More Information

Outcome loops in eTOM and Autonomous Networks terms

Autonomy is earned per action type and assessed, never claimed

Exosphere Agentic aligns every outcome loop to the industry's process map, runs it as a closed loop, and widens its autonomy one action type at a time, on evidence, with the operator making every move. Exosphere Agentic runs on Azure and integrates with the Microsoft AI platform, deployed into the customer's own tenant — so context, data and governance stay where they belong.

Updated 11 October 2026

Every loop sits on the industry's process map

eTOM, TM Forum's Business Process Framework (GB921), is the industry-agreed view of what a service provider does. In GB991 terms, its Operations area covers Fulfillment, Assurance and Billing, with Operations Readiness and Support shared with Strategy-to-Readiness. The end-to-end flows in GB921E, a non-normative application note, give each Exosphere Agentic loop a recognisable home.

eTOM end-to-end flow	Example loop	Open APIs in the path
Order-to-Payment	Enterprise quotes	TMF620 Product Catalog, TMF622 Product Ordering Management
Production Order-to-Acceptance	Order activation	TMF641 Service Ordering Management, TMF638 Service Inventory, TMF639 Resource Inventory
Request-to-Answer, Complaint-to-Solution, Problem-to-Solution	Bill explanations	TMF629 Customer, TMF637 Product Inventory, TMF656 Service Problem
Trouble Ticket-to-Solution	Proactive service-impact care	TMF656 records service impact, TMF621 drives resource repair, TMF724 gives the ITIL view across both
Usage-to-Payment	Bill-shock prevention	TMF635 Product Usage, TMF678 Customer Bill

Loops are aligned to these flows. They are examples, composed per operator against the existing estate. Start with one loop. There are many more. The right loop to start with is the one attached to the outcome you most need to improve.

OODA: the closed loop the industry chose, with a person at decide

TM Forum's closed-loop anomaly work compared several patterns and chose Observe, Orient, Decide, Act as the generic closed loop, with fast loops and slow loops for different decisions. eTOM's anomaly-management processes sit in the same stages. Exosphere Agentic runs each loop that way:

Observe. Signals arrive: alarms, KPIs, service problems, tickets, orders and bill events.
Orient. The telecom ontology answers who is affected, what they hold and what was promised to them.
Decide. The agent proposes with evidence. Policy decides: execute, route to the role with authority, or block.
Act. A governed call through TM Forum Open APIs, written once.
Confirm. The real outcome is recorded as applied, refused or unconfirmed, and feeds the next cycle.

In Exos terms: detect, understand, decide, act, confirm. The human sits at decide. Each answer follows the Sherpa envelope: Facts, Assessment, Recommendation, Action, Audit.

SID already gives the loop a place in the model. The Closed Loop ABE in the GB922 Patterns domain carries the loop with its goals, policy and responsible party role, so governance is modelled as well as enforced.

A closed loop with policy at decide

flowchart TB
    OBS["Observe<br/>alarms · KPIs · tickets · bill events"]
    ORI["Orient<br/>ontology: who is affected<br/>and what was promised"]
    DEC["Decide<br/>agent proposes with evidence"]
    GATE{"Inside delegated<br/>policy?"}
    HUM["Role with authority<br/>decides"]
    BLK["Blocked<br/>recorded with reason"]
    ACT["Act<br/>governed Open API call<br/>written once"]
    AUD["Confirm and audit<br/>applied · refused · unconfirmed"]
    OBS --> ORI
    ORI --> DEC
    DEC --> GATE
    GATE -->|"execute"| ACT
    GATE -->|"route"| HUM
    GATE -->|"block"| BLK
    HUM -->|"approved"| ACT
    HUM -->|"declined"| AUD
    BLK --> AUD
    ACT --> AUD
    AUD -->|"outcome feeds the next cycle"| OBS
    classDef exos stroke:#0077B6,stroke-width:2px
    classDef human stroke:#009688,stroke-width:3px
    classDef stop stroke:#E68A00,stroke-width:2px
    class OBS,ORI,DEC,GATE,ACT,AUD exos
    class HUM human
    class BLK stop

Observe, orient, decide, act: policy executes, routes or blocks, and every outcome is confirmed and audited.

