How Fyxer built its highest-CSAT product with Recall.ai

Executive Summary

Challenge

Fyxer is an AI assistant that helps users manage their inboxes and meetings. Meeting data is particularly valuable because conversations contain far more context than most emails, giving Fyxer the information it needs to generate meeting summaries, draft follow-ups, prepare users for future meetings, and improve the emails its AI writes.

Fyxer already offered bot-based meeting recording through Recall.ai, but saw growing demand for a botless experience. After surveying existing users and testing interest during onboarding, the team found that roughly 50% of users were interested in botless recording.

Building the underlying desktop recording infrastructure internally would have required Fyxer to solve difficult problems such as detecting meetings across different applications, capturing audio reliably, and accounting for edge cases across meeting platforms.

Solution

Fyxer used Recall.ai’s Desktop Recording SDK to add botless meeting recording to its desktop application.

The SDK handles the underlying recording infrastructure, including automatic meeting detection, audio capture, live transcription, and speaker identification. Fyxer could then focus its engineering resources on the user experience and on turning meeting data into useful outputs such as summaries, action items, and email drafts.

The team began dogfooding an initial version within two to three weeks. Fyxer estimates that building comparable infrastructure itself could have taken years.

Results

By building its desktop recording experience on Recall.ai, Fyxer:

• Shipped its desktop application in months instead of the years it estimates an internal build could have required
• Enabled a two-person note-taking team to build a product that can compete with much larger meeting-assistant products
• Avoided needing a dedicated engineering team for recording infrastructure
• Began dogfooding botless recording within two weeks
• Expanded the range of meetings it could capture to include ad hoc and Slack meetings that bots would otherwise miss
• Kept engineering focused on the Fyxer user experience instead of meeting-detection and recording infrastructure
• Used richer meeting context to power summaries, follow-ups, meeting preparation, action items, and better email drafts

Meetings give Fyxer context that email alone cannot provide

Fyxer started as an AI assistant designed to reduce inbox overwhelm. It categorizes emails, surfaces important messages, and helps users manage communication.

But email only reveals part of a user’s working life.

Meetings contain much more context.

“An email often is one to two sentences with very straight, direct text,” said Serafin, Product Engineer at Fyxer. “Whereas a meeting could be an hour of people talking about the why and the how.”

That context makes meeting data useful far beyond note-taking.

After a meeting, Fyxer can generate a summary and create a follow-up email draft within minutes. Before the next conversation, it can use the previous meeting to create a pre-read with relevant actions and insights. Users can also interact with their meeting data to recover decisions, insights, and action items that might otherwise be forgotten.

Meeting context also improves Fyxer’s core email product.

Many AI email assistants primarily understand users through their inboxes. Because Fyxer also has access to conversations, it can build a broader picture of what is happening in the user’s work and use that context to draft more relevant emails.

“Because we have the user's meetings recorded, it gives us a much broader context of the user and allows us to draft better emails with more depth,” Seraphine said.

Unlike an email, however, a meeting cannot simply be fetched later.

“If we didn't have the recording of the meeting, we wouldn't know what's happened in a large part of the user's day,” Serafin said.

For Fyxer, reliably capturing that data is therefore a prerequisite for everything it wants to build on top of it.

Half of surveyed users showed interest in botless recording

Fyxer initially used Recall.ai to power bot-based recording.

But the team saw the meeting-assistant market moving toward botless experiences and wanted to know whether its own users felt the same way.

Rather than immediately investing in a native application, Fyxer tested demand.

The team surveyed existing note-taker users and added a fake-door test during onboarding asking whether users would download a desktop application for botless recording.

Roughly 50% of users expressed interest.

Fyxer now had evidence that users wanted the capability.

The next decision was whether it made sense to build the underlying recording technology itself.

Fyxer tried building part of the desktop infrastructure, and quickly saw the complexity

Fyxer did not want to spend months building infrastructure before learning whether customers would use the product.

The team initially explored parts of the recording problem itself and quickly encountered the complexity behind seemingly simple functionality.

Meeting detection was one example.

Fyxer needed to reliably understand when a user had joined a meeting across different browsers and applications such as Zoom. Building that logic itself proved harder than expected.

“We tried to build it ourselves but then realized it is quite hard to detect all the different types of meetings that are happening in the user's browser [and] application,” Serafin said.

Other edge cases appeared as development continued.

For example, Fyxer wanted to automatically notify participants that a botless meeting was being recorded. Fyxer engineers attempted to build a system that would send a message into the meeting chat, but making it reliable introduced another set of problems.

Fyxer raised the issue with Recall.ai. Within days, the Recall.ai team had a working solution.

The experience demonstrated a broader build-versus-buy tradeoff.

