AI is probabilistic. Telecom billing cannot be.

The rise of the “agentic enterprise” promises automation at scale. But while AI thrives on probabilities, telecom revenue management demands certainty of outcome. Dominic Smith examines why charging and billing cannot be probabilistic – and why architectural discipline matters more than ever.

Artificial intelligence is rapidly becoming the orchestration layer of the enterprise.

AI agents can draft contracts, negotiate renewals, recommend bundles, generate marketing campaigns and even execute transactions across multiple systems via MCP integrations. The promise of the “agentic enterprise” is compelling – software that not only responds, but acts.

But in the telecoms industry, there is one area where this narrative needs careful examination: revenue management and monetisation.

Because while generative AI is probabilistic by design, charging and billing demand certainty of outcome, not probability of outcome.

The tension at the heart of AI adoption

Large language models do not calculate outcomes in the traditional sense. They predict the most likely next word or token based on statistical patterns. That makes them powerful for use cases such as content generation, summarisation, recommendation and advisory workflows.

It also means they are inherently non-deterministic: the same input can produce slightly different outputs; the model can refine its reasoning mid-flow and answers are contextual rather than fixed.

In most enterprise contexts, this is acceptable. If an AI agent drafts slightly different wording in a contract summary or marketing email, no harm is done. But billing is not a game of Numberwang, where random numbers are arbitrarily declared correct.

Financial systems cannot operate on vibes, probabilities or shifting interpretations. They require fixed rules and reproducible outcomes.

A bill must be the same every time based on the same inputs. It must be explainable. It must be auditable. It must stand up to regulatory scrutiny.

Revenue management systems are not advisory systems – they are financial truth engines.

Why billing is more than invoicing

For Communications Services Providers (CSPs), the process of “billing” is not simply invoice generation. It is the culmination of:

• Online charging
• Usage mediation
• Complex rating logic
• Discount hierarchies
• Taxation rules
• Bundling dependencies
• Contractual obligations
• Revenue recognition constraints

CSPs process millions of chargeable events every day. Small inaccuracies compound very quickly, and a minor inconsistency can lead to material financial exposure.

Regulators around the world have imposed significant penalties for billing inaccuracies. There is no tolerance for “close enough”.

This is why charging and billing engines are deterministic by architecture. The same charging data record and the same pricing rules must produce the same rated outcome every time. Reproducibility is not a feature. It is a requirement.

Where AI does add value

This doesn’t mean that AI has no role in telecom billing. It just means the role must be carefully defined.

AI can:

• Assist with configuration
• Explain complex bills
• Analyse anomalies
• Support dispute handling
• Recommend optimisations
• Automate operational workflows

It can accelerate human productivity around the revenue engine, but it cannot replace the deterministic core.

In practical terms, AI may assist in configuring offers, analysing revenue data or identifying anomalies, but the charging engine must execute rating logic deterministically and without interpretation. AI can draft a customer communication, but the underlying financial numbers must come from a governed system of record.

The real risk is architectural, not technical

Any suggestion that AI could replace billing systems is misplaced. The more realistic shift is architectural.

As AI agents become more capable, they increasingly orchestrate workflows across multiple enterprise systems, BSS/OSS included. The issue is not replacement but interaction: how AI-driven orchestration interfaces with revenue-critical systems.

In this environment, governance becomes paramount. Access to monetisation capabilities must be tightly controlled.

If agent-driven workflows are not properly constrained, probabilistic systems may begin influencing deterministic processes without sufficient guardrails. For example, if an AI agent dynamically alters offer parameters, discount thresholds or customer entitlements without strict policy enforcement, the integrity of downstream charging and billing can be compromised.

That is where architectural discipline matters. AI can orchestrate and accelerate, but the deterministic revenue core must remain governed, controlled and insulated from interpretation.

Determinism as a competitive advantage

There is a tendency to frame determinism as legacy and AI as inherently modern. However, in revenue management, the opposite may prove to be true.

As AI adoption increases, the value of deterministic execution, clear policy enforcement, structured product models and auditable workflows actually rises.

CSPs have spent decades ensuring that what is sold can be provisioned and billed, and that what is billed can be reconciled and recognised. It is not the time to abandon that discipline simply because an AI agent is now initiating the transaction instead of a human.

In fact, the need for structured, governed systems becomes even more critical when automation scales.

The balanced path forward

The future of BSS is not AI replacing billing. It is AI augmenting operations around a deterministic core.

As Microsoft has recently noted, “Agentic BSS is not about replacing BSS platforms. It is about elevating them – from transaction engines to intelligent, outcome-driven systems.” The emphasis, however, should be on preserving deterministic execution at the financial core.

Revenue systems must remain:

• Predictable
• Reproducible
• Transparent
• Governed

AI agents should sit alongside them, accelerating insight and automation, but never compromising financial integrity.

In an agentic world, the systems that endure will be those that integrate intelligent automation without weakening control over financial outcomes.

Charging and billing are built on certainty of outcome, not probability of outcome. This is not a limitation – it is a design principle that will continue to underpin trust in an increasingly automated industry.