How Mixpanel uses AI: Marketing edition

Here at Mixpanel, we run into a lot of the same problems that our customers encounter. We want to move quickly and make data-driven decisions that work well for our business. But like most companies, our resources are finite, and we can’t spend all of the time and budget we’d like on every single project. 

As part of the Mixpanel product marketing team, I use data to fine-tune our messaging, launch successful brand campaigns, and much more. But I also have to be realistic about the resources I have available to gather and analyze that data. Otherwise, my projects would end up late and over-budget.

The reality is that marketing projects like defining our messaging could require large amounts of representative data—data strong enough to take to market. Achieving that level of confidence is expensive, in both time and budget. While research should ideally inform every campaign, that's not always realistic.

Like many of our customers, we’ve been playing around with AI to speed up processes and overcome blockers. We’ve found that using AI as a thought partner has helped us gather directional, actionable data for projects we otherwise wouldn’t be able to dedicate time and resources to. 

It’s still early, but we’re excited to share what we’ve learned to date with our community and hopefully learn from what others are doing. (Feel free to share how you’re using AI in the Mixpanel Community here!)

Here are two specific ways the product marketing team has experimented with using AI to help us scale. 

Our product marketing team needed feedback on new messaging, but couldn't access enough input from our ideal customer profile (ICP) to validate the changes. Rather than delay the project, we used AI to simulate how our ICP would respond, essentially using it as a stand-in for the user testing we couldn't do at that moment.

Our goal was to use AI as something similar to a user testing interview, but on a larger scale. Fortunately, we already had a ton of internal documentation on our ICP, thanks to months of in-depth research. We uploaded that information to the AI so that it could replicate the ICP’s point of view and give us feedback on what resonates and what needs improvement. 

AI acted as an initial filter and thought partner to help us refine our messaging before it went out to be tested with real users (and eventually to the market). 

Looking back, AI gave us good directional signals, so we could build out more complete messaging that's ready to be tested against real users at the next stage.

In preparation for a new brand campaign we were launching, our product marketing team wanted to expand our understanding of how the market perceives Mixpanel; specifically, how we show up across key criteria. Rather than launching a full brand lift study, we partnered with AI to get a baseline we could measure against over time.

AI’s deep research capabilities were crucial here to evaluate Mixpanel against competitors and peers across key criteria to get our baseline score. We wanted it to scour publicly available content to return an assessment of our position. 

As a first step, our team worked without AI to set four evaluation criteria we wanted to track in the future. Then, we leveraged AI to define and refine these categories: 

Next, we asked AI to return results for these four criteria, comparing Mixpanel and industry peers. This provided us with a quick, consistent, and “directional” baseline score to measure future efforts against. In the process, we were able to set tangible goals for improvement (e.g., “aim to move the needle 10% by April”), which otherwise would have been impossible without a major resource investment.

As a reminder, we’re treating these ideas and processes as experiments. So far, we love having the chance to explore and learn what AI can do for our team and our workflows. 

This approach is meant to be directional. It gives us a good starting point, but it’s definitely not the end of the exploration. 

The goal isn’t to replace user research or market studies, either. Rather, we’re simply introducing an additional early layer to our methods for when resources or a small sample size present us with constraints we otherwise wouldn’t be able to address. 

Based on our experiments, we’d recommend using the kinds of AI analysis hacks we’ve described above, then testing your work once it’s possible to get more data from your users or from market research. 

The value of this approach comes from consistency. We didn’t rely on the absolute score as a definitive reality. Rather, we repeated the same process and re-ran our AI-based tests later to measure improvement over time. 

Furthermore, we’ve only used this AI scale hacking approach to aggregate or simulate qualitative user feedback and sentiment. We have yet to try this approach with actual user event data from product usage, which would probably look a bit different. 

Like many of our customers, we’re experiencing uncharted territory with AI and its capabilities. Here are our takeaways so far: 

There will be mistakes, and false starts, and ideas that won’t pan out. But there are also a lot of opportunities ahead, and we’re genuinely excited to explore them together. 

We’d love to hear how our readers are using AI. Tell us more about how you’re incorporating AI into your day-to-day to overcome resource constraints and improve your processes in the Mixpanel Community here.