Skip to content
AsiaInfo's Agentic BSS: Unifying Token Metrics, Benchmarks & Settlement for Telco AI
Industry Insights 24 min read

AsiaInfo's Agentic BSS: Unifying Token Metrics, Benchmarks & Settlement for Telco AI

Dr. Ouyang Ye outlines AsiaInfo's strategy to rebuild BSS as a token-driven agentic platform that establishes proprietary benchmarks, standardizes task specifications, and connects resource consumption, task delivery, and commercial settlement — turning telco domain experience into measurable, monetizable AI capabilities.

AsiaInfo Technology: New Tech Exploration
AsiaInfo Technology: New Tech Exploration
AsiaInfo Technology: New Tech Exploration
Context: This is the second article in the "Token Cognition Trilogy" by Dr. Ouyang Ye, CEO of AsiaInfo Technologies and IEEE Fellow. The first article proposed Token Validity (Token效度) as a unified measure for comparing models across standard tasks using full cost. This article advances from "measuring intelligence" to "monetizing intelligence" by designing a next-generation Token-Driven Agentic BSS that becomes a value exchange hub for intelligent services.

1. OpenAI's Recruitment Reveals New Competition for Professional Software Vendors

On October 6, 2026, OpenAI announced a partnership with contract-management software company Ironclad and invited professional software firms to participate in Computer Use research. They selected 11 tasks covering legal, commercial, and procurement work, each with 8–50 evaluation criteria, constructed synthetic training tasks in a software practice environment, and used reinforcement learning with practice and feedback to improve the model. Industry experience entered the model training process in the form of tasks, rules, acceptance criteria, and environments.

This signals that general models are moving deep into professional domains . Models are learning to follow business rules, operate professional software, and verify final results. Years of accumulated business experience in professional software are becoming a key source for model capability development.

The pressure on professional software vendors is obvious: as models gradually master professional knowledge, users may increasingly initiate work through Agents, with models orchestrating software execution. Competitive advantages built on software UI, operational proficiency, and implementation experience could vanish under the new capability-supply model. Every software vendor — including AsiaInfo — must answer: why should customers keep buying our products? How do we build our own moat?

Software vendors must not easily give away their industry experience. The same learning path OpenAI uses is available to us: we can use our own experience for controlled post-training, forming proprietary skills and industry execution capabilities. Base models can be continuously introduced; task definitions, training environments, evaluation methods, and training outcomes should become product assets we control. Model capabilities and professional systems together constitute deliverable industry intelligence.

In a typical BSS (Business Support System), customers, packages, orders, billing, and settlement have clear rules, and long-term operation has accumulated massive exception-handling experience. Many task results can be verified through business state and accounting records. These conditions make BSS suitable for converting industry experience into practicable, evaluable, improvable intelligent capabilities.

2. How Totogi Organizes Its Agentic BSS

Totogi serves as a valuable reference. It employs two entry modes: cloud-native BSS suite services, and adding ontology and Agent capabilities on top of operators' legacy systems. Both modes target different system conditions but share the goal of delivery speed and capability reuse.

2.1 Business System Provides Complete Commercial Chain

Totogi's Wholesale BSS covers CRM, product catalog, quote configuration, ordering, charging, billing, and partner settlement, plus AI tools for package design and recommendation. Its AWS-native, multi-tenant SaaS architecture lets multiple customers share the product base, with usage-based pricing. Product selection here involves both business functionality and delivery economics: how much existing capability can be reused when onboarding a new customer.

2.2 Legacy Transformation Relies on Executable Ontology & Business Models

Totogi's disclosed architecture includes a data layer, ontology layer, and AI layer. The data layer reads existing system structures, configurations, and processes. The ontology layer expresses entities, actions, constraints, and validity rules. Agents check proposed actions and coordinate cross-system operations. For example, a package upgrade must simultaneously check contract conditions, network resources, and billing-cycle restrictions, identifying conflicts before business submission. This is the concrete mechanism for injecting business knowledge into execution.

2.3 Business Agents Deliver Cross-System Work

In September 2026, Totogi disclosed seven new production deployments in the past year, including connecting call information, CRM, and CPQ to turn sales requests into quotes and orders, and handling order failures. These cases show Totogi entering brownfield environments with concrete business problems and expanding scope; actual effects still need customer-specific evaluation.

