Skip to main content

Work Discovery Administrator Guide

Administrators install Work Discovery from the Marketplace, set it up, take it live, and read the reports. Work Discovery is not offered on Strongly Cloud. The administrator who installs it becomes its owner: its workflows and model calls run as that administrator.

The plugin page​

Plugins > Work Discovery (/plugins/work-discovery) is for administrators only. It has:

  • the Enabled switch in the page header. While Work Discovery is not live the switch reads Not live; turning it on opens Go live. Turning it off opens the Turn off choices.
  • one row of tabs: Setup (before go-live only), Settings, then Task list and Submission health (once live).

The Settings tab holds every setting: Reporting, Models, Submission and retention, Onboarding, Performance and Capture, then Turn off at the bottom. Its one Save saves everything on the tab as a new settings version. Every change is recorded in the audit log with who made it, when, and the values before and after. Each submitted day records the settings version it was submitted under.

Until Work Discovery is live, only administrators see it, and nobody is captured until they join.

Setup and go live​

The Setup tab shows the checklist:

  • Labeling and embedding models chosen
  • Report detail level chosen
  • Auto-submit time set
  • Onboarding window length set

An unfinished item opens Settings. Once every item is done, go live. Going live makes Work Discovery visible to everyone signed in under Operations, lets them join, and opens the onboarding window.

Settings​

Reporting​

  • Report detail level:

    LevelWhat administrators see
    Aggregate (default)Totals by organization, department, role or team only
    PseudonymousTotals plus a row per person under a stable pseudonym
    NamedTotals plus a row per person by name. Requires an attestation that people were given notice and that a legal basis exists

    Days submitted under a stricter level are never shown at a looser one, and people are told when the level changes.

  • Time granularity (day, week or month; default week) and app detail (app names, or app categories only; default app names).

  • Minimum group size (2 to 100 people, default 3): the privacy minimum for aggregate reports. See Minimum group size and folding.

  • What submissions include: redacted window titles and paths (default off), redacted meeting titles (default on), and manual-time entries (default on).

Models​

  • One set of models for the whole organization, so the same task done by different people is grouped as one task:
    • Labeling model (chat): labels every session in English. High volume, about 25k tokens per person per day.
    • Embedding model: maps labels to the task list and suggests tasks during onboarding.
    • Analysis model (optional): writes the Insights summary.
  • Only models you can use in AI Gateway are listed. Changing a model creates a new version, and later labels record it.
  • If a model is unavailable, the stage fails and shows the error, and days wait. Nothing switches to another model.

Notice​

What people are told before they join is on What Work Discovery captures.

Capture​

One table lists every capture rule, with search, sort and paging: App or site, Rule (Excluded app, Excluded site or Meeting app) and Source (Built-in or Organization). Add a rule with Rule, App or site and Add; remove an organization rule with its delete button.

  • Always excluded, not removable: password managers, banking and payment sites, health portals, personal webmail, HR self-service, private and incognito windows, and Strongly Computer itself.
  • The organization's excluded apps and sites, added on top. People can add their own exclusions in My Work > Preferences.
  • Meeting apps: the apps that count as meetings while they use the microphone or camera.
  • Idle after: a session ends after this many minutes without input (1 to 60, default 5).
  • Session split: 1 to 60 minutes (default 5).

Submission and retention​

  • Auto-submit: days a person does not submit are submitted automatically this many days after the workday (1 to 14, default 3), at this time and time zone (default 18:00). Choose a time inside the low-usage window (see Performance).
  • Retention: submitted days are kept on the server this many months (1 to 120, default 12). A computer deletes a day as soon as Strongly confirms its submission; people see their submitted days in Strongly. Unsubmitted days are never purged.

Onboarding​

The onboarding window length (3 to 90 days, default 14), when it opened, when the task list was approved, and when it closed. During the window labels are provisional and the task list is built from them. The window closes when its length has passed and the task list is approved in Task list. It never closes on its own with an unapproved task list.

Performance​

Reports are computed in the low-usage window (default 20:00 to 06:00), and only while the platform is below these thresholds:

ThresholdRangeDefault
CPU5 to 95%50%
Memory5 to 95%70%
Running executionsshare of the busiest hour, above 0 and at most 10.2
Gateway requestsshare of the busiest hour, above 0 and at most 10.2

Labeling concurrency: how many labeling runs may run at once across the organization (1 to 200, default 10).

