Rewinding to Move Forward: How Agentic AI Returns Human Intelligence to Telecom Order Management

Introduction

The order management system (OMS) is the orchestrating brain of a telecom operator. Whether it’s activating a consumer mobile subscription or provisioning a business broadband service or end-to-end (E2E) fulfilment of a complex multi-site enterprise connectivity service along with various partners’ offerings like unified communication, the OMS’s capability to transform any customer ask into operational reality is the key to defining both customer and business experience.

Over the last few decades, order management in the telecom industry has evolved through multiple eras, driven by shifting business expectations and supported by contemporary technologies. Today, we may be on the brink of another transformation, powered by Agentic AI.

This blog aims to trace the journey from the early days of manual fulfillment, through various purpose-built automations, to today’s commercial off-the-shelf (COTS) order management platforms. Ultimately, it outlines an ambition for how autonomous AI agents can restore humans’ intelligence, adaptability, and collaboration while performing and scaling at the machine level.

Era of the 1980s - Manual Fulfilment, Driven by SMEs

In the initial years of telecommunications operations, Business Support Systems (BSS) and Operational Support Systems (OSS) existed in silos. CRM, inventory management, provisioning system, billing system, workforce management, and logistics operated independently and were managed by their respective Subject Matter Experts (SMEs). When a customer order was raised, the process was mostly manual.

• A sales representative entered the order into the order entry system (or CRM).
• A fulfilment SME checked resource availability, such as ports, IP addresses, and licences, by logging into the inventory UIs or, sometimes, even in spreadsheets.
• If hardware/device/CPE was required, someone entered the request into the workforce management tool, requesting a shipment and a technician’s visit.
• Provisioning staff manually configured network elements using provisioning systems screens or command-line access.
• A billing representative created the account, mapped the right tariff, and ensured invoicing was triggered.

This approach was slow, error-prone, and expensive - but it worked, as humans brought intelligence, adaptability, and collaboration to address the simplest to the most complex issues. These SMEs knew their systems inside out, and if an order didn’t follow a happy path, they figured out a remediation or workaround. If something failed, provisioning staff called inventory SMEs, discussed alternatives, and activated the service. Human intelligence absorbed complexity to cover the weaknesses of the systems. For example, if a broadband installation failed because the assigned static IP range was already in use, an inventory SME could manually allocate a new range and pass it back to provisioning without holding the order for days.

This approach was resilient but unscalable. It was impractical to scale as customer bases grew and services expanded (e.g., B2C services like mobile, DSL, or B2B services like VPNs). Also, there was a huge dependency on human SMEs’ knowledge, and replacing one person with another was not easy, considering the on-the-job learning the former had gained.

Era of 1990s–2010: Flow-Through Fulfilment

Operators built their own order automation to address scale, bringing all human SMEs’ knowledge into automated workflows. These were typically in-house projects or middleware-based integrations using platforms like TIBCO, BEA WebLogic, or IBM WebSphere. Customer orders followed predefined sequences, e.g., inventory check and allocation, shipping, provisioning, and billing as per the product sold to the customer. Integrations to spoke systems were mostly point-to-point, driven by custom scripts or middleware connections.

This significantly reduced manual effort and enabled much higher throughput. However, this introduced rigidity, and if an order deviated from the predefined flow, it failed. Exception handling often dumps orders into fallout queues, requiring human intervention. Rolling out new products or services required coding new workflows, testing them extensively, and usually months of project work. For example, an L2 VPN fulfilment workflow could fail if one site had no available ports, and the workflow engine would error out, sending the order to a fallout queue, and waiting for a human to fix it.

Industry frameworks began emerging in this era. TM Forum eTOM provided process models for fulfilment, assurance, and billing. SID (Shared Information/Data model) attempted to standardise data across OSS/BSS. With all these collaborative efforts, the OMS became scalable and automation-oriented, but rigid, brittle, and costly.

