BubbleUp has always been the fastest way to figure out what a group of outliers have in common. Draw a box around a band of slow traces, a cluster of errors, or any set of events you're interested in, and BubbleUp compares that selection to the baseline across every dimension you've sent us. It's how Honeycomb users find the "unknown unknowns" that dashboards can’t show you.

We recently shipped a big upgrade. BubbleUp now comes with AI-powered insights that summarize your selections’ most significant correlations, how they differ from the baseline, and how they are likely to be relevant. By surfacing these correlations immediately, AI BubbleUp can save hours of investigation time and make debugging insights available to team members who aren’t experts in your system’s telemetry.

The challenge with high-fidelity telemetry

Honeycomb captures wide events with as many dimensions as you want to send, at any level of cardinality. That's what makes BubbleUp powerful, but it can also make the results dense and sometimes challenging for non-experts to interpret. A typical environment has well over a hundred dimensions on its core spans. Some have many more.

To help you make sense of the results, Honeycomb has always sorted those dimensions by how different the outlier group is from the baseline. That works well for some data types, but statistical difference and causal relevance aren't always the same thing, and context always matters. Also, dimensions with very high cardinality, like user ID, may not have a meaningful ‘baseline’ to establish difference from, so they may not be ranked highly by the sorting algorithm. A strong correlation in a low-cardinality dimension, like a customer support tier, might rank in the first page of results, while a more diagnostically relevant dimension, like a single misbehaving user ID, sits several pages down. If you're not deep in the specific telemetry your services emit, the most significant dimensions can be hard to spot.

What AI insights add

AI insights read the BubbleUp results and surface the dimensions most likely to be relevant to the problem, even when they aren't the top results by raw statistical difference. A few things to note about how this works:

• AI is always an add-on, never a replacement. Raw BubbleUp results are still right there. You can scroll, sort, and verify everything the AI flagged and find things it didn't.
• The AI reasons over your real telemetry. Insights come from the same high-cardinality event data you'd inspect by hand, so the recommendations are relevant to what actually happened in your system.
• It works on any BubbleUp, no need for setup or configuration. Run BubbleUp the way you always have and the insights appear with the results.

AI helps most when it's given the right raw material, and for production debugging, that means high-fidelity event telemetry. BubbleUp was already one of the highest-leverage things you could do with a wide-event store. Adding AI on top means that even people who aren't experts in every dimension your services emit can still get to the relevant signal quickly.

Get started

AI insights in BubbleUp are now available to all Honeycomb customers who have enabled Honeycomb Intelligence. Run any BubbleUp in your environment and you'll see them in your results.

Check out the documentation for BubbleUp.