Email Overload

Outlook's Focused Inbox: How It Decides

Focused Inbox uses ML to split mail into Focused and Other. Here is what it actually considers, where it works, and where it falls short.

Outlook’s Focused Inbox has been the default attention-management feature in Outlook since 2016. The feature splits accepted mail into two views: Focused (priority) and Other (lower priority). Most users see it but few understand exactly how it makes the decision.

This post is the realistic 2026 explanation.

What Focused Inbox Is

Focused Inbox is a machine-learning model that runs on top of accepted mail in your Outlook account. It does not change which mail is accepted (that is the spam filter’s job). It changes how accepted mail is presented.

The output is binary: Focused or Other. Focused mail appears in the primary inbox view. Other mail appears in a separate tab, with a periodic notification summarizing what is in Other. The user can switch between the two views with a click.

The mechanism is per-account, probabilistic, and adapts over time as the user’s behavior changes.

What Signals the Filter Uses

The signals are not all public, but the documented and observed ones include:

Reply behavior. Senders you reply to often are routed to Focused. The signal is strong: a sender you have replied to in the last week is much more likely to land in Focused than a sender you have not.

Read promptness. Mail you open promptly after delivery is treated as more important than mail you ignore. The model uses time-to-open as an engagement signal.

Manual Move to Focused / Move to Other. Direct user input on which side of the line a sender belongs. The signal is the strongest because it is explicit.

Frequency of correspondence. Senders you exchange many messages with land in Focused. Volume creates strong signal.

Recency of correspondence. Senders you have heard from recently are routed more aggressively to Focused than senders you have not heard from in months.

Threading. Mail that continues an existing thread is treated as more important than fresh threads from new senders. Conversation continuity matters.

Sender characteristics. Beyond per-sender signals, the model considers sender domain reputation, sender authentication status, and other meta-signals.

Mailbox-wide patterns. Mail patterns at the mailbox level (typical engagement times, typical reading patterns) inform the threshold tuning.

The combined signal is a probability score that the message is engagement-worthy. Above a threshold, it goes to Focused. The threshold is calibrated by Microsoft to err on the side of Other for borderline cases. False negatives (mail in Other that should have been Focused) are common; false positives (junk in Focused) are rare.

Why the Filter Sometimes Gets It Wrong

The probabilistic nature of the filter produces predictable failure modes.

New senders without history. The filter has no behavioral data on new senders. The default decision for unknown senders is Other unless other strong signals push toward Focused. The new vendor whose first email is critical lands in Other.

Pattern shifts. When your inbox patterns change (new job, new project, new vendor relationship), the model lags behind. Mail from senders you would now consider Focused continues to land in Other for weeks.

Bulk mail with high engagement. Some bulk mail (newsletters you actually read, automated reports you care about) generates engagement signals. The filter sometimes routes bulk mail to Focused because of these signals.

Important mail outside engagement patterns. Mail received outside your typical engagement window (weekend, holiday, off-hours) gets less signal. Important mail received at unusual times sometimes lands in Other.

Conservative default routing. Microsoft’s calibration favors Other when in doubt. Many users report mail they would have wanted in Focused ending up in Other.

The filter is generally accurate for established correspondence patterns and less accurate for novel ones. The structural limit is that the filter is predictive and requires data to predict.

How to Train Focused Inbox

The filter improves with explicit feedback. Practical training:

Move to Focused / Move to Other. Right-click any message and select the appropriate option. The signal trains the model on this sender.

Always Move to Focused / Always Move to Other. A stronger version that creates a deterministic override for the specific sender. The mail bypasses the model and goes directly to the chosen tab. Useful for senders where probabilistic prediction is unreliable.

Reply to important senders. Reply behavior is one of the strongest signals.

Use mail flow rules (admin-level). For Workspace deployments, admins can create mail flow rules that route specific senders or domains to Focused or Other for the entire organization. Useful for vendor relationships that should be Focused for all employees.

The training takes weeks of consistent feedback. The filter is not retrained instantly; the signals accumulate.

Focused Inbox vs Junk Filter

The two systems operate at different layers and produce different decisions.

Junk filter. Decides whether mail is accepted or routed to Junk (spam). The decision is binary: deliver or move to Junk. Focused on identifying spam.

Focused Inbox. Decides whether accepted mail is shown in Focused or Other. The decision is also binary but operates after the Junk filter has accepted the mail. Focused on identifying engagement worth.

A message can be:

  • Rejected entirely (rare, mostly for blacklisted domains).
  • Routed to Junk (suspected spam).
  • Accepted and shown in Focused (predicted engagement).
  • Accepted and shown in Other (lower predicted engagement).

The two systems do not communicate directly. Both can have false positives and negatives. We covered the Junk side at Outlook’s hidden junk mail settings most people miss.

Where Focused Inbox Falls Short Structurally

The probabilistic mechanism has fundamental limits.

Predictive only. The filter predicts engagement based on past behavior. New senders, new patterns, and edge cases are predicted poorly.

Per-account adaptation. The filter is trained on your account specifically.

Inscrutable decisions. When the filter gets it wrong, the user has no way to see why. The chevron-equivalent is on or off; the reasoning is opaque.

Cannot enforce identity. The filter is about engagement prediction, not sender identity. There is no way to say “this sender always reaches my Focused tab” or “this sender always pays a cover charge before reaching me.” The mechanism is wrong-shaped for those queries.

Cannot reduce volume. Focused Inbox decides where mail goes within the inbox. It does not change what arrives. The volume problem is downstream of what Focused Inbox is designed to address.

We covered the related limits in the complete guide to Outlook rules in 2026.

How an Inbox-Layer Filter Composes With Focused Inbox

Rythm and Focused Inbox operate at different points in the pipeline:

Rythm. Sits at the inbox-arrival layer. Asks unknown senders for a small cover charge or holds them in a separate folder for review. Known senders walk in.

Focused Inbox. Sits at the engagement-prediction layer. Once mail is accepted into the inbox, it predicts engagement and routes between Focused and Other.

The two compose naturally:

  • New unknown senders go through Rythm’s cover charge gate or held-for-review folder.
  • Known senders (on the Rythm guest list) walk into the inbox.
  • Inside the inbox, Focused Inbox predicts engagement.

Rythm reduces the volume of unsolicited mail competing for Focused Inbox’s attention. Focused Inbox has cleaner signal because the sender population is cleaner. Both work better together than either alone.

A Specific Honest Note

Focused Inbox is a useful tool for the engagement-prediction layer. The probabilistic prediction works for established correspondence patterns and is predictably less accurate for novel ones.

What Focused Inbox cannot do is enforce identity-based filtering or change the cost of reaching the inbox. Those are different mechanisms at different layers. Rythm fills the identity-and-cost layer; Focused Inbox handles engagement prediction inside the inbox.

For the related guides, see the complete guide to Outlook rules in 2026, Outlook’s hidden junk mail settings most people miss, and how to whitelist senders in Outlook. For the equivalent post on Gmail, see Gmail’s Important filter: how it decides. Rythm is $1.65 per month, cancel anytime.

Ready to take back your inbox?

Secure My Inbox
outlook focused inbox focused inbox algorithm outlook priority outlook clutter ms 365 focused inbox