Muscle Memory
1. What it is
Muscle Memory watches closed operational arcs — an incident resolved and its corrective action holding, an automation with a track record of clean runs, a Service Desk pattern that keeps bouncing — and distills each into a lesson backed by the records behind it. No evidence, no lesson.
The app is licensed and opt-in. An admin enables it from the Apps Marketplace; it requires a license. Once enabled, distillers run nightly and a retroactive backfill scans existing history on first enable.
What it is not: Muscle Memory does not ask anyone to write things down. It learns from outcomes that already closed in other apps. Managers can add lessons manually, but the primary source is automated distillation.
2. Standing it up
- Go to Apps Marketplace → Muscle Memory and enable the app.
- The first enable triggers a retroactive backfill of your operational history. When the backfill finishes, admins receive an in-app notification summarizing how many lessons were found.
- Open Muscle Memory → Settings to review which distillers are active. Seven sources are on by default; the scheduling distiller is off by default and labelled experimental.
- Assign access. By default the app is available to all members. Lessons describe operations, never individuals, so every member with access can read them. To curate (confirm, correct, forget, or add lessons), a user must be a manager or above.
- Confirm it worked: open the Business Memory tab. Lessons appear automatically as the nightly distiller sweep closes arcs. To see the app fully rendered before real lessons arrive, go to Settings → Sample data → Load sample data.
If no lessons appear after the backfill: the distillers need closed arcs to work with. An incident whose corrective action has not yet completed, or one still inside its recurrence window, is not eligible. The default recurrence window is 60 days.
3. How it fits together
Lessons, evidence, and confidence
A lesson is one specific, operational statement — for example, “Incidents of type Slip/Trip at the Downtown Warehouse stopped after the non-slip matting was installed (no recurrence in 60 days).” Every lesson carries a confidence score between 0 and 1, a domain (such as Safety Hub, Scheduling, or Service Desk), and links to the records it was distilled from (its evidence).
Confidence is not static. When the same pattern is observed again, the lesson is reinforced and its confidence rises. When a manager confirms a lesson, confidence increases. When later incidents contradict a “held” lesson, confidence decays — and if it falls below your confidence floor (default 0.60), the lesson is marked Superseded. Superseded lessons stay visible with a badge but stop appearing in Ask AI answers.
Three statuses
- Active — the lesson stands and can inform Ask AI answers (if above the confidence floor).
- Superseded — later evidence contradicted it; visible with a badge, excluded from recall. A manager can reinstate it with the Still holds action.
- Forgotten — a manager archived it. Hidden from all surfaces. The distillers will not re-learn a forgotten lesson.
Business Memory vs Personal Memory
These are separate stores serving different purposes:
- Business Memory (the default tab) holds organization-wide lessons distilled from operational outcomes. Everyone with app access can read them; managers curate them.
- Personal Memory shows the durable facts the AI assistant remembers about the individual user across sessions — preferences, what they are working on. Only the user can see their own personal memory, and they can erase any item or everything at once.
Distillers — where lessons come from
Each distiller watches a specific type of operational arc:
| Distiller | What it watches | Default |
|---|---|---|
| Incident + corrective action outcomes | A corrective action completed and the recurrence window elapsed — verdict: Held or Didn’t hold | On |
| Automation (playbook) outcomes | Playbooks with enough run history — verdict: Adopted or Rolled back | On |
| Service Desk resolution patterns | Request types with a consistent resolution track record — verdict: Dependable or Bouncing | On |
| Leave/time-off decision patterns | Leave types with a decision record across approvals | On |
| Onboarding + training readiness | Learning paths with a completion track record | On |
| Won RFP bids | RFP bids that were won, along with library reuse data | On |
| Staffing/scheduling correlations | Correlations between staffing and outcomes (experimental) | Off |
Each distiller is individually togglable in Settings → What Muscle Memory learns from.
Ask AI recall
When a user asks a question in Ask AI, the recall service searches for relevant lessons (using semantic similarity and keyword matching) and injects them into the AI’s context. The AI cites the lesson and links back to it. This is how the business’s institutional knowledge reaches employees without anyone opening the help centre. Recall is on by default and controlled by the Recall business lessons in Ask AI answers setting.
