How we built AI in Dovetail—Channels

Great products and experiences come from teams that really know how to put their customers first in every choice they make. But how do the best teams pull this off? They make the most of customer feedback.

Customer feedback offers direct insights into user needs and preferences that can inform more customer-focused decision-making. For many teams though, extracting valuable insights amid a deluge of customer feedback can be like trying to drink from a firehose—overwhelming and futile. This is because often, organizations are dealing with loads of data—support tickets, NPS results, survey responses—coming from various sources. For teams wanting to make better decisions, transforming overwhelming feedback into clear and focused decisions is key.

Dovetail’s Channels can help you get there. Channels is for any team wanting to uncover real-time, actionable insights from continuous user feedback. It does this by identifying trends that are hidden in mountains of data, saving you hours previously spent sifting through spreadsheets or support tickets. With Channels, you can channel (pun intended) your energy into the most important stuff—enhancing customer satisfaction.

So, why would you use Channels? Channels is for CX professionals wanting to look for trends to help determine where to prioritize growth opportunities. It’s for designers and PMs wanting to understand feedback on key experiences and workflows to inform roadmaps. And it’s for stakeholders who want to quickly see emerging trends and make informed decisions.

How does Channels work?

Let’s take a look under the hood.

Channels is powered by generative AI. We build upon established technology from Anthropic, ensuring reliability and privacy, and utilize specific models to manage large data sets efficiently without training on your sensitive data.

Once data is uploaded or streamed into Dovetail, we have implemented purpose-built models that perform specific jobs to uncover patterns in large, complex data sets. Our AI performs the following tasks to help you discover trends quickly:

1. Preliminary theme identification—our models first find themes within streamed data, creating a foundation for further analysis.

2. Data classification—as new data comes in, our AI continuously classifies information into pre-defined themes, maintaining a dynamic understanding of feedback.

3. Subsequent theme detection—our AI regularly identifies new, relevant themes in recent data, updating its understanding as trends evolve.

4. Summarization—for each theme, Dovetail generates concise summaries using Anthropic's LLM, transforming raw data into actionable insights. And it’s not just one and done, as new data is classified into themes, our AI will continue to update this summary to reflect any changes.

Dovetail’s guiding principles for AI integration

When it comes to deciding how we integrate AI into Channels, we are guided by a few core principles.

Purpose-built beats a one-size-fits-all approach

During a time when many companies were rushing to slap AI onto their products faster than you can say generative AI, we knew that we wanted to take a more measured approach to integration. We focused on deeply understanding the primary challenges our customers were looking to solve, and built our solution to address these using a mix of the best technologies available today.

This is why when you create a channel in Dovetail, we ask you what kind of data is being analyzed. This way, we can apply the model that is built to specifically analyze, cluster, and summarize that type. Channels is designed to help you understand the voice of your customers across feedback types so when we ask you this question, it also helps inform the LLM to generate more relevant themes. It’s never one-size-fits-all.

Flexibility wins over rigidity

The world of AI development is fast-moving. That’s why we maintain an agile and flexible approach to our AI features. How do we do that? We’re committed to continuous theme improvement and our engineering team is continuously responsive to updates that can impact the customer experience. We know that AI can sometimes get it wrong, and needs constant care and feeding based on those real-world experiences. Our flexible design allows us to adapt while also sticking with our goal of providing consistent value.

Keeping humans in the loop is a must

Recognizing that AI doesn't always get it right, we’ve built all of our AI features with our human-in-the-loop methodology. This methodology refers to a system design approach where human input is invited into an automated process to make adjustments where necessary. For Channels, this means that users have control over themes—they can create more granularity by splitting themes, they can merge overlapping themes or create a new theme to inform the LLM.

Additionally, when you make changes to your themes, what you see is exactly what the AI sees, and this natural interaction ensures that you are always in control.

How can you work with our AI to get the most out of it?

Get your data structure right

Structure is key when it comes to your data. The right structure enables Channels’ AI to do its thing—classify your data reliably and continually—in a seamless manner so that you can get the insights you need, fast.

When setting up a new channel integration, be sure to include the relevant metadata fields which will be useful for filtering later on—such as metadata in your Intercom integration which can be ported into Channels. This allows you to easily filter by those fields and gain insights into your data.

Finesse your themes

Engage with the AI's classifications to tailor insights to your specific needs. After Channels classifies your data into themes, you can split, merge, rename, and create new themes.

When it comes to naming your themes, we recommend following these tips as a guideline:

• Differentiate your theme titles—avoid having multiple themes covering the same topics.

• Make theme titles context-free and evergreen—for example, use “bugs”, as opposed to “last quarter problems”.

• Keep theme titles simple—keep them focused on a single focus area rather than multiple.

What’s next for Channels?

We are excited to see Channels already providing value to our users, with over 1,000 channels created across our customer base and over 8.4 million data points streamed into and analyzed by our AI. But getting that data into Dovetail and transforming it into rich insights is one thing. In our conversations with early adopters, we also know that it’s critical we make that data workable to meet each specific use case, as well as accessible across the organization to help align teams and drive better decision-making.

The first piece of this is ensuring we give our users total control over the data in their channels. We are currently working to deliver sub-themes, which would provide an additional level of organization to how your data is organized. This allows you to keep all of your data in a primary theme, while also abstracting them further into smaller, more focused groupings, without losing that top-level focus.

Additionally, we want to ensure that the data in Channels has maximum impact by making it accessible to your key stakeholders. To help with this we recently made updates to include Channels data in magic search, as well as within Ask Dovetail queries and digests. Today we are building features that allow you to also add Channels data to an Insight report in your Projects. And for when you just want to quickly share Channels data, we are also working on more dynamic views and enhanced data visualizations that you can quickly share with your stakeholders. By making Channels data more dynamic and visually engaging, we aim to empower you to make smarter, faster decisions, based on deep customer insights.

Channels is already providing significant value to users. We’ve already been working with organizations like Okta, Shopify, and Walmart who are continuously streaming their feedback data into Dovetail to help understand what to build next. Want to try Channels for your organization? Try it today.