Provide your Rezzy AI Beta feedback here!

(page 1 )

• Page 1
• Page 2

Hello, Rezzy AI beta users!

Now that Rezzy AI has been out in beta for a few weeks (and while there are the "thumbs up" and "thumbs down" options in the Unified Inbox on Drafts), we wanted to open a forum to collect feedback from those using Rezzy AI during the beta.

So comment below, and thanks in advance for helping us make Rezzy better each week!

-We need the ability to copy FAQ's from property to property quickly and easily.  

-Rezzy refers to checking with the host quite frequently. This is jarring to guests who don't know that they are not talking with the host.  This should be able to be customized to the time of day.  If Rezzy doesn't know something, there should be an option for it not to respond.  

Andrew and Brittney, thanks for kicking off the feedback here!

-We need the ability to copy FAQ's from property to property quickly and easily.

Some significant enhancements are coming for FAQs and property application/copying and account-level options.

-Rezzy refers to checking with the host quite frequently. This is jarring to guests who don't know that they are not talking with the host.  This should be able to be customized to the time of day.  If Rezzy doesn't know something, there should be an option for it not to respond.

You can change the setting to refer to Team instead of Host and there is a signature line you can give Rezzy to call it whatever you want, if you want to disclose they aren't speaking with the host; however, time-of-day constraints and other enhancements are coming to help in this area as well.

Thanks again, and keep the feedback coming!

First of all - thank you!  We are LOVING Rezzy!  SO much better than some of the other tools we tried!

 

Here are some preliminary notes and thoughts from the first week:

1.  Send, Edit, Delete icons (and now thumbs up and down) are tiny and close together.  It's too easy to accidentally hit send instead of edit....  Could there be a confirmation before sending?

2.  Could we add a due date to a task?  and maybe even flag when due?

3.  Definitely need the ability to copy either all of the FAQ sections from one property to another or even more appropriate long term would be to copy a single question across all properties. 

4.  It would be helpful to have FAQ not just for properties, but at the account level as well.  

5.  It would be nice to have SELECT custom fields (not all) that are available to Rezzy.  This would be at the account level, property, booking, etc.  Just an option within each custom field to allow Rezzy to see/use it or not.

6.  When a task is created, the reservation/guest thread will link to it, but when you're in tasks, it shows the reservation, but doesn't link back.

7.  Sometimes Rezzy is too quick to validate the guest complaints in an over the top level of empathy.  Need to figure out how to tone that down.

8.  Would love to have the ability to channel the Rezzy conversations to an echo dot inside the property. :)

Thanks!

Jenny

 

 

Thanks for creating this for more real-time feedback rather than having us email the support desk.

--

Similar to ChatGPT, Rezzy is only truly helpful if it can actively learn from us. There are a few key ways this could be happening: 

1. The system could scrape all the message history from the Communications Log for each property. From there it could actively learn tone, length, answers to specific questions -- it could even auto-generate an FAQ for that property based on questions that came up repeatedly in the message history. 
2. If we edit Rezzy's draft, it should be learning from our edits. 
3. If we ignore Rezzy's draft, it should still be continuing to learn from the replies we are creating. 
4. There should be some account level FAQs that could be pre-populated based on our settings and the data from the Communications Log, and then we can of course still add to or amend the FAQs as needed. 

I also echo Jenny's comment about how close/crowded the action icons are within the tiny draft message box!

+1. We should be able to define FAQs in the Rezzy setting, and from there select which properties they apply to, so we don't have to duplicate them in any way (much easier to maintain too) 

I like the start of Rezzy and where it is going. However, for those of us who really want to drive direct bookings, I would like rezzy to help with my text messages on my google voice line. I try to divert conversations with guests to my platform and space so that they remember my brand and may possibly re book with me rather than the OTA's. Is there any thought going into having rezzy helping answer text messaging from guests from a phone number we designate? I would love to teach rezzy to move the converstion to sms for our business line and respond to the guest from there. This would also help for those who have on call service and want help with answering those redundant questions via text. I would like to see Rezzy handle all the prospect questions from the OTA platforms but once they book, it shifts to our business line of messaging. 

