What is the Loop App?
Loop answers one question the rest of the platform could not: did that actually help?
Several things already happen across MangoApps independently. Apps detect opportunities and raise suggestions. Distillers turn closed operational arcs into durable lessons. Playbooks run unattended. What never existed was a place where those threads met and got scored — so “the AI is working well” was an opinion, not a measurement.
Loop is that place. It records a measurement around each intervention — a baseline before, a score after a fixed window, and the delta between them — and it hosts the surfaces that produce and act on those interventions: authoring automations, reading their run history, reviewing outcomes, administering suggestions, and inspecting agent corrections.
Loop is opt-in and licensed. It is included with the Platform Standard and Enterprise tiers rather than sold on its own, and an admin enables it per tenant.
What Loop is not: it is not where you run or stop the AI. Approvals, pausing an agent or a whole domain, the global kill switch, and agent health all live in AI Operations under Admin, and they stay there whether or not a tenant enables Loop. Emergency controls never sit behind an app licence.
Key Concepts (How It Works)
These are the ideas the rest of the article assumes. Most confusion about Loop comes from conflating two of them.
Measurement
A measurement is one record answering “did this specific intervention help?” It has a subject (the thing that intervened), a metric domain (what is being scored), a baseline captured at the moment of intervention, a post-window score captured when the window closes, and a delta.
Measurements are seeded by the platform, never hand-entered. There is no create or edit screen, deliberately: a ledger you can type into is not evidence.
The three statuses — and why “inconclusive” is not a failure
| Status | Meaning |
|---|---|
| Measuring (pending) | The window is still open. No verdict yet; the row shows how many days the window runs. |
| Measured | The window closed with enough signal. A baseline, an after score and a delta are all present. |
| Inconclusive | The window closed but there was too little signal to say anything honest — for example too few resolved tickets of that type. |
Inconclusive means “we will not guess”, not “it failed.” Reporting a delta computed from three data points would be noise dressed as a result. This is the distinction users most often misread.
Improved / regressed vs. measuring / inconclusive
These are two different axes, and they do not add up together.
- Lifecycle — measuring → measured or inconclusive. Every measurement is in exactly one of these, so they sum to the total.
- Outcome — improved or regressed. This applies only within “measured”, and is shown as a sub-line under it rather than as a peer count.
A tenant with 5 measured, 5 measuring and 2 inconclusive has 12 measurements — not 17.
Intervention subjects
Loop’s ledger is deliberately cross-domain: a measurement’s subject can be a lesson (from Muscle Memory), an acted-on suggestion, or an engagement action. That is why one ledger can answer “did our interventions help?” rather than one per app.
If the subject record is later deleted, the row reports that plainly (“the record this measured no longer exists”) instead of vanishing or rendering a broken label.
Automation vs. run
- An automation is the standing instruction: this playbook, on this schedule or this event, with these inputs.
- A run is one execution of it, with its own per-step results.
One automation produces many runs. Pausing an automation stops future runs; it does not touch past ones.
Action History vs. Run History
Both are histories, and they answer different questions:
- Action History (in Admin → AI Operations) is one action per row — everything AI completed, skipped or sent for approval.
- Run History (in Loop) is one playbook run per row, each containing steps.
They read separate records on purpose. A playbook run is an orchestration; an action is a single thing done.
Adopt
The end of the loop. A lesson whose measurement came back positive earns a path to become standard practice — it graduates into an automation rather than staying a remembered anecdote.
At a Glance
| Who it is for | Admins; managers optionally, read-only |
| Licensing | Licensed, opt-in per tenant; included with Platform Standard and Enterprise |
| Tabs | Overview, Automations, Playbook Runs, AI Outcomes, AI Suggestions, Corrections & Learning, Settings |
| Mobile | Read-only measurement ledger at /m/apps/loop |
| Sample data | None, by design — see below |
Key Features
Measurement ledger (Overview)
The front page. A lifecycle summary — measured, measuring, inconclusive — over the full ledger, then the ledger itself: one row per intervention showing what it was, what was measured, baseline, after, delta and status. Rows are paginated, newest verdicts first.