Era of Today: Enabled by Commercial Off-the-Shelf (COTS) Products

During the 2000s, telcos realised that building and maintaining custom OMS was unsustainable. Vendors like Amdocs, Ericsson, Netcracker, and MetaSolv (Oracle) emerged with COTS order management products. These platforms introduced key fulfilment patterns:

• Configurable workflows instead of hardwired ones. Operators could configure flows using vendor tools and UIs.
• Catalogue-driven fulfilment decomposes customer orders into service orders, which in turn are decomposed into network-facing resource orders.
• API-based integrations, where spoke systems exposed APIs or messaging endpoints (via MQ, JMS, ESB), and the OMS consumed those.

In addition, the architectural practices like Domain Driven Design evolved, clearly defining the responsibilities of various participating systems. For example, CRM captured the order, the catalogue mapped the product, inventory mastered the resources, provisioning ensured activations, and billing ensured monetisation. The OMS itself was divided into domains of “Product”, “Service”, and “Resource”, giving rise to PSR modelling and COM, SOM, ROM-based componentisation.

Industry standardisation also matured in this era. TM Forum Frameworx (process, information, applications) finally evolved to ODA by mandating Open APIs and componentisation. Life Cycle Orchestration (LSO) from Metro Ethernet Forum (MEF) enabled APIs for the enterprise connectivity lifecycle.

Although this solved a few challenges, and the architectural patterns like RODOD and RSDOD, promoted by Oracle and TM Forum, aimed for zero-touch fulfilment, the reality on the ground is far from the vision. OMSs are still considered one of the most delicate systems to change regarding complex multi-site, multi-service, multi-vendor service offerings.

For instance, a global enterprise connectivity order (MPLS + Internet + SIP trunk + managed firewall) between two sites involves 5+ systems. Mapping all dependencies into workflows takes months to build and test, and they still fail on edge cases.

This is where we are today. Despite all these attempts and progress, one fact didn’t change - when an order fails, it still lands with a human SME.

Why Not Time Travel to the Era of the 80s - But Be Smarter.

True intelligence lies where human insight meets rational precision: Inspired by Russell & Norvig, Artificial Intelligence: A Modern Approach

If we zoom in again to the era of the 1980s, it clearly had a key advantage - humans could handle ambiguity, complexity, and exceptions. They didn’t need every scenario pre-coded; they could improvise when workflows broke, and most importantly, they could collaborate among themselves across silos without necessarily needing an intervening supervisor. The only drawback was scalability, as no operator could afford armies of SMEs handling thousands of orders manually.

What if we could replicate that intelligence and collaboration digitally, at machine scale? That’s the promise Agentic AI could bring.

Era of Agentic AI-Enabled Order Management

Agentic AI represents a paradigm shift. Instead of rigid workflows, we can deploy autonomous AI agents, software entities capable of logical reasoning, making decisions, and collaborating with other agents to achieve goals. They are digital SMEs who know the systems, can improvise, and can talk to each other. Let’s revisit the earlier example of an enterprise connectivity order between sites A and B.

Today:

• Workflows are very well defined. When an order is submitted, the process flow checks and allocates inventory, raises a shipping order for CPEs, sends a technician, ensures provisioning commands, and enables billing.
• The workflow fails if site B’s port is unavailable (due to a race condition or something). The order enters fallout. A human SME intervenes, reallocates resources, and restarts provisioning.
• Also, depending on how the order management was initially designed and has evolved over time, a new service or technology (internal or partner) introduction may mean modification of some or many fulfilment flows, especially to cater to multi-service B2B orders.

Future:

• The Order Capture Agent decomposes the CRM request into intents.
• The Inventory Agent checks availability. Site B has a port available.
• The Logistics Agent dispatches devices as needed.
• While provisioning, the Provisioning Agent gets a failure as the port was allocated to some other customer (perhaps due to an inventory sync issue).
• Instead of failing the order, the Provisioning Agent consults the Inventory Agent over an agent-to-agent (A2A) protocol, which suggests using a nearby switch.
• The Inventory Agent reallocates and informs the Provisioning Agent, who, this time, provisions the service successfully.
• The Billing Agent updates billing with the correct billing offer mappings.
• All of this happens autonomously, without rigid workflows.

Also, introducing a new technology could be as simple as introducing new agents, which is analogous to new technology SMEs in the era of the 80s.

