Muscle Memory FAQ
Answers to common setup and operating questions about Muscle Memory.
For what the app is and how to set it up from scratch, see the
Muscle Memory Overview.
Setup
Will the app be empty when I first turn it on?
No. Enabling the app triggers a retroactive backfill that scans your existing operational history — incidents, corrective actions, automation run records — and distills what it finds into lessons. Admins receive an in-app notification when the backfill completes, telling them how many lessons were minted. Tenants evaluating the app with no history can load sample data from the Settings tab.
Which distillers are on by default, and which are off?
Seven of the eight distillers are on by default: incidents and corrective actions, automations, Service Desk resolution patterns, leave/time-off decisions, onboarding and training readiness, and won RFP bids. The staffing/scheduling correlation distiller ships off — it is experimental and must be turned on explicitly in Settings → What Muscle Memory learns from.
Do I need to configure anything before lessons appear?
No. With the defaults, the nightly distillation sweep and the retroactive backfill will start minting lessons from your existing data as soon as the app is enabled. The only prerequisite is that the source apps (Safety Hub, Service Desk, etc.) already have operational records.
Permissions and access
Who can see business lessons, and who can curate them?
Every member the app is published to can browse and search business lessons — they describe operations, never individuals. That read is deliberate and is not configurable; if you need a narrower readership, restrict the app’s audience under Access Management in the Apps Marketplace.
Everything else is yours to set, on Settings → Who can do what:
- Who can curate business lessons? — confirm, correct, forget, reinstate, add a lesson directly, and export the corpus. Defaults to managers and admins, which is what the app enforced before this setting existed. You can also choose app admins only, anyone with direct reports, or specific groups.
- Who can see Muscle Memory analytics? — the Analytics tab and the same figures over the API and the Ask AI assistant. Defaults to the same people who curate, but you can grant it more widely (for example, everyone) without granting anyone the ability to change a lesson.
- Who can suggest a lesson? — defaults to everyone with app access.
Business admins and anyone you designate as a Muscle Memory app administrator can always do all three, whatever these settings say. Whichever policy is in force applies identically on the desktop tab, the mobile view, the Ask AI agent, and the REST API — all four ask the same question.
Who receives the weekly digest and alert notifications?
Managers, admins, and super admins receive the weekly Monday digest and the time-sensitive alerts (corrective action did not hold, lesson superseded, teammate suggested a lesson). Notifications are routed under the Business Memory Lessons notification preference, so each user can mute them from the standard notification settings page.
Day-to-day operations
How do I add a lesson that no system ever recorded?
Type one sentence in the Add a lesson field on the Business Memory tab. AI restructures it into a claim with a domain and optional location scope. Lessons must be between 15 and 600 characters. If the sentence is too close to an existing lesson, it reinforces the existing one instead of creating a duplicate. Lessons added by a curator are born verified at 80% confidence.
Can employees who are not managers suggest lessons?
Yes, by default. Any member with app access can suggest a lesson from the Business Memory tab or from the mobile view — the same one-sentence box, which reads Suggest lesson for them and Add lesson for a curator. (An admin can narrow this under Settings → Who can do what → Who can suggest a lesson?.) Suggestions are born unverified at 55% confidence — below the 60% recall floor — so they cannot shape Ask AI answers until a manager confirms them. Suggestions are rate-limited to 5 per user per hour. Managers receive an in-app notification to review each suggestion.
What happens when a manager confirms a lesson?
Confirming boosts the lesson’s confidence by 10 percentage points (capped at 100%) and records who verified it and when. If the lesson was below the recall floor before confirmation, the boost may bring it above the floor so it starts informing Ask AI answers.
What does “correct” do, and how is it different from editing a lesson?
Correcting a lesson attaches a note alongside the original claim — the note travels with the lesson everywhere it appears. The claim text itself is not rewritten. The correction counts as a human review (it records who corrected it and when) but does not boost confidence.
What happens when I forget a lesson?
Forgetting sets the lesson’s status to Forgotten. It disappears from all list views, stops informing Ask AI answers, and — importantly — the distillers will never re-learn it, because the original dedup key is preserved. Nothing is hard-deleted; forgotten lessons are archived.