Recording was only one layer of Fyxer’s product, but owning it internally would mean becoming responsible for the long tail of meeting platforms, operating environments, detection logic, recording behavior, and edge cases that came with it.

“Using the Desktop SDK was a no-brainer because it allowed us to focus on the more important things for our product, like the user experience,” Serafin said.

A two-person team could build what otherwise would have required a dedicated recording team

The economic value of that decision is especially significant for Fyxer because the team working on its note-taking products is small.

“As a startup, we're just a team of two working on the whole note-taker, whether it's web application or desktop app,” said Lucien George, Senior Product Engineer at Fyxer.

Without Recall.ai, George believes the company would have needed a dedicated team just to build and maintain the recording layer.

“I think it definitely cut down on engineering costs because otherwise you would have needed a fully fledged team just working on the recording infrastructure,” he said.

Instead, Recall.ai allowed two engineers to work across both Fyxer’s web and desktop meeting experiences while relying on Recall.ai for the specialized recording infrastructure underneath them.

That changed the startup economics of launching the product.

Fyxer could direct scarce engineering resources toward the areas customers actually experience rather than staffing an internal infrastructure team before proving demand.

“Having Recall.ai's ecosystem integrated in Fyxer really allowed this team of two to build a note-taker that can compete with other note-takers out there,” George said.

Fyxer went from idea to internal usage in weeks

The Desktop Recording SDK also shortened the feedback loop.

After Fyxer validated user interest in botless recording, the team wanted to get something working quickly enough to test internally.

Because Recall.ai already provided the recording layer, Fyxer began dogfooding its desktop application within approximately two weeks.

George said the Desktop SDK looked straightforward enough from the documentation and integrating was simple.

Over the broader development cycle, Serafin estimated that Recall.ai allowed Fyxer to ship its desktop product in months rather than years.

“Recall.ai helped us launch faster,” she said. “We've been shipping the desktop app in months instead of years, which probably would have taken us if we didn't use Recall.ai.”

For a startup, those years are not simply an engineering-cost calculation.

They represent years before customer feedback, years before the feature can contribute to the product, and years during which engineers cannot work on other priorities.

Recall.ai let Fyxer test the product opportunity without making that investment first.

Recall.ai handles recording so Fyxer can design the experience around the user

The product Fyxer wanted to create was not simply another recording application.

Its goal was to make meeting capture disappear into the user’s existing workflow.

With Recall.ai’s automatic meeting detection, Fyxer’s desktop application can run in the background, recognize when a meeting starts, and begin the recording experience without requiring users to manage the underlying recording infrastructure themselves.

Afterward, Fyxer can surface the transcript, summary, and action items in its web application.

The Desktop SDK also supports live transcription with actual speaker names rather than generic labels such as “Speaker 1” and “Speaker 2.”

That division of responsibilities gives each company a clear area of focus:

Recall.ai handles the complexity of capturing the meeting. Fyxer focuses on what users can do with the information afterward.

The result is a desktop experience designed around Fyxer rather than around the mechanics of recording.

Botless recording expanded the meetings Fyxer could capture

The Desktop Recording SDK also gave Fyxer access to conversations its meeting bots did not always reach.

That includes ad hoc conversations and meetings in environments such as Slack.

More recorded conversations mean more meetings Fyxer can summarize, more follow-ups it can draft, more information it can use for future meeting preparation, and a more complete understanding of what happened during the user’s day.

Across Recall.ai more broadly, Fyxer says it has recorded millions of meetings.

Its Recall.ai-powered Note-Taker has also become Fyxer’s highest-CSAT product, according to Serafin. Its popularity was one of the factors that encouraged the company to expand from meeting bots into botless recording.

For Fyxer, buying recording infrastructure meant investing more in the product

Fyxer knew meeting data was strategically important.

That did not mean recording infrastructure needed to become a Fyxer core competency.

The company had already made the same decision with meeting bots: building and maintaining provider-specific recording infrastructure was a large amount of work for a problem that was not central to what differentiated Fyxer.

Desktop recording presented the same choice.

Fyxer could devote a team to solving meeting detection, recording, transcription workflows, provider differences, and the edge cases that would emerge over time.

Or it could use Recall.ai and put those engineers into the experience built on top of the meeting data.

For a startup with limited time and engineering capacity, George sees that tradeoff clearly.

“A startup doesn't have the luxury of time,” he said. “Building with Recall.ai really allowed us to ship something very quickly.”

That is the ROI Recall.ai provides Fyxer: not simply fewer lines of infrastructure code, but the ability for a small team to launch a new recording experience, validate it with users, capture more of the customer’s day, and keep investing engineering resources in the product only Fyxer can build.

‍