Be the first to know what's breaking
Trends watches what customers are writing about and tells you when the mix changes — a login outage, a failed delivery run, a billing job that didn't fire. It recomputes every five minutes and shows the counts behind every claim.
500 conversations, no credit card.
What's in Trends
- Moving away from usualThe topics pulling away from their usual share of the window, ranked.
- Total volumeWhether the number of conversations arriving is unusual for this day and hour.
- TopicsWhat your customers write about, in your own words. You can rename, merge, split and retire them.
- TagsWhere a window has no topics yet, Trends detects on the tags you already have.
- DetailsThe counts behind a signal, its channels, and the conversations themselves.
- Time windowLast 2 hours, 6 hours, 24 hours or 7 days.
- SensitivityLow, Normal or High — how sure Trends has to be before it lists a move.
- What is not settled yetHow much of the window it could place, and what it left off the page.
- FallingTopics moving the other way, and channels that were closed.
- Trend Spike DetectedAn automation trigger, so a rule can post a signal to Slack or open an incident.
Every number, and where it came from
Fourteen customers in the last two hours, against a usual 1.4 — ten times the normal rate, and half of everything in the window. It started about two hours ago. That sentence is assembled from the counts the detector used; no model writes it.
Details opens the rest: the last seven days against what is usual for those days, when it was first detected and whether it is still going, the biggest fifteen minutes, the split by channel, language and source, how many were handed to a human, and the conversations themselves.
- Signals are found in Topics — built from your own conversations, in your own words — and you can rename, merge, split and retire them
- Where a window has no topics yet, Trends falls back to the tags you already have and says so
- The window picker holds the last 2 hours, 6 hours, 24 hours and 7 days
- The list is ranked by human work above usual, and the heading says how many of the total it is showing
14 customers in the last 2 hours, against a usual 1.4 — 10.1× the normal rate, 48% of everything in this window. It started about 2 hours ago.
It knows the difference between a busy morning and a bad one
Volume and composition are separate questions, so Trends asks them separately. Is today's traffic unusual for a Tuesday at nine? That's one event, reported once. Then, given the volume you actually received — not a prediction of it — is any topic taking a share it shouldn't? A morning that's evenly busier moves no topic's share, so it raises one flag instead of 150.
Yesterday's incident doesn't become tomorrow's normal
Leave a week of login failures in the history and the detector learns that login failures are normal — so the next outage passes unremarked. Windows already flagged as incidents are lifted out before what's usual is worked out.
With a limit, because the correction feeds itself: every removal lowers the bar, which flags another window. Removal is capped at about a fifth of the sample — over sixty-two days of real traffic, the cap gave up nothing in recall.
- What is usual is read from the 28 days before the window
- A guard band around each flagged window keeps its edges out too
- The charts draw the same window a week earlier beside today's
It tells you what it cannot see
Under thirty conversations in a window, no test can separate a change in the mix from noise — so Trends says exactly that and reports total volume only.
Same for the rest of its limits: how much of the window carried a topic or a tag, that it's still reading your history, how many signals didn't make the page. Nothing is quietly truncated.
- Sensitivity is yours to set — Low, Normal or High — and the page says what each level is hiding
- Falling lists the topics moving the other way, and names a channel that was closed
- When nothing clears the bar it says so rather than filling the page
Trends reads the traffic the rest of the platform is already handling. Analytics tells you how you're doing; Trends tells you what just changed.
Conversations
Every conversation across every channel is the traffic Trends reads. The ones behind a signal open from the card.
Analytics
Analytics shows how the operation is doing over time. Trends shows what changed in the last two hours.
Incidents
When a signal turns out to be real, publish it and keep customers posted through to resolution.
Monitor
Watch the spike land in the live queue, and who is picking it up.
What changed, and how sure we are.
See Atender in action
Try the full platform for free. Or book a demo and we will walk you through it.
500 conversations, no credit card.