How do I bring back a lesson that was marked superseded?
A manager can reinstate a superseded lesson from the Business Memory tab (or via the Ask AI agent or REST API). Reinstating lifts the lesson’s confidence back to the confidence floor and returns its status to Active, so it re-enters Ask AI recall. Reinstate works only on superseded lessons — it cannot undo a forget.
When something looks wrong
Why isn’t a lesson showing up in Ask AI answers?
A lesson must meet all of these conditions to be recalled into an Ask AI answer: its status must be Active (not superseded or forgotten), its confidence must be at or above the confidence floor (default 0.60), and the Recall business lessons in Ask AI answers toggle must be on in Settings. Sample lessons (loaded from the Settings tab) are always excluded from recall.
A lesson says “Superseded” — what does that mean?
Later incidents of the same type at the same location contradicted the lesson’s claim. Each contradicting incident reduces confidence by 10 percentage points. When confidence falls below your confidence floor (default 0.60), the lesson is marked Superseded: it stops informing Ask AI answers but stays visible on the Business Memory tab with all its evidence preserved. Managers see an in-app alert when a lesson is superseded by recent contradictions.
My corrective action resolved the problem, but I don’t see a lesson yet
The incident distiller waits until the full recurrence window has elapsed after a corrective action completes — the default is 60 days. Only then is the verdict final (held or did not hold). The distiller also runs on a nightly sweep, not in real time, so a newly eligible arc may take up to a day to appear.
I can’t find the lesson a notification linked me to
Notification deep links carry the lesson’s ID, and the list sorts that lesson to the top of the page. If the lesson has since been forgotten, it will not appear. Check the Forgotten count on the Analytics tab to see whether someone else has already reviewed and removed it.
What is the Learning tab?
It shows how your business is learning, in order, where the Business Memory tab shows what it knows. Patterns forming lists the records Muscle Memory is watching that do not have enough history for a lesson yet — counts only, for example “Service Desk · Access — 7 of 10 resolved requests needed”. Below it, a timeline shows each lesson being learned, confirmed by a curator, relied on by the assistant, tested in Loop, or retired. The patterns panel follows the same permission as Analytics; everyone who can open the app sees the timeline.
Why does a lesson appear on the leave request form or an approval queue?
When a lesson is about exactly what someone is doing, it is shown where they are doing it, with a link to the records behind it. A person requesting leave sees the lesson for that leave type beside the form. An approver sees it above the queue they are about to work: leave approvals (the lessons for the leave types waiting on that page), timesheet approvals and shift claims. Only active lessons at or above your confidence floor are shown, and only to people who could also read them in Muscle Memory. An admin can turn this off with Show a lesson where the decision is made in Settings.
Which approvals does Muscle Memory learn from?
Leave decisions, Service Desk resolutions and Frontline Execution hand-offs have their own lessons. Beyond those, it counts how dependably each kind of request is approved in the apps you have switched on: purchase requisitions and purchase orders, expense reports, shift claims, schedule approvals, timesheets, Service Desk access requests, asset requests, profile-change requests and form reviews. These lessons are made of counts only — never who asked, who decided, or anything that was written on the record. Only decisions a person made are counted: an approval made automatically by a rule, an instant grant or an AI agent is left out, and so is a record that was created already approved. Profile-change requests are counted per field, and never for a sensitive field (for example gender, date of birth, disability or veteran status, pay, or a custom field). Some records are never learned from by design, whatever is switched on: HR files, disciplinary actions, benefits, resignations, recruiting screenings, performance reviews, incident investigations, banked-time payouts and pay changes.
What does “Apply this change” do?
Some lessons propose one Frontline Execution setting change (for example, raising the daily directive budget). A Frontline Execution admin can apply it in one click from the lesson or from Loop: the setting is saved exactly as the settings page would save it, and Loop starts a 30-day measurement of whether it helped. If Loop is not available to you, the change is still made and the message says it is not being measured.
Who gets the weekly email?