4. Running it
Reviewing lessons
Open the Business Memory tab. Filter by status (All lessons, Needs review, Verified, Superseded), by domain, or search by keyword. Sort by Most relevant (last recalled, then confidence), Highest confidence, or Recently learned.
Managers see curation actions on each lesson:
- Confirm — the lesson is true; raises confidence.
- Correct — amends the lesson with a note (kept alongside the claim).
- Still holds — reinstates a superseded lesson and lifts its confidence back to at least your confidence floor.
- Forget lesson — archives the lesson permanently. It will not be re-learned.
Adding a lesson manually
Managers type one sentence into the text box at the top of the Business Memory tab and click Add lesson. The platform rewrites it into a structured claim with a domain and optional location — no form to fill. Near-duplicates reinforce the existing lesson instead of creating a second one. Sensitive text (passwords, ID numbers, account numbers) is rejected.
Non-managers see a Suggest lesson button instead. Suggestions are born unverified and sit below the recall confidence floor until a manager confirms them.
Exporting lessons
Managers can download every lesson as a CSV from the Export button on the Business Memory tab. The file includes each lesson’s claim, domain, confidence, verification history, correction notes, evidence, and measured outcome lift.
Weekly digest
Every Monday, managers and admins receive an in-app notification summarizing lessons minted in the past week and any manager-added lessons that have not been re-confirmed in 6 months.
Notifications
Muscle Memory sends in-app notifications for:
- Backfill complete — admins, when the first-enable retroactive scan finishes.
- Corrective action didn’t hold — managers, when an incident arc closes with a negative verdict (action request).
- Lesson superseded — managers, when contradiction decay drops a lesson below the confidence floor (action request).
- Teammate suggested a lesson — managers, when a non-manager submits a lesson for review (action request).
Users control these from their notification preferences under the Business Memory Lessons category.
5. Settings
All settings are at Muscle Memory → Settings. Only users who can edit the app see this tab.
| Setting | Card | Default | What it changes |
|---|---|---|---|
| Muscle Memory AI assistant in Ask AI | Muscle Memory Agent | On | Enables the Muscle Memory agent in Ask AI — browse, search, and (for managers) curate lessons conversationally |
| Distill incident + corrective-action outcomes | What Muscle Memory learns from | On | Incident distiller runs on the nightly sweep |
| Remember durable facts for each user (personal memory) | What Muscle Memory learns from | On | The assistant stores per-user facts across sessions |
| Distill automation (playbook) outcomes | What Muscle Memory learns from | On | Playbook-outcome distiller runs |
| Distill staffing/scheduling correlations (experimental) | What Muscle Memory learns from | Off | Scheduling distiller runs — experimental, opt-in |
| Distill Service Desk resolution patterns | What Muscle Memory learns from | On | Service Desk distiller runs |
| Distill won RFP bids into lessons | What Muscle Memory learns from | On | RFP win distiller runs |
| Distill leave/time-off decision patterns | What Muscle Memory learns from | On | Leave decision distiller runs |
| Distill onboarding + training readiness patterns | What Muscle Memory learns from | On | Onboarding readiness distiller runs |
| Recall business lessons in Ask AI answers | What Muscle Memory learns from | On | Lessons are injected into Ask AI answers when relevant |
| Keep a history of where each lesson was used | Lesson usage history | On | Records a usage event each time a lesson enters an answer |
| Also record who received the answer | Lesson usage history | Off | Adds the recipient’s name to each usage record — off by default because lessons describe operations, not individuals |
| Keep lesson-usage history for (days, 0 = forever) | Lesson usage history | 365 | Usage records older than this are deleted on the nightly sweep; 0 keeps them indefinitely |
| Minimum confidence to keep a lesson (0–1) | Thresholds | 0.60 | A lesson below this confidence is dropped rather than kept |
| Recurrence window for incident lessons (days) | Thresholds | 60 | How far back the incident distiller looks for a recurring pattern |
6. More help
- Muscle Memory FAQ — specific setup and operating questions.
- Ask AI — the assistant answers questions about Muscle Memory from these articles, and the Muscle Memory agent can list, search, and help curate lessons conversationally.