Task list​

Once live, Task list holds:

  • Suggested tasks the engine found in what people submitted: accept, rename, reject, or select several and merge them.
  • The accepted task list: labels are mapped to these tasks. Aliases are other names people use for the same task.
  • Rejected tasks.

Approve task list closes the onboarding window once its length has passed. After it closes, computers label against the approved task list, and provisional labels in submitted days are mapped to it (people's submitted days themselves are not edited).

Submission health​

Once live, Submission health shows:

  • participants, and who submitted last week
  • held days, with their reasons
  • every computer, with old versions flagged
  • labels not in the task list
  • the onboarding window
  • the engine's last run and the low-usage signal
  • labeling failures

In Admin > Users, a Behind badge means a day is held: waiting for labeling, or a device is behind.

Turn off​

Turn off, at the bottom of Settings (also opened by switching Enabled off):

  • Disable keeps the plugin installed so it can be turned back on. Uninstall removes it. Capture stops on every computer either way.
  • Keep the data: everything stays, unreadable while Work Discovery is off, and returns when it is turned back on.
  • Delete the data: every submission, report, task list and setting is deleted, and joined computers delete their local Work Discovery data. This cannot be undone.

You must choose what happens to the data, and confirm, before Turn off proceeds. Work profiles (role, department) are core platform data and are not affected.

Reports​

Administrators see the report tabs on Operations > Work Discovery, after the My Work tabs: Overview, Insights, By group, Task catalog, and People (only at the pseudonymous or named level). Each report tab has the period picker (from, to, last 7, 30 or 90 days) and Refresh. Every report follows the report detail level and only uses submitted days.

TabShows
OverviewTotal hours submitted, top categories and top tasks, meeting share, and coverage
InsightsAutomation opportunities, biggest movers, the summary and common task sequences
By groupHours, people, meeting hours and change vs prior period per group, with hours by category
Task catalogEach task with category, hours per month, people, runs per week, median minutes per run, spread (hours per person) and apps. Open a task for its detail: distribution across people and groups, apps, weekly trend and people's notes verbatim
PeoplePer-person totals by category, listed alphabetically, never ranked

How figures are computed​

  • Coverage: every report is based on the people who submitted in the period out of everyone who joined, and the hours submitted (for example "142 of 180 joined people, 3,410 hours submitted").
  • Level: at the aggregate level no row is about one person; people are shown as counts. At the pseudonymous and named levels People lists everyone in alphabetical order (never ranked), under a stable pseudonym or by name. A day is never shown at more detail than the level in force when it was submitted.
  • By group needs a work profile field marked for reports (Admin > Users > Work profile fields).
  • Refresh queues every submitted day; they are recomputed at the next low-usage window.
  • Onboarding: while the onboarding window is open the task list is not final and labels are provisional.

Minimum group size and folding​

At the aggregate level only:

  • Every report row totals at least the minimum group size (Settings > Reporting, 2 to 100 people).
  • A department, role or other group smaller than that is folded into one Other (small groups) row. That row is shown only if it reaches the minimum itself; otherwise its hours are left out, and the hours left out are counted.
  • A task done by fewer people than the minimum is not shown.
  • Automation opportunities and task sequences need at least this many people too.

The pseudonymous and named levels show every group, including a group of one person.

Insights​

  • Automation opportunities: Score = total hours x repetitiveness x people, where repetitiveness = occurrences per person per week x regularity (regularity is the share of the period's weeks each person did the task).
    • A task is a repetitive pattern at 3 or more occurrences per person per week in at least 50% of the weeks, with under 50% of its time in meetings.
    • The saving estimate assumes 50% of a repetitive pattern's hours can be saved, and claims no saving for other tasks.
    • Open a task for its detail.
  • Biggest movers: hours per week against the prior period of the same length, for categories and tasks.
  • Summary: written by the analysis model (Settings > Models) from the findings, through the summarize workflow. With no analysis model set, only the deterministic findings are listed.
  • Common task sequences: tasks done one after another within a day, across people (process hints). Shown when seen in at least 2 people's days, and at least the minimum group size at the aggregate level.