How we built AI in Dovetail

In 2024, we announced a host of new AI-powered capabilities in Dovetail, also known as our Magic features. It was a culmination of extensive work by our engineering team, and came to life across several existing Dovetail features, as well as brand-new products such as Channels and Ask Dovetail.

It’s been over two years since the initial release of ChatGPT, which unleashed the new age of generative AI. It can be difficult to remember those early days when AI and ML transformed overnight from aspirational buzzwords to the most exciting (and controversial) technology in recent memory. That is partly because the hype cycle for AI seems to be heading quickly for the dreaded “Trough of Disillusionment” (in fact, Gartner indicated that by November 2024 GenAI has already passed the “Peak of Inflated Expectations”). We believe one reason for this is that companies, eager to seize on that initial excitement, rushed to “integrate AI into their core product experiences.” But when digging into these new products, they were simply a ChatGPT-style wrapper bolted onto a product, making experiences feel clunky and disjointed.

At Dovetail, we wanted to avoid that pitfall. That is why we took the time to deeply understand the advantages and limitations of AI, understand how our users felt about this technology, and how both should inform our long-term product strategy. In this blog post, we want to share our learnings from building our magic features in hopes it will help other teams in their AI journey.

AI is best suited for speeding up manual tasks

While we all simultaneously dream of and fear for a future where AI does all of our work for us, the fact is that the technology is still quite limited. By taking the time to dig into AI and ML and pushing them to their limits, while also listening to our customers to understand where they felt blocked in their Dovetail experience, we determined the boundaries of various use cases and how AI can help accelerate key workflows.

For example, we knew our users were big fans of our transcription capabilities as a way to save time taking and uploading their notes. However, we heard lots of feedback that users (especially those who aren’t professional researchers) were frustrated with the manual nature of highlighting key moments and pulling out the takeaways from each call. This was a perfect opportunity to start experimenting with AI.

Today, our users can leverage AI automation to create summaries of calls, as well as searches within their Dovetail workspace. Additionally, they can work hand in hand with AI to more quickly identify highlights within call transcripts, as well as uploaded documents. It was a great reminder that any solution, even one built on a nascent technology like AI, should be rooted in clear customer pain points to ensure lasting success.

‘Human in the loop’ is a necessity

Even with the acknowledgment of AI’s power to accelerate manual tasks, our research showed that there is still some hesitation in the market. For example, it is well documented that LLMs tend to “hallucinate” and make up answers in the absence of data, and as a result very few people are ready to completely hand over tasks to AI. We wanted to avoid this pitfall by ensuring our models were well-trained, but knew that if we spent too much time waiting for the “perfect AI” we would lose out on seizing a clear market moment.

So we made the conscious decision to include human quality gates in our AI features, allowing our users to accept or reject AI-powered highlights and give feedback on the quality of AI summaries. This provides the user an extra level of transparency and agency to incorporate these features into their workflows as they see fit, and has a knock-on benefit of giving us the real-time feedback needed to improve our own models.

Secure by design

Keeping our customers’ data safe and secure is always at the core of any feature we build and it is no different with our magic features. There is no shortage of think pieces around the legal and ethical concerns of data privacy and AI, and we knew that to maintain trust with customers we had to drive clarity on how we were implementing AI in ways that protect their most sensitive information.

That is why we designed these features with a few overarching principles in mind:

• We don’t use any customer data to train our models. Instead, we run our own pre-trained models in all of our magic features.

• We select a model that is suited to the task at hand. From summarizing to classifying, we never use a one-size-fits-all approach when it comes to AI.

• All of our models run in our own infrastructure, so you can rest assured your data won’t end up elsewhere.

AI is not a feature, but our foundation

While it’s tempting to jump on the bandwagon and slap AI onto our product as quickly as possible, we knew instinctively that integrating this technology would fundamentally change how we approach our roadmap. So we wanted to make sure that we were setting ourselves up for success for the long term.

To this end, we are laser-focused on building AI into our product which is deeply rooted in the problems our customers are facing every day. These include:

• Uncovering insights faster,

• Driving alignment between teams through a shared understanding of customers, and ultimately

• Building better products

Dovetail prides itself in putting the customer first (it’s one of our most beloved values, after all). By reminding ourselves of this principle, we believe we are well-equipped to continue building delightful experiences with AI that truly make our customers’ lives better and can avoid the pitfalls of incongruous, bolted-on technologies that aren’t truly rooted in the challenges they face. In upcoming blog posts, we’ll be doing deep dives into our AI products and features to help our customers understand them and get the most out of them including Channels, Ask and magic analysis in projects. So stay tuned.

These magic features are only the beginning of our AI journey, and we are excited to continue to build on this foundation in the months to come. If you’d like to try them out for yourself, sign up for a free account today.