top of page

Suggested Tasks in Outlook

A look at Microsoft's mission to help

you keep your promises at work

It's pretty hard to remember everything you commit to.

Think about how many emails you send and receive at work. There's a decent chance it's a lot. Now, think about how many of those emails either contain someone asking you to do something, or a promise you've made to someone.

How many times have you forgotten about a commitment or request in email? I do it all the time. There's just too much going on to keep track of it all.

What if your assistant kept track of your tasks for you?

My role

In 2018 I was tasked with taking what we learned from shipping predictive assistance in Skype and applying it throughout other Microsoft products, specifically in the form of suggested tasks within Outlook.

I was the responsible for defining AI's role in the M365 task lifecycle, shipping several successful new experiences and establishing a framework for ML-driven task extraction.

What I did

  • Interaction design

  • Product vision

  • Defining design principles

  • Sprint facilitation

  • Lifecycle mapping

  • ​Research coordination

  • Executive presentations

  • Fostering relationships with partners

  • Documenting patterns and best practices

Aligning on goals

Before beginning any design work, I brought together PM, research, and engineering leaders together as part of a 5-day sprint in Bellevue, Washington. People joined from Berlin, Sunnyvale, Tel Aviv, and Redmond, with the goal of agreeing to a unified vision.


How do email users currently triage their incoming tasks?

At the time, if a user received a task in Outlook, their primary method of saving that task if by flagging for follow-up. Yes, that famous little red flag. I had so many of them littering my inbox. 🚩

One of the problems with this flag is that users effectively set it and forget it. It's not until much later after the user has dropped out of the funnel that they remember, and sometimes forgetting to follow up comes with a cost, like reputation and job performance.

Competitive analysis

The only real competitor Outlook has in this space is Gmail. Recently they introduced a feature called "nudging", where users receive an inline hint after several days of inaction. MyAnalytics also has a nearly identical experience already within Outlook.


Gmail also offers the ability to drag emails into the sidebar to instantly add to Google Tasks as a file attachment. A drawback to this method is that the user still has to manually enter what the task is.

What role could Cortana play in the task lifecycle?


Offer action

Cortana helps you complete your task quickly and in-the-flow, or save it for later


Infer intent

Cortana finds requests and commitments made in your messages and email

Cortana helps you stay honest and productive by making sure you don't forget

Prevent drop-off


Go where the users are

Email is the lowest common denominator when it comes to how people communicate at work. It’s typically the first thing an enterprise user checks when they sit down at their desk, and Cortana is uniquely positioned to leverage this platform in the workplace. This was the basis of our hypothesis for explorations.

In Outlook, there's already a space for nudging and AI insights...


“You replied 3 days ago”


“This email has been flagged for follow-up”


“This message is high priority"

What if this was also the space for predictive assistance?


Add to tasks

Block time

Set a reminder


We can do better inbox nudging too

Cortana can provide powerful inline nudge actions on inbox list items as well.

User journeys and promises


Keeping you honest


Helping you stay focused

Being where you need us


As we moved towards making Suggested Tasks a reality, we narrowed our scope down to 3 hero scenarios that highlighted our user promises:




Very organized; Getting things done leads to high sense of accomplishment and peace of mind


Spontaneous achievers

Struggles to balance work and personal life; longs for technology that make this easy for them




These people are a little bit of both. They might be very structured in some ways yet struggle in others.

Who are we designing for?

There are three core archetypes that we design for as we think about M365 scenarios for the modern workplace.

Android Copy 16.png
Android Copy 19.png
Layer 3.png

When Jen opens her inbox, she sees a new message from her boss.


Before she opens the message, she notices a whisper from Cortana, letting her know that the message may contain a request for her.

When Jen taps on the whisper, she is shown the extracted request along with options to act.


This preview provides Jen with a scoped view of what she needs in order to act, saving her time from reading an entire email.

Later after lunch, Jen receives a reminder notification about the task she had created earlier.


Meet Jen

Jen is a structured achiever who likes to feel in control of her workload.

While she’s out to lunch, she gets a notification about several new emails...

Scenario 1

Helping you stay focused


Meet Sarah

Sarah is a spontaneous achiever who juggles alot at once and sometimes loses track of her tasks.

Late in the day her boss sends an important request that gets buried in her inbox...

Scenario 2

Keeping you honest


When Sarah gets to work the next day, she finds an email from Cortana in her inbox.


In the email, Sarah sees the missed request from her boss with options to act immediately or triage for later.

MS Todo Suggestions Copy.png

Darren notices his suggestions drawer has a message from Cortana containing a suggested task from his email.

When Darren adds Cortana’s suggestion, it’s added to this daily task list, which syncs across all of his M365 endpoints.


Meet Darren

Darren is a hybrid achiever who checks his tasks at the beginning of every day.

Before he gets out of bed in the morning, he opens his To Do app...

Scenario 3

Being where you need us

Mapping an experience framework

These user journeys are great and all, but without some rules that govern how and when we show a suggested task at a given time, this will all be a chaotic mess. As part of our ongoing sprint, myself, my PM team, the engineering team, and the data science team all gathered together to align on a framework.

This framework builds upon the logic derived from our smart suggestion model and helps to determine a criteria by which actions are surfaced inline vs when they are sent via email the next day.


Validation & iteration

Given its low dev cost, the first experience we had an opportunity to validate was our email experience, named "Heads Up". We tested a variety of options related to information density, progressive disclosure, and context-setting language.

2 design options tested

Basic Email_Desktop.png

Model A


Model B

We quickly learned that all participants preferred Model B due to its simplicity and the perceived value and importance of the information it conveyed. Many users held the opinion that calendar insights would not be valuable in the context of an email.

Rolling out to G.A.

While we are still hard at work building our other two experiences, Heads Up email has already shipped to Cortana users with an O365 account.

Of our potential users, 70% continue to open Heads Up week-over-week.

Of that 70%, we receive about 15% SATs and only 4% DSATs, with nearly all related to the accuracy of extraction.


This success has lead to the adoption of Heads Up by Outlook, with plans to ship to all O365 subscribers.

Moving forward

Heads Up continues to shine for Outlook and Cortana, with steady progress being made as we build smart suggestions into Outlook.

In the future, look forward to seeing actionable insights in Heads Up mail, with focus shifting away from task triage and more on the instant resolution of tasks, such as file sharing, instant replies, calendar management, and reminder creation.

I'm very excited for what the future holds for this product. Feel free to give it a try 😁





Productivity voice skills for Cortana


Smart replies & predictive assistance


Conversational UI

for Cortana

bottom of page