We are excited to launch AI agents and automated resolutions!

Hey @eric75 

For AI agents Essential, the best source of truth on each AI agent's volume of Automated Resolutions is the Insights tab in Admin Center. Time based analysis isn't possible at the moment (only the last 30 days), however Zendesk is investing more in H2 2025 to improve this reporting experience. We'll be adding a reporting dashboard in Admin Center to make it easier to analyse AR volume, including across multiple agents /  brands. We are also investing in reporting improvements in the AI agents Advanced experience to make it easier to analyze AR volume on a single, or multiple, AI agents.

@ryan48 I look forward to learning more about “email support for AI agent Essential”. At this point, I am on the verge of disabling “autoreplies with articles” due to the false-positive rate. The current behavior has the potential for actual reviewer issues to be missed entirely, since the ticket goes directly to Solved. Fortunately, we email our Agents when new tickets are received, and they will respond, if necessary, even if the system has marked the ticket as Solved improperly. In theory. The first time I discover that a ticket was missed due to an autoresolution false-positive, the feature is getting disabled.

@ryan48 I'm reading the docs on the new “AI agents for email and web form” capability, and I'm going to start testing it out. I didn't see anything in this documentation on how autoresolutions work for this new model, and it doesn't look like the About automated resolutions for AI agents page has been updated. There is a section on email AI agents under “Automated resolutions in advanced AI agents”. Does this now apply to Essentials as well as Advanced, or are these new AI Agents handled differently for autoresolution determination?

Hi Everyone,

We're evaluating Zendesk's AI Bots (Automated Resolutions) and would really appreciate your insights, especially around cost-effectiveness.

Many users have raised concerns about:

• The pricing model based on “automated resolutions”, which can be unpredictable;
• Lack of visibility into how a resolution is counted (vs. real customer satisfaction);
• Difficulty justifying ROI, especially compared to other AI platforms with broader features and lower costs;
• The need for ongoing maintenance in the knowledge base to keep performance stable.

If you've used Zendesk’s AI Bots (excluding Copilot), could you share:

1. Has the cost aligned with the value your team expected?
2. Were there any surprises or challenges in billing/reporting?
3. How do the results compare to what you'd get with external bots or platforms?
4. Do you feel the investment is sustainable at scale?

Thanks in advance!

@vinicius11 

1. Has the cost aligned with the value your team expected? This is difficult to answer due to the recent false positives, billing challenges, and the time/money spent trying to rectify those. Until recently, I would say so, but we didn't get switched to being charged by resolution until our renewal last month, and things have not been great since then.

    

2. Were there any surprises or challenges in billing/reporting? Yes. In fact, I am currently working with a couple people on the Zendesk team regarding allotment confusion, both on my part and Zendesk's. It appears we were vastly overcharged, but I am currently awaiting confirmation.

    

3. How do the results compare to what you'd get with external bots or platforms? I have been looking into this solution heavily since the pricing change, given that API calls for certain LLMs are literal pennies compared to the ridiculous $2 per resolution overage charge of Zendesk's solution, and those AI models tend to be much more robust. Ultimately, I think a self-hosted AI solution using Ollama will be our path forward given that there will be next to zero continuing costs once the upfront cost of a server and setup is complete, unless the pricing of Zendesk's solution is so attractive that it makes me not want to invest the time and money into doing it myself.

    

4. Do you feel the investment is sustainable at scale? Not with the current pricing structure. Scaling at Zendesk's prices versus using API calls for existing LLMs or self-hosting would be much more expensive. Zendesk's AI was great for getting our feet wet in the AI world, but I don't see it being the ideal solution for us forever without some major changes.

Feel free to reach out directly if you'd like any other feedback. While my responses might seem negative, that's in the context of the questions posed here. For the most part, Zendesk AI has been good to us and I would prefer to see it continue to improve rather than spend my time implementing alternative solutions due to cost. 

Hello @adam14 .

