Thank you detector not working as expected...

Hi! 

It's pointless to try fixing this until Freshworks starts actively maintaining it—and right now, they aren't.

We've been using the system for about two years, and I can confidently say that the Thank You Detector project has been neglected. The website https://thankyou.freddyproject.com/ is practically dead, and there hasn't been any development or improvement in the past two years.

In our experience, the Thank You Detector fails in every fourth email. For example, it triggers on any "Thanks + question" message or even on a blank email with an attachment. The last change Freshworks made to this project was simply moving it under the paid Copilot licenses and slapping a "beta" tag on it.

Honestly, it's better not to rely on it. It's a trial and error experience that can be painful for a business, and it's practically useless since the Freshworks isn't maintaining it.

Hello @raghavkannan @Mykola Zhuravel - I’m sorry to hear that. I’m tagging some of our SMEs to help out if they any any updates for you @sukanya.arunachalam @Aswin Ayapparaj @Preethi Cx expert @mahendarsingh 

Has there been any movement on this - no replies since ​@alyssia.correa tagged the SMEs ?

It seems the silence speaks volumes. Two months have passed since @alyssia.correa tagged the so-called SMEs, yet no response from them. Clearly, it's not just the Thank You Detector that's been abandoned—the forum itself seems to be in a similar state.

The Thank you detector continues to perform as poorly as 2 years ago, and this lack of engagement only reinforces what many of us have suspected: Freshworks has no real interest in fixing this or even acknowledging the issue.

​@Mykola Zhuravel have you logged any support tickets about the issue?

​@RobCrossHM, I raised this with Freshworks support back in early 2023, and their response was, and I quote: "You can start a topic on the developer forum, where our product team and development team will review and validate the request." 😂

Two months ago, we asked through our partner about the status of this functionality, and the answer was blunt: there are no updates planned. The feature isn’t being developed further; they’re only analyzing possibilities for its future development. According to our partner, there’s no expectation of any changes in the near term.

So sure, you can log 10 support tickets if you want—but what’s the point? Freshworks has clearly abandoned the Thank You Detector project. Don’t waste your time here, and definitely don’t try to fix it yourself.

​@alyssia.correa could any of the SMEs review this?  I’m having the same problem with the thank you detector, and it’s really frustrating.

I know that this used to work, as I used it at another organisation while it was in Beta

​@Mykola Zhuravel 

I’ve followed this up with Support, and it seems that I’ve got it to work.  this is my setup, what are you using?

Thanks ​@RobCrossHM , but the problem isn’t in the setup — it’s in what the system defines as a “Thank you”.

Pretty much everyone uses this feature to avoid reopening tickets when a customer replies with a casual “Thanks” — that’s the core idea behind the Thank You Detector. And in theory, it sounds great.

In practice, though? Here's an example from just 2 hours ago on our end:
A customer wrote us asking to change their password — no “thanks”, no appreciation, just a request. Freddy detected it as a thank you note.🤦🏻

I'll attach a screenshot to show you exactly what we're dealing with.

Wow, in my testing I've seen it completely miss thank you notes, but not get it entirely wrong.  I have flagged this thread to the community leaders, as it’s not getting the attention it needs.  See what that brings!

Thanks ​@RobCrossHM  — I can actually share a bunch more examples like the one I mentioned earlier. This isn’t just about occasional misses — it’s about wildly inaccurate detection.

We’ve seen it trigger on completely irrelevant replies like:

• A customer simply responding with “Deposit insurance”
• Or replying “Yes please” to a question from an agent like “Do you agree?”
• Or writing “Haha good one… Ok cools…” in an ironic tone
• Even angry messages like “You guys s*ck ballz f**kers” can sometimes trigger the tag
• And if a client writes something like “Chiudete il mio conto, grazie” — boom, the system slaps on the thank you note tag just because it sees “grazie” in there, context be damned

There are borderline cases too — like “All good” — which you could maybe stretch to interpret as gratitude, but even then, it’s shaky.

Freshworks tries to present this as some kind of AI-based smart feature, but that’s just not true. If you put these same inputs through actual AI models like ChatGPT, Gemini, or Copilot, none of them would mislabel them as thank-you messages. That’s why I strongly believe this is just a rules-based system behind the scenes — probably something as basic as word matching or character count triggers.

The real kicker? This feature has worked the same broken way for two years. Then out of nowhere, Freshworks decides to tuck it under the Freddy Copilot branding and slap a beta tag on it — which feels more like a strategic move to avoid accountability than any real product improvement.Appreciate you flagging it to the community leaders — but honestly, if they haven’t taken it seriously by now, I doubt anything will change without public pressure.

