Howdy hunters, Matt here from the Intercom product team.
Just a couple of short months after the launch of Custom Bots, we’re thrilled to announce Answer Bot–the newest addition to our family of conversational bots powered by Operator™.
It combines a best-in-class machine learning engine with a powerful curation tool to automatically identify your most common customer questions and answer them before your sales and support teams even see them.
We’ve been building Answer Bot for the past year and from the private beta we’ve seen that on average it can instantly resolve 29% of your most common customer questions and reduce your response time by 44%.
Here’s a quick summary what you get with Answer Bot:
Instant resolutions, 24/7
Give your customers precise answers even when your team is busy or offline.
Control exactly what Answer Bot says, so your answers are always on brand.
Richer answers with apps
Show people the status of their order, embed help articles, let people register for webinars, and more.
Immediate efficiency gains
Improve your customer response time by 44% while freeing up your team for more valuable work.
No dead ends
Always provide an option to wait for the team and talk to a human to get help.
You can add Answer Bot to any paid subscription, and we have special introductory pricing, starting at $49/mo for a limited time. It’s free to try for 14 days, and you can set your first answer live in just 5 minutes. Head over to our website learn more.
One of the best parts about working on Intercom’s support team is being able to beta test our new features early. With Answer Bot, we got to work with it extensively, and see how powerful it can be, before it went live. Answer Bot has made our job a lot easier, and I’m sure our customers will feel the same!
Answer Bot has been a game changer for us here at Intercom and has quickly became my new favorite coworker! I’m so excited to see the different use cases and personalities - this bot really does it all.
Answer Bot has changed the way our team works. From training the bot to answer our customers' most pressing questions to being able to instantly see reporting on the metrics that matter most, Answer Bot has it all and then some. This is not just a product; it’s the evolution of how we work and Intercom is leading the way.
It’s been a great adventure building this with an amazing team of people across London, Dublin and San Francisco. Always exciting to get to build something this innovative!
Thanks everyone for checking this out!
We’re excited. The results we saw in the beta were wild. Real prospects and customers getting real answers to real questions. Instantly. And without anyone in the company even having to know about it.
My favorite part about Answer Bot is the way it’s connected with the rest of the Intercom platform. It can reply with just text (and emoji, gifs, links, etc.), but can also launch apps from our App Store. E.g. If a customer asks about the status of their order, Answer Bot can launch the Shopify app to show them real, live, instant data, right inside the Intercom Messenger!
This is a BIG, deep product that the team put a lot of love and care into to make simple. We’ve been iterating hard on it for over a year. I hope you’ll try it out and let us know what you think and how we can make it better for you.
@eoghanmccabe @mattnhodges congrats on the launch!
I know we had a fiesty exchange a year ago, but today I've had my coffee and disabled RAGEMODE so hopefully we can have a more cordial discussion. (And FWIW, I could have been much calmer in my criticism (which I clarified later in the back and forth) so I apologize for the hostility in my tone.)
Presuming bygones can be bygones — I'm curious about some of the machine learning that you're doing — given that I was working on something related to questions and answers... Are you using pre-trained models that are better suited for customer queries? Or are you using a some kind of generic question-answer model? IOW, given the data that intercom could see across all the customer support interactions on its platform, it could build a pretty robust model for inbound customer inquiries (requiring an effective privacy framework, but I digress)... From the docs, it seems like it's pretty manual — similar to wit.ai. I think I'm asking about the underlying machine learning approach that you're taking to Q & A with Answer Bot.
I'm also wondering about one keeps answers up to date... considering the wide range of questions that customers may have that may cover a broad range of product versions that might be in the wild (iPhone X vs iPhone XS, etc)? How does Answer Bot not end up taking a customer down a long tree of qualifications to figure out what model of a product they might be asking about? Easy questions about resetting a password are pretty evergreen and universal, but what about variant responses that change based on a specific product SKU?
What do you see as the long term aspiration for Answer Bot? How much is about just dealing with the top of the funnel questions vs actually going deeper into specific customer questions or providing personalized guidance and advice (which then, I suppose, is when you fall back to the human-in-the-loop)?
Lastly, I found a missing page linked at the bottom of this page ("You can read more about creating effective answers here.")!
@eoghanmccabe @mattnhodges @chrismessina
We love fiesty exchanges! Thanks for starting good debates.
1. Are you using pre-trained models that are better suited for customer queries? Or are you using a some kind of generic question-answer model?
