Dominic Jainy is a seasoned IT professional with a profound command of artificial intelligence, machine learning, and blockchain. With a career dedicated to navigating the intersection of emerging technologies and complex enterprise systems, he has become a leading voice on how high-stakes industries can adopt innovation without sacrificing structural integrity. In this discussion, we explore the delicate balance between the probabilistic nature of AI and the rigid, deterministic requirements of telecom revenue management, highlighting why the “human-like” intuition of AI can be a liability in the world of financial truth engines.
Billing requires exact outcomes every time, but large language models function on statistical patterns. How do you reconcile these opposing architectures, and what specific steps can ensure that an “almost correct” billing outcome never reaches a customer’s invoice? Please share an example of the potential financial fallout.
The reconciliation happens by recognizing that AI should never be the execution layer for charging; it must remain an advisory or observational layer. In telecom, billing is a financial truth engine where the same data record and pricing rule must yield an identical result every single time, whereas LLMs are designed to predict the most likely next step based on a bell curve of probability. To prevent “almost correct” outcomes, providers must maintain a hard boundary where AI can suggest a discount or identify a pattern, but the actual calculation is handled by a rules-based, deterministic engine that ignores statistical “noise.” If this boundary fails, the fallout is devastating: imagine an autonomous agent misconfiguring a discount tier for an enterprise client by just two percent. Over a three-month billing cycle, that small probabilistic “guess” could lead to a seven-figure revenue leak and a regulatory nightmare where the operator cannot explain the logic behind the error during an audit.
When autonomous agents begin orchestrating workflows like adjusting discount thresholds in real-time, where do the seams of integration become dangerous? Could you walk through a scenario where probabilistic decisions bleed into the deterministic engine and the resulting impact on customer trust and regulatory standing?
The danger lies in the “contamination” of the revenue stack, where an AI agent initiates a transaction faster than a human can review it. Consider a scenario where an agentic system dynamically modifies a customer’s data entitlement or adjusts a threshold based on what it perceives as “commercial intent” or “loyalty trends.” If that AI-driven parameter is fed directly into the charging engine without a governed policy check, the engine will faithfully execute a flawed command, leading to systemic overcharging or under-provisioning. This immediately erodes customer trust because the bill becomes unpredictable, and from a regulatory standpoint, you are left defenseless. Regulators have zero tolerance for billing inaccuracies, and if you cannot reproduce the exact logic of a transaction because a “black box” model made a judgment call, you face massive fines and a damaged reputation.
The EU AI Act labels AI in financial decision-making as high-risk, carrying penalties of up to 3% of global turnover. What governance frameworks must be in place to meet these transparency obligations, and how do you maintain an auditable trail when models cannot explain their specific logic?
To navigate the EU AI Act, operators must implement a framework that treats every AI intervention as a high-risk event requiring strict human oversight and transparency. This means you cannot allow a model to act as a final decision-maker in the revenue chain; instead, you use structured product models and governed policy enforcement to act as a “safety net.” To maintain an auditable trail, every action initiated by an AI agent must be logged against a fixed set of business rules that provide the “why” behind the “what.” Since models often cannot explain their internal weights, the governance layer must translate the AI’s output into a deterministic command that is then verified against pre-defined financial guardrails before it ever touches a customer’s account.
Some view deterministic, rules-based systems as legacy constraints, yet they remain the foundation of financial reporting. How can operators use this fixed core as leverage when scaling automation, and what metrics should they track to ensure that speed does not compromise the integrity of the revenue stack?
Determinism isn’t a legacy burden; it is actually the greatest competitive differentiator an operator has because it provides the “trust layer” for all innovation. By having a rock-solid, rules-based core, operators can experiment with high-speed automation at the edge, knowing that the underlying financial reporting remains unassailable and compliant. To ensure integrity, operators should track “reconciliation accuracy” and “outcome reproducibility” metrics, ensuring that 100% of automated transactions can be traced back to a specific, non-probabilistic rule. When speed increases through AI, these metrics act as a dashboard to ensure that the “acceleration” of the business doesn’t lead to a “crash” in the audit trail or the revenue recognition process.
If AI belongs at the edge for tasks like anomaly detection rather than at the engine level, how should a provider map out their implementation roadmap? Which high-leverage workflows should be prioritized to enhance the customer experience without risking the accuracy of the underlying charging data?
A provider’s roadmap should focus on “outside-in” implementation, starting with workflows that support the customer without altering the core financial logic. Priority should be given to AI-driven anomaly detection, which can scan millions of records to find billing errors that human teams might miss, and natural language “bill explainers” that help customers understand complex charges. Another high-leverage area is offer configuration and operational triage, where AI can speed up the design of a promotion, but the final deployment is still handled by the governed, deterministic BSS. This approach allows the operator to capture 46% of the efficiency gains predicted by industry surveys while keeping the “financial engine” protected from the unpredictable nature of generative models.
What is your forecast for agentic BSS?
I believe the future of agentic BSS will be defined by “governed acceleration,” where AI agents act as the connective tissue between disparate systems but are strictly prohibited from making final financial judgments. Within the next two years, we will see a shift where the most successful operators are those who refuse to let AI replace their billing logic, choosing instead to use it to accelerate everything surrounding that core. My forecast is that the industry will move away from the “black box” approach and toward a “glass box” architecture, where AI-driven speed is balanced by a transparent, rules-based foundation that can withstand the scrutiny of both regulators and customers alike. The real winners will be the ones who realize that in an automated world, the ability to stand behind every single cent on an invoice is the ultimate form of brand equity.