User experience
and interface design
in Pasadena, California

Once a product has more than one person working inside it, a plain email stops being enough. An activity feed, paired with a notification centre, gives a person one place to see what changed since they last looked. No digging through every screen required.

The two things are related but not the same. An activity feed is a record. It is a chronological list of events, useful for understanding what happened and when. It stays useful even for events a person already saw. A notification centre is a queue of things that need attention now. Once read or acted on, an item should leave the queue rather than linger.

Grouping matters more than almost anything in a feed that gets busy. Twelve lines saying a comment was added to the same document read as noise. One line saying twelve comments were added, expandable on demand, reads as information. Follow what a person cares about, usually the object being acted on. Not the raw order events arrived in.

Every entry needs a clear actor, a clear action, and a clear object, in that order: who did what to which thing. Passive phrasing that hides the actor, such as a document was updated, forces a click through just to find out who to ask. Name the actor plainly. It saves that step every time the feed is read.

Read and unread states need to be obvious without shouting. A small dot or a weight change on unread items is usually enough. Save a bright colour for items that truly need action. If everything in the feed looks urgent, a person soon learns to ignore the whole thing.

Timing decisions belong here too. Some events deserve an instant push, such as being mentioned by name. Others, like a routine status change nobody is watching, are better batched into a quieter summary sent once a day. Push every event the instant it happens, and people mute the channel within a week.

Preferences should be specific, not one all-or-nothing switch. A person might want an instant alert for direct mentions but a daily digest for general activity. Split these into a short list of toggles, grouped by event type rather than delivery channel. A person tunes the feed once. Then mostly forgets about it.

An empty state for the feed is worth designing on purpose, not left as a blank rectangle. A short line explaining what will appear once something happens turns an empty screen into a small nudge rather than a dead end. Add a suggestion for the action that would create the first entry.

Retention depends on how the feed gets used. A feed for catching up needs only a short window, perhaps a month, kept fast to load. A feed doubling as a record for disputes needs a longer memory and a way to search it. That is a different technical shape. Plan it as one from the start.

A home screen that shows different things to different people, based on what they did before, can genuinely save time. It can also quietly narrow a product. A person cannot find something that used to be easy to reach, simply because the system decided it was no longer relevant. Design its edges as carefully as its centre.

The cold start problem comes first. A brand new account has no history to personalize from. Showing a thin personalized surface to a first-time visitor is a poor start. Build a sensible default view instead. Base it on what most people need first. Let it carry a new account until real behaviour gives the system something to work with.

Personalized results need a visible reason attached where practical. A line such as shown because you viewed this recently turns a mysterious ranking into something a person can trust and adjust. Without that line, a personalized list can feel arbitrary even when the logic is sound. There is no way to check it otherwise.

A full, unfiltered view of everything should always exist nearby. Keep it one step from the personalized version. This matters most when the stakes are real: a missed item in a task list, a hidden setting, or an old document a person needs today. Personalization should narrow the default path. It should never close off the rest of the building.

Recommendation lists risk feeling repetitive if the same handful of items keep resurfacing simply because they scored well once. Rotate in items that are merely relevant, alongside the top scoring ones. That keeps a feed feeling alive rather than stuck on a loop. It also gives newer items a fair chance to be seen at all.

Negative signals are as useful as positive ones. They are often skipped in early designs. A simple way to say show me less of this, attached to an item rather than buried in settings, teaches the system faster than passive behaviour alone. It gives a person real control over the surface.

Explaining personalization is not the same as exposing every detail of how it works. A person rarely wants a breakdown of a scoring model. They want enough to feel the system is reasonable. And enough control to fix it when it is wrong. Detail slows a screen down. Trust does not need a diagram.

Turning personalization off should be a real, working option. Not a setting that quietly leaves half the behaviour in place. Some people, particularly those doing careful or comparative work, prefer a stable, unranked view every time. Respect that preference fully. Do not treat it as an edge case to tolerate.

Measuring whether a personalized surface works needs more than a click rate on the top item. A surface that gets clicks because it is the only thing shown differs from one that saved real time over the plain alternative. Compare task completion and return visits against the unpersonalized default. That is the honest test.