cs.AIMar 16, 2026
SaveFrom Workflow Automation to Capability Closure: A Formal Framework for Safe and Revenue-Aware Customer Service AI
Organizations: Minerva CQ, 114 Lester Ln, Los Gatos, CA 95032
Abstract
Customer service automation is undergoing a structural transformation. The dominant paradigm is shifting from scripted chatbots and single-agent responders toward networks of specialised AI agents that compose capabilities dynamically across billing, service provision, payments, and fulfilment. This shift introduces a safety gap that no current platform has closed: two agents individually verified as safe can, when combined, reach a forbidden goal through an emergent conjunctive dependency that neither possesses alone.
Figures & tables
| ID | Capability | Description |
|---|---|---|
| IntentClassify | Classify customer intent from natural language | |
| CustomerLookup | Retrieve customer record, tier, and history | |
| BillingRead | Read itemised billing history and current balance | |
| DisputeLog | Log a billing dispute and open a case | |
| CreditEligibility | Check eligibility for a billing credit | |
| CreditApply | Apply a billing credit to the account |
Table 1: Telco capability set ( ). Forbidden: (PaymentModify), (BulkAccountExport).
| Arc | Tail Head | Semantics |
|---|---|---|
| Intent classification enables billing and service reads | ||
| Identity-confirmed intent enables payment read | ||
| Billing read + credit eligibility credit application | ||
| Catalogue + eligibility service provision | ||
| Customer lookup enables downstream checks | ||
| [UNSAFE] Billing + payment read bulk export |
Table 2: Telco capability hyperedges. Arc is the unsafe hyperedge.
| Event | Closure update | Safety check? | Goal set effect | Telco cost |
|---|---|---|---|---|
| Agent join | from | Required | Can only grow | |
| Agent leave | , deferrable | Not required | Can only shrink | |
| Cap. gain (lazy, ) | Not required | Unchanged | ||
| Cap. gain (active, ) | Required | Grows by | ||
| Cap. loss | Not required | Shrinks by |
Table 3: Agent and capability event costs and safety re-check requirements. Telco values: , , , .
| Failure category | Annual cost | Derivation basis |
|---|---|---|
| AND-violation correction | $14.6M | Proved (Theorem 6.3, Spera 2026 ); rate from Fini Labs Research (2025) |
| Workflow ambiguity excess handling | $9.9M | Proved; rate from Fini Labs Research (2025) |
| Compliance documentation overhead | 2.0M | Industry estimate |
| Expected breach cost (tail risk) | $15.6M/incident | IBM 2025; GDPR fine schedule |
| Operational total (excl. breach) | 26.5M | Recurring annual |
Table 4: Annual cost of failure under workflow-based CS automation (10M transactions/year).
| Line item | Conservative | Moderate | Validation status |
| Operational cost reduction | $3.81M | $7.62M | Proved (Thm. 6.3) |
| Compliance documentation avoidance | $0.50M | $2.00M | Industry estimate |
| Emergent upsell discovery | $0.40M | $0.61M | Projection; A/B testable |
| Churn reduction via NMF | $12.24M | $12.24M | Projection; pilot needed |
| Compliance-enabled expansion | $5.25M | $7.50M | Projection |
| Total annual value | $22.20M | $29.97M |
Table 5: Consolidated annual business case (10M transactions/year, conditional projections).
| Assumption | Base | Stressed | Relaxed | NAV range |
| NMF identification accuracy | 85% | 60% | 95% | 22.3M |
| Churn retention conversion | 15% | 8% | 22% | 23.0M |
| Automation gain (pp) | 5pp | 2pp | 10pp | 24.5M |
| Upsell conversion uplift | 28% | 15% | 40% | 20.9M |
| Sessions at churn risk | 5% | 2% | 10% | 24.2M |
| Compliance expansion conv. | 17.5% | 10% | 25% | 22.3M |
Table 6: Sensitivity analysis: net annual value under assumption stress. Base: $20.7M.