I think Rezzy if off to a great start, and I have been able to send its suggestions without any edits about 30% of the time. The other 70% the responses are a) too long - no human would write a whole paragraph the way Rezzy does, and it's pretty obvious an AI bot is talking, b) as others stated, WAY too empathetic. I get that customer empathy is important, but it's over the top. Less is more. I second what others are saying - the icons are super tiny and in light gray, and close together - you have users over the age of 40!

Came here to give my suggestions but it seems everything I was thinking has already been suggested.  The account level FAQs/ability to copy FAQs to other properties is a biggy... its taking such a long time to put these in each property... I'd say it's a couple of days of manual work!!

 

Also I would love to be able to export the inbox to help me think of FAQs... when you have a lot of properties its tricky wading through everything.. Of course there are some obvious ones but there should be more.

 

Would be great if Rezzy didn't agree so much with the guest... I haven't used one of Rezzy's replies yet, they've all been deleted.  But that may be because filling in the FAQs takes so long!!

 

I use chatGPT to rewrite my emails to guests on a daily basis and it now knows my tone and company values etc so well that i have to change very little.  I hope Rezzy can get to this level as it would be so much easier.

Hi - I noticed Rezzy is getting the day of the week wrong when she created a draft reply. I read the draft quickly and sent it. I assumed she knows the dates and she is accurate. This has happened to me twice. 

Here is what happened today. A guest checks in Monday Sept 1 for one night. Rezzy said Sunday Sept 1 in the message and I didn't catch it. 

-We need the ability to copy FAQ's from property to property quickly and easily.  

-Rezzy refers to checking with the host quite frequently. This is jarring to guests who don't know that they are not talking with the host.  This should be able to be customized to the time of day.  If Rezzy doesn't know something, there should be an option for it not to respond.  

i agree with the above. I think there is / should be a way to se the voice to that of the host. Rezzy is also too formal for a typical 'friendly chat style' and honestly goes into too much detail - repeating the guest issue and repeating what they are looking forward to in a very bot like manner. I understand it's nice to be chatty, but you can really see the bot. I find that unfortunately I can't use much of the text the bot proposes. 

 

Hi - I noticed Rezzy is getting the day of the week wrong when she created a draft reply. I read the draft quickly and sent it. I assumed she knows the dates and she is accurate. This has happened to me twice. 

Here is what happened today. A guest checks in Monday Sept 1 for one night. Rezzy said Sunday Sept 1 in the message and I didn't catch it. 

Hey Mike,

This is a bug we will be addressing soon. Thanks for bringing this to our attention.

I've created a Latest Updates forum topic for Rezzy AI so that everyone can stay informed with what's new and what's coming down the line soon. Check it out here, and add yourself as a watcher if you'd like notifications when updates are provided!

https://www.ownerrez.com/forums/rezzy-ai/latest-updates 

Despite asking Rezzy to reply in English, as the guest asked a question in Spanish, Rezzy is replying in Spanish. Also, when a guest writes in a foreign languabe (like German, Portuguese, Spanish), Rezzy is not translating ther question to English, so we know what the guest is asking.

Despite asking Rezzy to reply in English, as the guest asked a question in Spanish, Rezzy is replying in Spanish. Also, when a guest writes in a foreign languabe (like German, Portuguese, Spanish), Rezzy is not translating ther question to English, so we know what the guest is asking.

Hey Ariel,

Thanks for reporting this! I do see in your Rezzy logs where Rezzy appears to have responded in Spanish at least three times even though it's set to respond only in English. We'll need to research this and see what's going on there.

At present, Rezzy does no translation for the host. This will be implemented at a later date, but it's a work in progress at this time.

I really appreciate the progress with the Rezzy AI beta so far and am thrilled to be part of the test group. I wanted to share a few observations that I hope may be helpful, as well as add a +1 to some comments others have already raised.