2.4 R&D Agents Change Capability Delivery

An AWS–Totogi joint case organizes requirements analysis, architecture design, development, code review, and testing into a multi-Agent workflow using Claude on Amazon Bedrock, with workflow managing state and feeding review/test results back to development. Business Agents handle runtime work; R&D Agents participate in software changes; both paths jointly affect customer transformation costs.

Totogi's layout now exceeds a single ontology product. It aims to let industry knowledge serve business execution, system integration, and R&D delivery, then expand reuse via a shared base. AsiaInfo already possesses CRM, Billing, SBC, AITOP and other capability foundations. The next step is to achieve complete integration on a unified task foundation: ontology and rules constrain execution; post-training guides and evaluates model ability to complete professional tasks; verification systems judge effectiveness of both; and a unified measurement standard establishes a value settlement center. All these must enter our planning.

3. Building AsiaInfo's Own Benchmark During BSS Reconstruction

The AI-Native reconstruction of BSS gives AsiaInfo a simultaneous opportunity to build both product and industry capabilities. During construction of the next-generation Token-Driven BSS, we inherently need to decompose tasks, organize rules, define acceptance criteria, and conduct testing. In this phase we should precipitate task specifications, rule configurations, execution records, and acceptance cases together. Thus, every business process advanced in reconstruction adds a set of tasks that can be practiced and evaluated. Running different models on these tasks and verifying them gradually forms our own BSS Benchmark.

3.1 From Process Language to Task Language

eTOM once provided the telecom industry with a common reference for process classification and hierarchical decomposition; its basic unit is the process. In the Agent era, the basic unit may further become the task . Past IT systems were organized around processes; future intelligent systems may be organized around tasks. Therefore, eTOM once defined the process language of telecom software; the AI era needs to redefine the task language of telecom intelligence.

On this basis we can build an internal L1–L5 task system linking business domain, process, task, subtask, and atomic action. Take package change as an example.

L1-L5 task hierarchy example for package change
L1-L5 task hierarchy example for package change

3.2 Five-Step Benchmark Construction

Map Current Systems, Build Task Map: Using eTOM as process reference, identify tasks (order processing, bill disputes, partner reconciliation) and cross-system dependencies from the customer's job-to-be-done. Internally adopt L1–L5 to associate business domain, process, verifiable tasks, subtasks, and atomic actions; this hierarchy expresses business granularity, while difficulty, risk, and autonomous execution scope are marked separately.

Unified Task Specifications & Acceptance Standards: Each task type uniformly defines inputs, environment, tools, constraints, completion criteria, KPIs, Cost, Token, and Outcome. Completion criteria confirm delivery qualification; KPIs observe quality, speed, stability; Cost accounts for full cost; Token retains each model's original measurement; Outcome is verified by actual business data. Task specifications provide repeatable training and evaluation standards. Specific task scope, authorization, budget, deadline, acceptance, and responsibility are confirmed at acceptance, forming a task contract.

Build Practice & Verification Environments: Construct normal and abnormal scenarios from authorized configurations, desensitized cases, and synthetic scenarios covering resource shortage, interface timeout, duplicate submission, billing-cycle conflicts. Verification checks final business state, necessary execution evidence, and exception handling. Example: package activation must link contract, order, resource, and service acceptance; cannot rely solely on Agent-reported completion. Deterministic checks and expert acceptance jointly provide evaluation basis; industry evaluation models assist professional result checking.

Test Multiple Models & Compute Token Validity: Under identical task, quality, and time requirements, repeatedly run different models; confirm qualified result count R; calculate total cost of ownership (TCO) within agreed scope; compute Token Validity. Failed calls, retries, tool usage, verification, human intervention, and rework all count toward TCO. Completion time, stability, and exception-handling records verify delivery conditions and analyze differences. Specially optimized Agent solutions are evaluated separately, preserving versions of model, tools, rules, configuration, and cost caliber. Each task computes validity separately; results show applicable business conditions for each solution.

Apply Evaluation Results to Production, R&D, & Operations:

Model Selection: Judge by task, difficulty, and business conditions which models meet delivery requirements, which are superior in cost or speed.

Production Routing: Convert evaluation results into actual scheduling strategies — routine work uses economic qualified solutions; complex cases escalate to higher models or human; continuous calibration with production feedback.