Thanks for the reply.
From what I understand, the main issue still seems to be cost, rather than the tool’s effectiveness.
I imagine they might be getting charged for cases that aren’t actual automated resolutions, but rather customer abandonments.
That’s particularly concerning to me, as we have many curious customers who could end up triggering charges for interactions that didn’t really lead to anything.

Hey @david22 

Yes, we’ve just launched AI Agents Essential for email and webform channels. Our recommendation is to migrate to this new capability which will have a much better user experience, automation rates and resolution evaluation. Thanks for calling out the lack of information on the definition in the article, I have the team working on an update to that article now. In the meantime, here’s the definition of an automated resolution for AI agents Essential on Email and Webform:

1. User sends message
2. AI agents Essential generative replies trigger executes and responds with a relevant response based on the knowledge base
3. There are no public agent responses within 72 hours
4. AI evaluation is run on the conversation to confirm that the response to the initial message is relevant and no subsequent user comments seeking additional information or help would negate the automated resolution. For example, the user could respond with “Thank you!” and the AR would count, but if they say “This isn’t what I’m looking for. I still need help” it would not. 

Hey @adam14  and @vinicius11 ,

Love the discussion here and giving your input. A few thoughts from my side if it's helpful

1. $2 per resolution - The current list price is definitely below $2 and this is only if you go into overages above your subscription. You can always limit to using ARs that are in your subscription and there is a free allowance available in every customer's subscription based on your plan level and number of agents so getting started is always free. 
2. Adam - Very sorry to hear about your billing issues, if you'd like me to check in with your account manager on this I'm happy to follow up and make sure we take care of this. 
3. Self-hosting or build it yourself - You definitely need to do what you feel is right for your business. Just keep in mind the infrastructure around AI Agents isn't just calls around LLMs. It requires building a robust Retrieval Augment Generation (RAG) pipeline that's grounded in your knowledge base, tuned to speak many languages, evaluating it's output for accuracy, hallucination and constant upgrades and tuning to that pipeline. As you move to AI Agents Advanced, this includes flows and integration capabilities into third party services, multi-agent systems that cooperate to solve problems. All of this requires technical folks to implement this infrastructure and be able to evaluate the data feedback loops of how it performs and perform ongoing optimizations. If you take the regular case of upgrading to a new model (GPT-4o to GPT-4.1 as as recent example), knowing that this new model will perform well for the current task or prompt is not guaranteed. Being able to evaluate it's performance before you put it in production and tuning prompts and inputs with confidence before going live requires a fair bit of know-how. Be prepared to invest in this work if you go down this path. 

Thanks!

Hi @ryan48 ,

Thanks for your reply and the clarifications.

That said, I want to respectfully push back on a few points from a customer’s perspective:

• While you mention the list price is “definitely below $2,” Zendesk's own pricing page (screenshot attached) shows a clear $2 pay-as-you-go rate for automated resolutions. That’s the number we see, and plan around.
• The “free allowance” per agent/month is extremely limited. For larger teams or support, heavy operations, this barely scratches the surface and ends up pushing customers into overage charges very quickly.
• In theory, charging per resolution sounds fair, in practice, it becomes unpredictable and expensive. For example, if a company has 8,000 tickets/month and 5,000 are resolved by AI, at $1.50–$2 each, we’re talking $7,500 to $10,000/month, just for automated resolutions.
• Even assuming 100% accuracy (which isn’t the case), this cost is far higher than other market tools offering comparable performance, some charging around $700/month flat, regardless of volume.
• We want to use native Zendesk solutions. But it’s becoming harder to justify when the cost structure seems misaligned with what competitors are offering for the same (or better) results.
• Lastly, just reading through the public threads and community forums, it's clear we’re not alone, pricing concerns are one of the most common and consistent complaints around AI Agents.

We genuinely want Zendesk AI to work for us, but for that to happen, cost transparency and global affordability need to be part of the roadmap.

