Back to blog
Product

Introducing ML insights: forecasts, spikes & segments

Zayd
Jun 28, 2026·3 min read

Most link tools stop at counting clicks. That's useful, but it leaves the hard questions unanswered: what is actually happening beneath the total, is the change I'm seeing real, and what happens next? A number on a card can't tell you whether Tuesday's spike is a trend or a fluke, or whether the audience clicking today is the same one that will convert tomorrow.

Answering those questions has traditionally meant exporting your data, wiring up a warehouse, and asking an analyst to build models. That's out of reach for most teams — and overkill for a link. So we built the analysis into the product. Today we're shipping ML insights: a set of models that run over your click data automatically and surface the patterns a careful analyst would look for, without any of the setup.

Trends you can actually trust

The word "trend" gets thrown around loosely. A dashboard that shouts "+24% this week" feels great until you realise the week before was down 20% and you're just watching noise bounce around. To fix that, the trend in ML insights now ships with a significance test.

We fit a linear regression over the last 90 days of daily clicks and report the slope alongside a p-value. When the p-value is small, the movement is statistically solid and worth acting on. When it's large, we tell you plainly that the change is within normal variation. The practical effect is that you stop chasing random spikes and start responding to real signal — which is where your time is best spent.

Forecasts with an honest range

A single forecast number is a trap: it looks precise and is almost always wrong. Our forecast model instead projects the next seven days with a confidence band, so you get a realistic range rather than one optimistic point. If your campaign is trending up, you'll see how fast and how uncertain that growth is — which is exactly what you need to plan capacity, budget, or a follow-up push.

Anomaly checks every hour

Running alongside the forecast, an exponentially weighted anomaly detector checks each link's daily clicks every hour. It keeps a moving expectation of normal behaviour and flags days that land far outside it, and the same score powers spike alerts. When a link suddenly takes off because someone influential shared it, you hear about it within hours, not in next month's report.

Visitor segments, discovered automatically

Averages blend everyone together and hide the interesting structure. Using k-means over three behavioural signals (how often a visitor clicks, how many different hours of the day they show up, and how recently they came back), TRIMS sorts your audience into groups such as power users who return often, casual visitors, and traffic that looks automated.

Seeing those groups separately changes decisions. You can weight your reporting toward real humans, notice when a campaign is inflating on low-quality traffic, and recognise the small segment that drives most of your conversions.

  • Significant-trend detection with p-values so you know what's real
  • Seven-day click forecasts with confidence bands
  • Spike and decline alerts, checked every hour
  • Behavioural visitor segments via k-means clustering
  • Peak hour and busiest day of the week
  • Dark-social estimation to explain "direct" traffic
The goal was never a smarter dashboard. It was fewer meetings that end in "let's keep an eye on it."

All of this is available on the Pro plan and runs without any configuration — no warehouse, no notebooks, no data team. Open a link's analytics and the insights are there, refreshed every hour for any link with more than ten human clicks in the last 90 days. If you've ever stared at a click chart wishing it would just tell you what to do, this is for you.

Zayd
Founder & Engineer, TRIMS

Building TRIMS — a fast, honest link platform with analytics that actually help you decide.

Next up
Why we made the most generous free tier in the market