting purposes, or if those customers should anticipate buying a large volume of ARs as a one-off during those time periods.

Along simlar lines, as I understand the documentation currently - customers who were using the  Legacy Bot Builder prior to February of 2025 will continue to have access to it for a currently unspecified amount of time without needing to upgrade to the AI Agents - Advanced experience.  However, once those contracts renew after March 25th, 2025, they will begin to incur usage deductions for Automated Resolutions and, if they choose not to allow for overages, their Bot Builder flows & answers will pause and revert to a ‘transfer to human’ flow-step until the ARs are replenished.  Have I got that right?

Brandon

Hi @brandon12 - Thanks for the questions. 

Re: AR pooling on annual contracts - Customers on annual billing plans receive their entire plan allowance and pre-purchased automated resolutions at the beginning of the contract. They are able to use them at any time over the course of the contact, so can flex them between months within the contract. No need for customers to purchase one-off amounts to handle seasonal surges. 

Re: Legacy Bot Builder - Your understanding is correct. Renewals at March 25, 2025 will see these flows result in automated resolutions and count towards a customer's entitlement. If a customer chooses to not allow overages, when they reach their entitlement flows will revert to a ‘transfer to human’ step. 

Please let us know if we can answer any more questions

Thanks!

Hi, Our business have multiple brands and to each, their respective support teams. Similarly, they have their own AI agents. 

I'd like to know how I can get a report for the number of Automated resolutions consumed by each AI agent to facilitate cost calculations, preferably in varying intervals like week of year, custom dates, or months. Anyone know how?  

When using “autoreplies with articles”, if the end user clicks “Yes” to the “Does this article answer your question?” prompt, that immediately counts as an autoresolution. This is documented behavior. However, if the end user reopens the ticket (e.g., “wait - that didn't actually fix my issue”), an Agent is involved and works the ticket as normal. Yet it still counts as an autoresolution. IMHO, this is broken behavior, because we are charged for an autoresolution even though the end user's issue wasn't actually resolved.

Another way autoresolution is completely broken for “autoreplies with articles”: We have a common support case that requires Agent interaction, as they have to update a user setting in the system. We have a Help Center article that triggers from the error the user reports, and tells them that an Agent needs to update this for them. When the user clicks “Yes” on "Does this article answer your question?” prompt, this gets counted as an autoresolution even though the article states in so many words that the issue cannot be resolved without Agent interaction. The article is correct, and it's necessary, so that users know what's happening, and it is absolutely ridiculous that we are charged an autoresolution for this scenario.

Bottom line - charging an autoresolution simply because an end user clicks Yes to this prompt is broken. Autoresolutions should only be charged after evaluating the ticket stream. There are too many scenarios where autoresolutions are charged when the issue is not resolved without Agent involvement.

Hey @david22 

Thank you so much for sharing this feedback and for the clear examples. You're right that today, with “autoreplies with articles,” a “Yes” click on the article prompt does count as an autoresolution, even if the issue later requires agent involvement. This is expected behavior today and aligns with how resolutions are measured. That said, we completely understand how this can feel misaligned when articles are used to inform — and may not resolve — or when customers reopen tickets. These edge cases are one of the reasons we’re shifting our focus toward launching email support for AI agent Essential, which uses a more robust resolution model based on a 72-hour window without agent replies and AI evaluation. This helps ensure tickets that require follow-up aren't mistakenly marked as resolved. This is on-track to launch by the end of July and we are recommending that customers adopt this new generative AI capability.