Why we killed our real-time activity feedWhywekilledourreal-timeactivityfeed

We shipped a real-time activity feed in our dashboard 18 months ago. It was a live scrolling ticker showing every scan, count, adjustment, voice command, and order event across all the warehouses a customer ran. We thought it was going to be a flagship feature for operations managers. Last week we removed it.

This is the story of why, because we think there's a pattern that applies more broadly than dashboards.

What we were trying to do

The pitch was straightforward. A warehouse operations lead has a hard time knowing what's happening across the floor at any given moment. They can walk the floor, but only one floor at a time. They can call a supervisor, but only one supervisor at a time. A real-time view of what's actually happening, scrolling past in front of them, should help.

We built it in three weeks. The animations were polished, the typography was good, the batching was smart enough that the feed didn't feel like noise. We were proud of it. Customer demo feedback was positive. We shipped it as the default landing page for managers.

What actually happened

For about six months we patted ourselves on the back. Customers told us they loved the feed. Sales used it heavily in demos. We saw screenshots of it in customer marketing materials, sometimes.

Then we started looking at the data.

Managers spent a median of 14 minutes per session on the dashboard with the feed open. That, by itself, is fine. What wasn't fine was that during those 14 minutes they took an action (clicked, drilled in, navigated to a related view) about 0.7 times. Median. Most sessions ended with the manager closing the tab.

For comparison, on a dashboard view we built around weekly summaries, managers averaged 4 minutes per session and 3.2 actions. The feed was three times stickier and four times less productive.

We thought maybe the feed was helping in ways the click data couldn't show. The manager watches the feed in the background, notices something off, then walks the floor. The action would have happened in the physical world, not in the app.

We interviewed thirty managers. Not one of them described that workflow. What they described was something else: the feed was a comfort object. It made them feel informed. It did not make them more informed in any way they could articulate. Several said they kept the tab open during meetings because the motion was calming.

A few specifically said they got anxious when the feed scrolled fast and they could not keep up. One manager described it as "a stock ticker for problems I might already be missing." That was bad.

What broke our resolve

We sat on this data for about six months. We tried to redesign the feed (slower scroll, prioritized events, filters). It got marginally less anxiety-inducing and did not get more useful. We added a "summary mode" that compressed the feed into a once-an-hour digest. About 8% of managers turned it on.

What broke us was a separate signal. When we shipped the anomaly detection layer in April, the alerts it produced were better signals than anything in the feed had been. The feed had been showing every event with no judgment about what mattered. The anomaly layer was showing the few events that actually did.

With anomaly detection live, the real-time feed was redundant for the things that mattered and noisy for the things that didn't. So we removed it.

What replaced it

The new manager landing page is a once-a-day digest of what happened in your warehouse yesterday: high-level totals, the anomaly alerts that fired, the cycle count recommendations the system produced overnight, and a short list of what needs attention today. The page does not auto-refresh. Looking at it more than once a day produces no new information.

For the rare cases where someone genuinely needs to see live activity (a supervisor watching a difficult shift, a manager investigating a specific incident) we kept the feed view, but it's no longer the default and you have to navigate to it explicitly. Usage is down 94% since we made the change. We expect it to settle around 5% of its old volume, which seems about right.

What we kept from this

Real-time is a default that should be earned, not assumed. In our experience, the things that genuinely need real-time visibility (a tank approaching critical temperature, a fire alarm, an order pickup at the dock) are very few. The things people enjoy watching in real time (events scrolling past, dashboard numbers ticking up) are mostly entertainment.

We think this is true for a lot more software than warehouse dashboards. We have gotten quietly skeptical of any feature that's primarily justified by "you can watch it happen live." Live is not the same as useful.