Thanks again for listening.

@ryan48 Thanks for the reply.

1. Yes, I am aware that $2 is only for overages and stated it was for overages. That said, $1.50 is still far too expensive in my opinion, especially with the false positive rates we are seeing, and 10 free automated resolutions per month per agent license feels insultingly low, doubly so when those get used up for false positives. Considering we had unlimited resolutions for the same price when we were pushed to this product, it seems like a massive bait-and-switch, which would be easier to swallow if the way that it evaluated what was an automated resolution and wasn't was accurate, but it really isn't, at all.

2. Thank you, but I am already engaged with my account rep, and he involved someone else whose name escapes me at the moment, but was more involved with the AI side of things. They were both puzzled by what they were seeing as much as me. I will hopefully hear back from them soon.

3. Yes, self-hosting or building out another solution ourselves is complex and costly, which is why I am here providing feedback hoping that Zendesk comes up with something more customer-friendly so I don't need to go through the effort. I am familiar with RAG and training and prompting other solutions (ChatGPT, Claude, Retell), and in fact we have already done so with an AI phone agent (a significant gap in Zendesk's AI offering that I am sure is being worked on). But at a monthly cost of $50 per agent per month, plus automated resolution pricing, we are already paying thousands of dollars a year. Investing a few thousand more dollars up front for another solution will be far better on pricing in the long run as we continue to scale, and if we're being honest, a far better customer experience if we're using one of the traditional LLMs that can have a real conversation rather than treating every message as independent. Sure, we'd lose things like auto-assist, entity detection, etc., but to be honest, we pay for the AI for the chat bot. Losing the other features would hardly impact us.

Look anywhere on the internet. Whether on this post (with 9 pages of negative customer feedback), Reddit threads, or competitor websites highlighting this pricing structure as an example of how their product is superior,  the automated resolution pricing is being discussed extremely negatively. There are NO positive reactions. That should be a HUGE red flag to Zendesk.  We all know that AI is the future, and actively steering customers away from your AI product does not feel like a sound business decision to me. When we first adopted Zendesk's AI, we LOVED it. We WANT it to succeed. I'm not sitting here typing all of this just to be like “This ain't it; we're using something else”. We want THIS. We just want it in a pricing model that makes sense. Your customers are enthusiastically providing you feedback that other companies would kill for. Please listen to them.

@ryan48 Thank you for your response. It definitely sounds like the autoresolutions model here will address the false-positive and manual-Agent-steps-required edge cases we've been dealing with via “autoreplies with articles”.

The new AI Agents Essential for email and webform channels was activated in our instance yesterday and I started playing with it. I'm seeing behaviors that aren't working quite as I expect/as documented, and I'll likely follow up with the support team to drill into it.  I'll also post comments on the appropriate pages, rather than cluttering up this thread which is about autoresolutions.

"If you want to turn this feature off for now, see Turning the AI agent tickets feature on or off. However, note that starting on:

February 18, 2026, this feature will be switched on by default but can still be turned off. 
May 4, 2026, this feature will be on by default and cannot be turned off."  Reference Link 

We are concerned the impact this is going to have on our support team, and the admins and managers that have to work ensuring tickets are created accurately.  We try to run a high touch support for our customers, even in Chat/Messaging, so we do not want AI Agents handling tickets except in a very few edge cases.  I think making this feature a requirement and unable to turn off or remove is an injustice to your customer base.  You are taking control of what works or doesn't work for us and forcing us down a path that could be detrimental to our relationships with our customers.  
We request that you reconsider the inability for Zendesk customers to be able to turn off this feature.

@will14 that's not entirely true. AI agent tickets are being created regardless of them being escalated to a human! 
Zendesk actually shows and inform that the ticket is not editable and stay open for a while.
https://support.zendesk.com/hc/en-us/articles/9204149016346-Understanding-AI-agent-tickets-for-AI-agent-only-conversations