Every Monday, people who curate lessons and managers receive “What your business learned this week” — new lessons, how often the assistant relied on them, and (for curators only) how many lessons are waiting for review. It is its own notification category, Weekly Learning Digest, so anyone can turn the email or the in-app copy off without affecting review requests.
What is the difference between “used in an answer” and “offered to the assistant”?
Every time someone asks Ask AI a question, Muscle Memory offers the assistant the few lessons that look relevant. Most of the time the assistant does not need them. When a lesson actually shapes the answer, the assistant records that, and only then does it count as used — that is the number beside each lesson, the one on the Analytics tab, and the only case in which the answer shows Informed by N business lessons. “Offered” is shown next to it so you can see both. Uses recorded before this distinction existed show as offered.
Can I tell the assistant to remember something?
Yes. Say “remember that I prefer CSV exports” and it is saved to your private Personal Memory straight away, with a Saved to your memory note under the answer. If what you want remembered is about how the organization works (“month-end close is on the 5th working day”), the assistant tells you it will be sent to your lesson curators as a suggestion with your name on it; it appears under Needs review and informs answers only after a curator confirms it. Asking the assistant to remember something you previously erased puts it back.
What are playbook lessons, and does Muscle Memory read what my team writes?
Not unless an admin turns it on. Two settings — Draft playbooks from Service Desk resolution notes and Learn from the reasons people give when they reject the AI’s drafts and proposed actions — are off by default. When on, those notes are sent to your AI provider to find the practice they have in common (“when an access request comes in, the team confirms manager approval, then…”). The draft never names a person, arrives under Needs review, and is never used in an answer until a curator confirms it. A curator’s wording is never overwritten, and a forgotten playbook is not drafted again.
Why does the “Used in N answers” count show zero for a lesson I know was used?
If the Keep a history of where each lesson was used toggle is off in Settings, the app does not record individual usage events — the count reads zero because nothing was recorded, not because the lesson was never recalled. Turning the toggle on starts recording from that point forward; it does not backfill.
Personal memory
What is personal memory and who can see it?
Personal memory stores durable facts the AI assistant remembers about each user across sessions — it is strictly private. No manager, admin, or coworker can read it. Each user can review every remembered fact on the Personal Memory tab and erase any single fact or everything at once.
Can I turn personal memory off for the whole organization?
Yes. The Remember durable facts for each user (personal memory) toggle in Settings controls personal memory tenant-wide. Turning it off stops new facts from being stored without affecting business memory. Existing personal facts remain until the user erases them.
Licensing and limits
Does Muscle Memory require a licence?
Yes. Muscle Memory requires a licence and is opt-in per tenant — it ships disabled. An admin enables it from the Apps Marketplace. That enablement is the single consent moment for both memory layers (business and personal), which matters because the app reads real operational data.
Are there limits I should know about?
| What | Limit |
|---|---|
| Hand-written lesson length | 15–600 characters |
| Suggestion rate limit (non-curators) | 5 per user per hour |
| Lessons recalled per Ask AI answer | 4 |
| Confidence range | 0.00–1.00 |
| Default confidence floor | 0.60 |
| Default recurrence window | 60 days |
| Reinforcement confidence boost | +0.05 (capped at 0.95) |
| Confirmation confidence boost | +0.10 (capped at 1.00) |
| Contradiction confidence decay | −0.10 per contradicting incident |
| Manual lesson initial confidence | 0.80 (added by a curator) / 0.55 (suggestion) |
| Usage history retention | 365 days by default (0 = keep indefinitely) |
| Distillation schedule | Nightly sweep, plus a retroactive backfill at enablement |
| Re-attestation nudge | Hand-added lessons not re-confirmed or recalled in 6 months |
Who can export lessons?
Anyone your Who can curate business lessons? setting reaches — managers, admins and super admins by default — can export the full set of visible business lessons (active and superseded) as a CSV from the Business Memory tab. The export includes evidence links, verification details, correction notes, and measured outcome data.
More help
- Muscle Memory Overview
- Ask AI — the assistant answers Muscle Memory questions from these articles.