Advantages of Agentic AI Over COTS OMS

• Adaptability: No need to predefine every workflow. Agents can reason dynamically.
• Resilience: Instead of breaking on edge cases, agents explore alternatives.
• Collaboration: Agents talk to each other, resolving conflicts (just like SMEs).
• Reduced Integration Complexity: Instead of fragile BPMN flows, agents interact flexibly with each other and with spoke systems, using modern agentic protocols like A2A or MCP.
• Continuous Learning: Agents improve over time, learning from past fallouts. And most importantly, this also helps with knowledge retention beyond human SMEs. This can make systems immune to staff turnover.

Challenges and Considerations

Agentic AI-enabled OMS, once achieved, could be just plug-and-play; however, it has a long journey to cover to reach there. Operators must address issues like:

• Trust and governance through guardrails to ensure agents act within policy.
• Agents still rely on clean and standard APIs, and industry initiatives like TM Forum ODA, Open APIs, and MEF LSO will be critical to be adopted by each component.
• Workflow engineers must remove the fear of replacement and become AI trainers and observers.
• Data quality must improve significantly; otherwise, bad knowledge could lead to bad outcomes.

Despite these challenges, there is hope, and TM Forum’s vision of Autonomous Networks and MEF’s intent-based APIs already point towards agentic orchestration.

A Phased Roadmap

Transforming order management won’t happen overnight, and a phased approach will be needed, with each phase building confidence and capability, and finally achieving a seamless transition from today's rigid OMS to fully autonomous order management.

Phase 1: AI-Assisted Workflows

In this phase, OMS can be augmented with AI agent-assisted workflows. AI agents can act as assistants within existing OMS workflows with specific focus areas. For example, Exception handling and fallout management, data validation and enrichment before handing to OMS, and gathering important learnings from fallout logs. This benefit is reduced manual effort in exception handling, which is today’s biggest pain point. For example, a provisioning failure triggers an AI agent, which checks inventory, applies a known fix, and retries without routing to a human.

Phase 2: AI Co-Pilots

In this phase, agents can become co-pilots along with OMS workflows and take more proactive roles, e.g., Generative AI (Gen AI) can interpret customer intent and dynamically configure OMS workflows, or ML agents can predict potential failures before they occur, e.g., inventory shortages. Humans can remain in the supervisory loop but rely on AI agents for most operational decisions. For example, a new enterprise L2/L3 VPN order doesn’t require a full new BPMN design as the AI co-pilot can decompose it dynamically and orchestrate across Open APIs provided by the spoke systems.

Phase 3: Fully Autonomous Orchestration

In this phase, AI agents can completely take over. Workflows may no longer be predefined, and orchestration becomes adaptive and goal-driven.  Spoke system agents like inventory, provisioning, billing, and logistics can collaborate and negotiate in real time. In addition, agents can continuously improve based on historical fallouts, service assurance data, and customer feedback. Humans can oversee governance, compliance, and exceptions beyond the AI scope. For example, an AI-driven OMS can take a global enterprise order, decompose it into sub-orders across multiple operators (via MEF LSO APIs), negotiate with partner systems, and provision the service, without human intervention.

Summary

Agentic AI is not just the next step in OMS, it’s a reimagination - from rigid workflows to intent-based fulfilment where operators define outcomes and agents figure out how. In doing so, telcos can finally realise the long-promised vision of zero-touch fulfilment, where even the most complex enterprise orders flow smoothly from quote to cash, without brittle integrations or human bottlenecks.

Ready to bring human-like intelligence back to your order management at machine scale? Connect with our experts at reachus@covalensedigital.com or complete our contact form to discover how Agentic AI can eliminate fallouts and enable zero-touch fulfilment.

Author

Deepak Bhatt, Vice President - Architecture & Solutions
Deepak is a seasoned technology leader with extensive experience in driving end-to-end BSS/OSS transformations and greenfield initiatives across the telecommunications and SaaS domains. He successfully blends deep solution architecture expertise with a strategic focus on AI-driven automation and intelligent operations, delivering scalable platforms that boost monetisation, agility, and customer experience.