 

Thanks ​@RobCrossHM  — I can actually share a bunch more examples like the one I mentioned earlier. This isn’t just about occasional misses — it’s about wildly inaccurate detection.

We’ve seen it trigger on completely irrelevant replies like:

• A customer simply responding with “Deposit insurance”
• Or replying “Yes please” to a question from an agent like “Do you agree?”
• Or writing “Haha good one… Ok cools…” in an ironic tone
• Even angry messages like “You guys s*ck ballz fkers”** can sometimes trigger the tag
• And if a client writes something like “Chiudete il mio conto, grazie” — boom, the system slaps on the thank you note tag just because it sees “grazie” in there, context be damned

There are borderline cases too — like “All good” — which you could maybe stretch to interpret as gratitude, but even then, it’s shaky.

Freshworks tries to present this as some kind of AI-based smart feature, but that’s just not true. If you put these same inputs through actual AI models like ChatGPT, Gemini, or Copilot, none of them would mislabel them as thank-you messages. That’s why I strongly believe this is just a rules-based system behind the scenes — probably something as basic as word matching or character count triggers.

The real kicker? This feature has worked the same broken way for two years. Then out of nowhere, Freshworks decides to tuck it under the Freddy Copilot branding and slap a beta tag on it — which feels more like a strategic move to avoid accountability than any real product improvement.

Appreciate you flagging it to the community leaders — but honestly, if they haven’t taken it seriously by now, I doubt anything will change without public pressure.

Hello Folks!

Sorry about this experience — and thank you for taking the time to explain it so clearly. I have already flagged this to the Product team and they are taking a closer look. I’ll circle back here with the updates I receive.

Hello everyone,

I've just given a very negative feedback to our TAM regarding the Thank you detector (or rather "no response to bring" detector because it doesn't only detect thank you...). 
I'm very sad to come across this article and discover how this feature, which seems so “basic” to us, is not being prioritized and improved in 2025 under Freddy's label... The idea behind it is very interesting and much needed by all your customers, but unfortunately the results aren't there.

We decided to measure the success rate of the thank you detector. In short, our Team Leaders check every ticket detected as “thank you” and indicate in a field whether it's OK or KO. 
Our results over the last 3 months have been rather disappointing... 78% OK in March, 74% OK in April and 70% OK in May... We're losing 4% every month, while the system is supposed to be learning from previous tickets...

Is there any real plan to improve this feature?

​@SophiaEM this is some great insight - I’ve created automation to add a tag each time the thank you detector kicks in, need to wait a few weeks to see what our numbers are like.

If anyone from Fresh reads this thread, its’ this experience with the thank you detector, that has put me off paying extra for the agent Freddy licences.

Hey Folks,

I’ve escalated this to my leadership, and it will be looked into. Please allow me some time to gather the updates and get back to you.

This is a great example of it not working  - this is definitely a thank you note

​@Kamakshi V did you get any feedback from Leadership?  Here’s another example of it not working

Another example of it just not working.  Is anyone from FreshWorks looking at this thread?

​@Kamakshi V  This isn’t just an annoyance, this means that my FCR/Ticket reopen stats aren’t reliable

Hello, 
for information, our TAM advised us outright to stop using the Thank You detector, as we had more problems than solutions with this feature, unfortunately. I have to admit it's extremely frustrating and a shame because it could be so useful to all your customers. 

Hi Everyone,

I’m Sripriya, one of the SMEs working with the product team. I want to start by sincerely acknowledging the concerns many of you have shared here. We completely understand how important it is for the “Thank You Detector” to work reliably, and I apologize for the inconvenience and disappointment this has caused.

This feature is currently in Beta, and while the intention is to reduce noise and make your workflows smoother, we clearly recognize the gaps you’ve highlighted — from accuracy drops to false positives in different contexts. Your examples are extremely valuable, and I’ve already shared them with our product and engineering teams.

To clarify, this isn’t the final version of the feature. The team is actively reviewing how the detection logic works and exploring improvements to make it more context-aware. I’ll continue working closely with them and will keep this thread updated as soon as we have more concrete progress to share.

We truly appreciate your patience and honest feedback — even when it’s tough to hear. It helps us prioritize better and push for improvements that matter most to you.

Hi ​@Sripriyadarshini Rajendran 

Thanks for replying.  Do you have any idea of an ETA on these improvements?