 

 

• +1 on improving the FAQ module so it’s easier to create once and apply across multiple properties. I see strong potential for using “tags” to extend attribution options. We already tag by community and geographic area, so being able to parse FAQs by tags (for example, beach access details or resort rules) would save a lot of duplication instead of tagging 20+ properties individually.
• I’ve noticed cases where an FAQ I created isn’t referenced in draft responses. For example, guests sometimes report the refrigerator “not cooling” when it simply needs time to bring newly added warm items down to temperature. I wrote an FAQ for this and tested it in the sandbox, but Rezzy didn’t pick it up the way I expected. I experimented a bit with different ways to pose the question, but did not get it to trigger the way I had hoped. Can/should the “question” be blown out to include things like keywords or differences on phrasing, etc.?
• +1 on Rezzy improving each time through individual response feedback, not just taking a thumbs down, but also learning from the corrected response, (which it might already be doing). I’m not sure if Rezzy is building a RAG dataset behind the scenes, but all those past guest conversations would make a powerful resource for training future replies. 
• I’ve leaned heavily on ChatGPT over the past year to craft consistent guest responses, even building a ~2000-character “brand voice” prompt that I adapted for Rezzy and inserted into the “conversation tone” field. Responses, length, style, etc… are now more closely aligned with the tone and style I wanted, and I imagine many others could benefit from using this same feature.
• In some cases, it seems like when I delete an FAQ and recreate a slightly different version, Rezzy still defaults to the original draft logic. Longer-term, it would be great if the FAQ module could detect conflicts or overlaps between older and newer entries and even invite the user to merge similar FAQs that could be offering conflicting information.
• One small but noticeable item: some draft replies have responded in a guest draft something like “Sorry. I don’t know who Shelena is.”, when the initial message was a guest thanking my wife Shelena. A master field for things like staff names might solve this and keep replies more natural. Along similar lines, it would be useful if Rezzy could also store/reference company basics such as hours, address, and critical contact info so those get woven consistently into responses.
• Rezzy created a really solid response to a guest who asked if they were able to just go straight to the property and check in, to which the draft included the door code and basically invited him to come on over. Really great response, but it might be a good thing to have a security override option, or override by time frame, tied to check in and check out times that would not allow certain specific details like door codes, or even addresses to be shared. (Full transparency, we include door codes, stored in the Internet field of the property, along with all other arrival information that we then merge into the outbound welcome message about half an hour before arrival, which is likely where the door code was pulled from.)
  
• I love the tasks concept, but might be able to get more use out of the feature if it could automatically create tasks that would integrate with our breezeway implementation as an option for certain things like maintenance. In the current form, it would be more helpful for us to disable the tasks feature as it does create some additional overhead along with notification overload if kept enabled.
• The new Directives feature sounds awesome, but I am not totally sure how to best use it yet. Do you have some examples that you could provide and what is being processed on the backend?

That’s all for now! Thank you again for allowing me to be part of the beta test and keep up the great work!

 

Michael

Epiqhost

Santa Rosa Beach, FL

Hey, Michael! I'll speak to a few of your points here.

+1 on improving the FAQ module so it’s easier to create once and apply across multiple properties. I see strong potential for using “tags” to extend attribution options. We already tag by community and geographic area, so being able to parse FAQs by tags (for example, beach access details or resort rules) would save a lot of duplication instead of tagging 20+ properties individually. - A recent update introduced new shared FAQs, accessed from the global Rezzy settings area, to see and work on shared FAQs across properties.

Mapping FAQs to tags is an interesting concept. Gives us something to think about!

I’ve noticed cases where an FAQ I created isn’t referenced in draft responses. For example, guests sometimes report the refrigerator “not cooling” when it simply needs time to bring newly added warm items down to temperature. I wrote an FAQ for this and tested it in the sandbox, but Rezzy didn’t pick it up the way I expected. I experimented a bit with different ways to pose the question, but did not get it to trigger the way I had hoped. Can/should the “question” be blown out to include things like keywords or differences on phrasing, etc.? - Taking a look at the Rezzy logs in your account, it sounds like she's getting confused because no property was selected when the sandbox request was made. I did some testing in my own account and received a similar response for one of my FAQs with no property selected.