We'll likely post a blog with details. But at a high level, the ML we're using here is custom-built to deal with how people ask questions in messengers. We tried using models trained on sources like Wikipedia, but we got better results when training on conversational data.
> From the docs, it seems like it's pretty manual...
Machine Learning just isn't yet powerful and reliable enough to let bots fully automatically just generate answers to questions. To get great experiences, you have to curate the bot. But writing manual rules to respond to messy conversational language is too much work. With our Answer Bot you aren't manually specifying exactly what it says and to whom. Instead, you train it by giving it example questions for each Answer, and the machine learning figures out when its right to give them. We think this combination enables people to use powerful machine learning to deliver well-curated answers. We also learned during an extensive private beta that businesses wanted control, so we feel we’ve struck a good balance. We built our own custom curation tool and ML engine because we couldn’t find anything in the market that addressed exactly this.
2. I'm also wondering about one keeps answers up to date...How does Answer Bot not end up taking a customer down a long tree of qualifications to figure out what model of a product they might be asking about?
Keeping answers up to date requires a human! There’s no magic to that. But as for long conversation trees, the way we prevent it is that we only let Answer Bot “disambiguate” once. This means Answer Bot will only proactively ask “Did you mean a, b, or c?” one time. If the user chooses to keep asking for more options though, we’ll let that go as long as Answer Bot has more options for the user to choose. Our testing has shown users are comfortable going a couple levels deep, but rarely further.
3. What do you see as the long term aspiration for Answer Bot?
For sure the best place to start with Answer Bot is the most common questions you get where the answers are pretty straightforward. But we’ve already seen in our beta where customers used Answer Bot for questions that required a human, but Answer Bot could ask for more details to help speed up the conversation. We haven’t captured that value in our reports, but a lot of times Answer Bot is helpful even when it doesn’t full resolve a question. By adding apps to answers, we think that you can already personalise answers (for example using the Shopify app to let people check the status of their order). As for the long-term future, honestly we want to see where customers decide to take this themselves. For some companies, Answer Bot will be able to shoulder a big burden of their conversations. For other companies, it will carve out a very discrete part of their conversations. Either way, we’re optimistic Answer Bot can be a critical parts of most companies’ support tools. Down the line -- humans will need to be less and less involved to make this work well.
@eoghanmccabe I love Intercom, and I've been hoping for a product like this since you guys launched Operator. Awesome, it looks great!
The problem is, your pricing model - charging per user stored in Intercom, regardless of whether they're active or not. We now have ~25k users in our database, but most of them are completely inactive. The launch of this product got me to check - we're now paying Intercom $1,000 per month, which is a lot for a small startup.
To keep using Intercom, we're going to have to build something that automatically deletes users from Intercom (your reps actually suggest this form of gaming) if they're not active for some period of time. But it diminishes the value we get out of the service, because it means we won't have that history with that customer if they re-engage, and it means that we can't use the chat system.
IMO you should charge based on active users in the system, or charge based on the quantity of communications being sent to users.
@andrewmason - Hey Andrew! Another Andrew here, and I'm one of the Support Engineers at Intercom. Just wanted to chime in really quickly to let you know that we'd never want you to lose that connection with your customers while using Intercom. If your team does choose to go about the route of regularly archiving inactive users from your Intercom Workspace, with a robust Intercom integration, any users that do eventually become active should have their historical data (conversations, custom attributes, tags etc.) restored. Definitely feel free to hit us up to make sure your installation is suited to handle this behavior!
@eoghanmccabe Love it. This is the future! Intercom already powers most of the internet, so it was only a matter of time, but man, this is going to make customer support so good!
@eoghanmccabe @andrewmason I'm with Andrew on this, loving the product, trying to get more out of it but the pricing model doesn't make sense to us =/.
Not that we don't want to pay more it's just not charging us vs the value we receive, it's charges too much for just stored data (inactive users).
Wow, this looks really smart and easy! Looking forward to trying it out.
Looks very powerful! Have you tried it in other languages than English?
@the_minh 👋 just English for the moment. We ship early in order to learn–it's faster to ship in one language and validate the product, make it better, before going multi-language. I know that that doesn't help a lot but it's the rationale at least. 😄
Welling up with pride and appreciation for all the people and teams across Intercom who along with our customers brought this to life. Excited to see how people use it and excited to further evolve in the weeks, months and years ahead. 😍
Total world domination, but in a good way. Thank you Intercom, only you can do it the right way when it comes to business <> human communication.
Boom! This is awesome. I was working on custom bots in Intercom a few days back. This ML addition will take away the pain of support team who often needs to answer the same question, over and over again.