Analytics

Product and web analytics has spent several years reorganising itself around two pressures: privacy regulation, and the realisation that most teams were collecting far more data than they ever looked at. The launches in this subcategory reflect both, and the interesting ones are notably smaller and more opinionated than the previous generation.

The privacy-first group is the clearest trend. Tools that avoid cookies, do not track individuals across sites, and aggregate at collection rather than storage can frequently run without a consent banner, which has a measurable side benefit: you see all your traffic rather than the sixty or seventy percent who accepted. For most sites the loss in analytical depth is smaller than people fear, because the questions that actually get asked — where did traffic come from, which pages get read, what converts — do not require individual-level tracking.

For product analytics specifically, the meaningful decision is event granularity, and the common failure is tracking everything. A schema with four hundred events is one nobody can query with confidence, because no one remembers which of the three similar events is the one that fires. Twenty well-named events covering the actual funnel produce better decisions than four hundred covering everything.

Self-hosted options are unusually strong in this category and worth considering, because analytics data is exactly the kind you may prefer not to hand to a third party, and the workload is light enough to run cheaply.

The thing no tool provides is the question. Analytics answers questions and generates none, and a dashboard opened daily with no specific question is a habit rather than a practice. The teams that get value from this category tend to work the other way: write down what you believe, decide what number would change your mind, then go look.

Attribution deserves a warning. Multi-touch attribution models are assumptions rendered as charts, and different models will confidently give you different answers from identical data. They are useful for comparing periods under a consistent model and misleading when read as truth about which channel deserves credit.