Email research by Victoria Bellotti from PARC

PIM Research at PARC

Victoria BellottiPrincipal Scientist ([email protected])

Overview

• Personal information management (PIM) in the wild– And overload

• Embedding resources in email• Activity management

• What is PIM?– Personal information management means dealing with documents,

messages, scheduling events, to-dos, contacts, notes– Essentially the work we do to make it possible to do our work

© 2010 PARC | Confidential

Postulating PIM

3 of 25

The Reality of PIM

PARC | 4

Overload: Analysis of Time Spent in Email

• Microanalysis of samples of video observation of email triage– The time that people are focused on dealing with incoming email– Heavily interleaved with:

• Reading, skimming, editing, organizing, prioritizing, phone calls etc.

• Breakdown of time spent– 23.1% reading email– 6.2% scanning inbox– 2.4% deleting messages– 2% looking for messages– 9.5% filing messages– 1.1% spent adding attachments– 0.8% opening attachments– Most of the rest spent writing email and editing documents

• 20% of time looking around, searching for and organizing information– This likely overflows into the rest of the day since email is an archive

5 of 25

Overload: Analysis of Thread Complexity

• Quality not quantity– ~50% messages are threaded– Index of complexity

• No. of threads X (days per thread/steps per thread)

– Seems to be a better indicator of overloading than quantity

– Obviously because there’s more to remember to keep track of

6 of 25

Active threads of the manager who complained the most about overload

Personal Knowledge Pad

Snapshot To-do Study

• Average about 70 to-dos and 11 places• Only 14% of to-dos on paper-lists and e-lists• 2/3 online, 36% in email, 12% in e-calendar

– Distributed across the workplace and elsewhere

• The to-do doesn’t describe the task– Natural language may not be used– Contextual and personal cue

• To-dos have multiple roles:– Reminders: “I would like to remember to do this at an appropriate time”– Planning tools: “What must I do next?”; “What needs doing soon?”– Status indicators: “Done”; “Important”; “Priority”– Indices: “What content is involved in this task?”; “How do I access it?”

• A significant minority of to-dos may not get done

All(most) in the Head• Non specific• Acronyms• Incomplete sentences• Nonsense• Illegible

A relatively tidy and

explicit list

An untidy and less explicit list

“Beth blah blah”

Manager at PARC

To-dos in the Wild

• We interviewed people in detail about their to-dos once a week for four weeks with a final 5th interview.

• We classified them– What they were about and where they were stored– We also coded them for about 30 factors that might affect their getting

done, e.g., importance, consequence of not doing, difficulty, etc.

• Each week we asked whether the last week’s to-dos were done

PARC | 10

Factor Significance (random chance of data)

Urgency <0.1%

Customer <0.1%

Is a meeting <0.1%

Involving others (not mtg) <0.1%

Importance 0.1%

Non-discretionary 1.5%

Common 5.6%

Significant Determinants of Prioritization: Getting Things Done in a Week

Can’t-do-it-now tasks

Hard-to-forget tasksHaving no reminder 1.2%

On a to-do list negative <0.1%

Social

Conclusions

• People are good at prioritizing– Only 1% of cases of dropping the ball (but none high priority)

• They just need more help with the PIM– Resources need to be embedded in their work habitat

PARC | 12

Embedding Resources in Email

16 of 32

17 of 25

TaskMasterIn a small trial half of its users continued using it for months after end of study even though it lacked many features of Outlook

Optimizing for Activity Inferencing(under DARPA CALO Program)

PARC | 18

Project Objectives

• Goals– Simplify PIM and activity management– UI that increases explicitness of activity context for better ML

• Design Innovation– UX construct “Activities” that people can interact with

• System offers different human-meaningful ‘types’ (e.g. meeting, hiring)– User creates instances of each type

• System populates the instance with predetermined containers & behaviors– When user drags content to activity good stuff happens– Meanwhile machine learns about this instance of the human

activity

• RQ1. Will users adopt pre-designed structures?

• RQ2. Can we incent users to label their content precisely?

Activity-Centered Task Assistant (ACTA) embedded in Outlook

Drag-and-drop anything into Activity: automatic organization into contacts, documents, correspondence

Pre-designed folder component structure

TV-ACTATaskVista (TV) to-do list

Drag-and-drop or type-in to-do and Promote to Activity

Paper to-dos

More Features: Unified Content Collection

Structured Documents

Drag-and-drop Agenda with Attendees and Final Materials Presentations and Documents

Structured Email:

One menu-selection to

email agenda to all

Attendees

Useful Activity-Related Forms Links

Instant Map

No need to type in address again; address came from agenda

Evaluation

• RQ1. Will users adopt pre-designed structures? – Yes, more Activities created than folders

• RQ2. Can we incent users to label their content? – Yes, users selected specific Activity types and used components

• Users find Activity template approach appealing in spite of bugs and even without ML benefits– Justifies further exploration of this approach

Ongoing Research: Logging and Visualizing plus Activity Inferencing

PARC | 27

Hybrid Field Research

PARC | 28