Closing the velocity gap: Experimentation in the AI era

Building fast is the baseline. The teams pulling ahead are the ones who also learn fast, and learn early enough to change what they ship next. 

That was the central argument in Mixpanel's recent webinar, hosted by Emma Janiszewski (Solutions Engineer) and Russell Loube (Senior Product Manager). They walked through what they call the "velocity gap": the divide between shipping quickly and learning effectively through experimentation. The session included live product demos, a look at what's on the roadmap, and practical strategies teams can apply now.

Emma opened with the core problem. Even teams that ship quickly often operate in reactive cycles.

"You launch a feature, wait weeks for data to trickle in, and realize too late that it didn't move the needle."

The fix isn't slowing down. It's building systems where experimentation is continuous, integrated, and connected directly to the product cycle. That shifts teams away from blind bets and toward hypothesis-driven development.

The MCP (Model Context Protocol) server connects natural language AI tools like Claude and GPT to your Mixpanel data. Emma described it as a universal adapter. It started with analytics and now supports experimentation and feature flags, so PMs, engineers, and analysts can:

This removes the need to stitch together reports manually or jump between tools. Emma showed how users can tap into pre-built "skills" inside MCP, including a brainstorm skill, to surface behavioral flow issues, pull session replays for context, and rank opportunities with real metrics. That covers the connection layer: external AI tools talking to your Mixpanel data. Mixpanel Agent works from the other direction, bringing AI directly inside the product rather than routing through an external client.

Russell introduced Mixpanel Agent, which brings AI assistance directly into Mixpanel for more granular insight synthesis. Two things it does well: 

"We want to give you the speed to synthesize, but also ensure you remain close to your customers and data."

Mixpanel's feature flags connect directly to behavioral cohorts from Mixpanel analytics, something other platforms currently don't offer. Teams can use the same cohorts for experiments and feature rollouts with only minimal code required.

Russell walked through how teams can configure onboarding experiences or target first-time users with runtime event capabilities, a feature built specifically to streamline personalization.

"We provide unparalleled depth of targeting with behavioral cohorts. What sets Mixpanel's feature flags apart is their ability to connect directly with your analytics workflows, ensuring seamless coordination across experiments and rollouts."

A consistent thread throughout the webinar was Mixpanel's focus on combining accessibility with depth. Emma described how the tooling lowers the barrier to experimentation without sacrificing statistical rigor.

Russell pointed to recent investments in advanced statistical features, including Bonferroni corrections, winsorization of outliers, and stratified sampling, to make data interpretation faster and more reliable.

He also shared Mixpanel's vision of letting teams simulate experiments with synthesized user data, so they can get earlier signals and validate hypotheses before deployment.

"Mixpanel balances simplicity and accessibility for new users with the power and depth needed by experienced data scientists."

Emma and Russell both pushed the idea of embedding experimentation into team culture, not just tooling. Russell framed it around participation: "The more you can promote participation, the more momentum you'll drive. At Mixpanel, we have company-wide Slack channels where anyone can contribute growth ideas, and it's hugely exciting for the team."

Failed experiments aren't wasted ones. As Russell says: 

"Most experiments fail, but all of them present lessons. Sharing inconclusive results is equally important to showcasing wins."

Both speakers returned to the same point throughout: the right tooling lowers the cost of starting, especially for teams with limited experimentation infrastructure. Experimentation often stalls not because teams lack ideas, but because the setup, analysis, and coordination overhead is too high to sustain. When the infrastructure lives inside the same tool you're already using for analytics, the barrier to running your first test, or your fiftieth, drops considerably.

Here's where things stand for teams ready to start:

Russell closed by encouraging attendees to put the tools to work: "Please give the tools we've discussed today a try. They're designed to empower faster decision-making, reduce workflow friction, and provide clarity to your team."

The teams closing the velocity gap aren't just shipping more. They're learning faster, acting on better signals, and building that feedback loop directly into how they work. Emma's line from the session captures it well: "Speed is table stakes, but insights are the edge." 

To watch the full session, go here or to see how Mixpanel's experimentation tools can close the loop for your team or get a demo.