R&D & Post-Training: Distinguish model, rule, tool, and orchestration issues from failure logs; choose worthwhile skill or training directions; verify improvements with independent tests.

Cost Optimization: Examine gains from caching, retries, task splitting, model substitution while confirming business quality remains qualified.

This Benchmark accumulation process is exactly AsiaInfo's transformation of BSS experience into proprietary intelligent capabilities. Its uniqueness comes from the joint precipitation of business conditions, exception experience, reproducible environments, and acceptance bases. Achieving this step gives the previously proposed measurement a runnable product form: the Benchmark becomes not only a BSS test set but a record of Token Validity for key BSS tasks , continuously logging qualified outputs, full costs, and validity performance of different models and Agent solutions under identical delivery conditions. These records directly inform model selection, production routing, post-training, and cost optimization, becoming AsiaInfo's continuously accumulated industry assets serving R&D and operations.

4. Token Validity Supports Token Trading & Clearing/Settlement

After establishing measurement, intelligent service operations still face a problem: model suppliers provide resources by token or compute; Agents and tools jointly complete tasks; customers purchase services per contract. Each party's measurement, charging, and delivery responsibilities differ; someone must map them together.

The next-generation BSS should assume this role, connecting model-side resource accounts, business-side delivery records, and customer-side contract accounts. Benchmark helps select delivery solutions; production records prove actual completion; contract rules determine charging and revenue sharing. Token Validity is used for solution comparison and cost management; customer charging follows contract-agreed measurement units and pricing rules. BSS links them to simultaneously support ROI verification and commercial settlement.

Billing for intelligence: value exchange hub for internet of agents
Billing for intelligence: value exchange hub for internet of agents

AsiaInfo's existing Billing supports charging, accounting, and multi-party settlement; SBC supports multi-dimensional charging; AITOP covers token operations, model routing, and multi-party reconciliation. On these foundations, "task measurement and intelligent service clearing/settlement" can become an explicit product module with four interlocking capabilities:

Service Catalog & Ordering — Define What Customers Buy: Define deliverable tasks as services, agreeing scope, acceptance criteria, timelines, prices, responsibilities. Confirmed customer goals become service commitments accepted by the delivery side; each acceptance further specifies objects, authorization, budget, forming a task contract linked to an order. Benchmark provides basis for supply solutions; actual commitments are determined by the service operator.

Fulfillment Evidence & Task Measurement — Link Consumption & Delivery: Establish unified task identifiers linking orders, contracts, participants, resource usage, and execution records. Execution results must be cross-checked with authoritative business state, then accepted per agreement. Example: package change must confirm package effective, pricing correct, related account statuses compliant. BSS receives verified delivery records, making consumption, results, and responsibilities traceable.

Operational Accounting & Task Billing — Separate Input Efficiency & Commercial Fees: Supply side retains each model's original tokens, compute, and tool usage; customer side uses subscription, usage, qualified standard results, or combined pricing per contract. When charging by standard results, agreed acceptance results serve as billing basis. Same standard results with same agreed unit price can correspond to same customer fee despite different token consumption; underlying model calls still settle per supplier's actual rules. Linking customer revenue and full delivery cost per task lets the operator see service gross margin.

Multi-Party Clearing/Settlement & Dispute Handling — Confirm Receivables/Payables per Commercial Relationship: Confirm receivables/payables per party contracts and fulfillment records; complete fee aggregation, clearing, reconciliation, settlement. Model/tool procurement costs, partner revenue shares calculated separately; task reopening, delivery remediation, fee adjustments retain corresponding bases. Responsibilities and revenues per party per task are determined by commercial agreements.

Priority is to build Token-Driven BSS with continuous operation services. Operators define customer-facing service commitments; AsiaInfo provides task acceptance, measurement, accounting, clearing/settlement products and operational support; each party bears responsibilities per contract. By reusing task specifications, rules, and verification assets, reduce delivery costs for new customers and scenarios; extend mature capabilities to enterprise intelligent service operations.

General models can learn the industry, but having industry experience does not equal having a moat in the intelligence era. Experience must transform into learnable capabilities, verifiable delivery, and sustainable revenue. AsiaInfo's next-generation BSS must accomplish this transformation. The professional software vendor's industry position must be re-won through new products and operational results.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

BenchmarkingAgentic AIToken EconomicstelecomBSSAsiaInfoAI MonetizationBusiness Support Systems
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.