Generally speaking, a guest with an existing booking will have a property association that Rezzy can pick up on and answer appropriately. Have you tried selecting a property in the sandbox to see if the results were different?

+1 on Rezzy improving each time through individual response feedback, not just taking a thumbs down, but also learning from the corrected response, (which it might already be doing). I’m not sure if Rezzy is building a RAG dataset behind the scenes, but all those past guest conversations would make a powerful resource for training future replies. - This is something we're working toward! 

In some cases, it seems like when I delete an FAQ and recreate a slightly different version, Rezzy still defaults to the original draft logic. Longer-term, it would be great if the FAQ module could detect conflicts or overlaps between older and newer entries and even invite the user to merge similar FAQs that could be offering conflicting information. - Great idea! We'll take this into consideration. 

The new Directives feature sounds awesome, but I am not totally sure how to best use it yet. - I think you'll find this article helpful: https://www.ownerrez.com/support/articles/rezzy-ai-common-issues-questions#directives

 

For now, thumbs up / down is how we're gathering intel for learning opportunities. Please do make use of those thumbs! You're also welcome to share specific thread links and messages here in the forum if there's something particular you'd like us to drill into!

So far I think it's pretty useful. I think this will be great for larger scale users with dedicated staff to monitor the task board frequently.

Larger Portfolios
At scale I think you will realize the most value from RezzyAI. This probably is a fairly obvious observation. But the main value I see from it is that it will prevent follow ups from falling through the cracks. The ability to live on the task board instead of having to read EVERY message becomes a lifeline. You can let the automation do the heavy lifting in this instance and focus on the things that actually need attention instead of having to worry about being a super engaged chatty Cathy. This is one of those things that becomes necessary when you have dozens of listings to monitor.

Smaller Portfolios(my scenario)

Unfortunately, I think the value proposition dwindles a bit for smaller accounts. My account has barely over 10 properties. And accordingly, it feels like the AI creates clutter for me. Nearly every message generates a task. Even if it's just a comment about it being their wedding anniversary or something. Perhaps that means I need to spend a bit more time marking the tasks and actually clearing the task board. Our messaging volume is not that high. So it feels like the AI is overcompensating a bit. 

Communication Style

I would like it a bit more if the suggested draft replies were less verbose. AI tends to try and add too much detail and wing it on the details sometimes. And it just feels like more work to craft the narrative to my style. I'm not sure how I can train the AI to be closer to my communication style. I probably need to tinker some more.

AI Data Access

I recently had an amusing yet concerning interaction with Rezzy. Somebody addressed me by my first name, just saying thank you, and Rezzy generated a task "guest is thanking some person named Justus. Need to figure who that is and why they are thanking them for context." Kind of funny, but also concerning that Rezzy can't figure out that I am the name on an AirBNB API, VRBO API and that I am also an Owner in the PM module as I have my personally owned properties under me for reporting filtering/grouping and other stuff. This begs the question, what is RezzyAI limited to in regards to the system data it looks at?

Sometimes Rezzy creates replies that are not accurate, I suppose due to not correctly interpreting my Directives or not having the context of the ongoing conversation...e.g., we have an automated process where guests request an early check-in via a webportal, upon which we send an OR message confirming the request was recorded in our system. If we finish cleaning the house before 4PM, we send another message, also thru OR, granting access to the house. 
Here you have an example, where the guest asks if we know about the early check-in request, and Rezzy says, incorrectly, that it was already granted (which is not true)

Here is a more accurate response from a competing AI tool, Conduit

Rezzy's task-creation is out of control. So many tasks for all.the.things!! It might be helpful to have the thumbs up/down for tasks so we can let the system know when it's unnecessarily creating tasks. Like this one after a guest messaged just to say she checked in and loves the space!