Beneath it, a summary of the loop’s own shape for this business: how many acted-on suggestions are being measured, how many active lessons exist, how many kinds of outcome have been measured here, and how many improvements are ready to adopt. Every number is scoped to your business, not a platform total.
Automations
Author what runs unattended. Pick a playbook, then either schedule it or bind it to an event, set the input values it runs with, and require approval before its risky steps execute if you want a human in the loop. Automations can be paused and resumed without deleting them, run on demand without waiting for a schedule, and installed from a system template in a paused state so you can review before enabling.
Playbook Runs
Every execution with its per-step results and timings. Filter by status, playbook or date range. A failed run can be retried — steps that already succeeded are skipped rather than repeated — and a stuck run can be cancelled. Run history exports to CSV for compliance review, and the page rolls up which tools have been failing most often recently.
AI Outcomes
Business results the AI delivered over a date range you choose, with the evidence behind them — the fuller counterpart to the fixed completion snapshot shown in Admin → AI Operations → Trust & Safety.
AI Suggestions
Administration for the proactive suggestions shown across apps: per-rule thresholds, enable and disable, telemetry on how often each rule was shown and acted on, plus review of AI-proposed new rules.
Corrections & Learning
What agents learned from feedback and failed actions — each correction with its confidence, how often it has been used, and its success rate.
User Roles & Permissions
| Who | Access |
|---|---|
| Business admins | Full access to every tab |
| Loop app-admins | Full access (granted per app, without full business-admin rights) |
| Managers | No access by default. With Allow managers to view the Loop overview enabled, they get the read-only overview; the interior tabs remain admin-only |
| Members | No access |
Because the interior tabs keep their own admin gates, a manager granted overview access sees the ledger and the pipeline summary and is not offered doors that would bounce them.
Settings Reference
| Setting | Default | What it does |
|---|---|---|
| Allow managers to view the Loop overview | Off | Extends the read-only overview to managers. The interior tabs stay admin-only regardless. |
Sample Data
Loop ships no sample data, deliberately. Its ledger is a measurement surface, and fabricated before/after deltas are exactly the kind of invented number the platform’s outcome metrics exist to prevent. To see the ledger populate, load sample data in an app that feeds the loop — Mango Signal or Muscle Memory — and measurements flow in from there.
How Loop Relates to Other Apps
- Muscle Memory — supplies the lessons. Muscle Memory is what the business remembers; Loop is how it learns. A lesson appears in Loop’s ledger only once it intervened and was measured.
- Admin → AI Operations — runs the AI. Approvals, pausing, agent health and the kill switch live there permanently.
- Ask AI and the app agents — generate the suggestions and corrections Loop administers.
When Loop is enabled, the admin URLs for the surfaces it absorbed redirect into the app, so each has exactly one home and existing links and bookmarks keep working. When Loop is disabled they render in admin as before.
Getting Started
- Enable Loop from the Apps Marketplace (admin).
- Open Loop → Overview. On a new tenant the ledger is empty — that is expected; measurements appear as the loop intervenes.
- Visit Automations to schedule a playbook, or install a suggested one in its paused state and review it first.
- Check back after a measurement window closes to see the first verdicts.
- If your managers should see the receipts, turn on manager access in Settings.
Best Practices
- Read inconclusive as a prompt, not a failure. It usually means the window was too short or the volume too low for that metric.
- Let a window finish. Deltas mid-window are not yet meaningful; the row tells you how long it runs.
- Adopt what measured well. A positive delta is the signal to turn a lesson into a standing automation instead of relying on people remembering it.
- Keep the ledger honest. There is no way to hand-write a measurement, and that is the point — resist the urge to treat the absence of a verdict as a bad one.