Earning autonomy, one action type at a time

Autonomy in Exosphere Agentic is set per action type, not per agent or per model. A goodwill credit and a plan change can sit on different rungs of the same loop.

Rung	What happens	Autonomous Networks level, by analogy
Always approve	Every case goes to a person with authority	L2 Partial autonomy
Approve above a threshold	Policy executes routine cases; a person approves above the threshold	L3 Conditional autonomy
Execute within policy, sampled	Policy executes; people review a sample	L4 High autonomy
Autonomous within policy	Executes inside the delegated policy; intent stays shared with people	L4 High autonomy

What moves a rung

Promote when disagreement between agent and approver is low, the reasons are understood, and sampled review agrees.
Demote when disagreement rises, a new reason appears, a policy changes, or an incident or reversal occurs.

Nothing promotes itself; the operator makes every move. The ladder stops short of full autonomy by design, so people can always constrain, correct or stop a loop.

The earning-autonomy ladder

flowchart BT
    subgraph LAD["ONE ACTION TYPE · E.G. A GOODWILL CREDIT"]
        direction BT
        R1["Always approve<br/>by analogy · AN L2"]
        R2["Approve above threshold<br/>by analogy · AN L3"]
        R3["Execute within policy<br/>sampled review<br/>by analogy · AN L4"]
        R4["Autonomous within policy<br/>intent still shared<br/>by analogy · AN L4"]
        R1 -->|"promote"| R2
        R2 -->|"promote"| R3
        R3 -->|"promote"| R4
        R4 -.->|"demote"| R3
        R3 -.->|"demote"| R2
        R2 -.->|"demote"| R1
    end
    EV["Evidence<br/>disagreement rate and reasons<br/>sampled review"]
    OP["Operator<br/>approves every move"]
    EV --> OP
    OP -.->|"promotes and demotes"| LAD
    classDef exos stroke:#0077B6,stroke-width:2px
    classDef human stroke:#009688,stroke-width:3px
    class R1,R2,OP human
    class R3,R4,EV exos

Each action type climbs on evidence and drops on a new pattern; the AN levels are an analogy, not an assessed level.

Autonomous Networks levels: assessed, not claimed

TM Forum's Autonomous Networks programme gives the industry a shared yardstick, and Exosphere Agentic speaks its language.

Six levels: L0 Manual, L1 Assisted, L2 Partial autonomy, L3 Conditional autonomy, L4 High autonomy, L5 Full autonomy.
IG1252, the Autonomous Network Levels Evaluation Methodology, defines the six-level model and how a network, or part of one, is evaluated task by task.
IG1218 sets out the business requirements and framework: autonomous domains related to business, service and resource operations, connected by closed loops.
Assessment instruments: scenario-specific evaluation questionnaires (ANLET) and independent validation (ANLAV).
The journey: pick a high-value scenario, evaluate it, validate it, implement with solution packs, then contribute back.

As the level rises, execution, awareness, analysis and then decision pass from people to systems, and L4 is where systems begin deciding independently. That is why the step from L2 to L4 is a semantics problem before it is an AI problem: a system decides well only when it knows who is affected and what was promised. The telecom ontology for BSS and OSS supplies that meaning.

Exosphere Agentic maps each rung to the level vocabulary by analogy, so an AN programme can read it at a glance. The level itself comes from the operator's own assessment under IG1252, and an Exosphere Agentic loop fits straight in as a high-value scenario.

Exos Systems is a TM Forum member and aligns to ODA. Standards bodies named here describe the context we work in; they are not an endorsement by those bodies.

Take the next step

Pick one high-value scenario with us and see where each of its action types starts on the ladder.