AP mission runtime Decision processing SAP remains system of record

Sphere AP Processing Mission Runtime

Transform Accounts Payable from document processing into decision processing. The runtime detects invoices, understands context, executes controls, resolves exceptions, recommends actions, learns from outcomes, and produces auditable evidence while SAP remains the source of record.

Back to architecture pack Open P2P mission Open P2P guide Run AP runtime

North star

Invoice -> Evidence -> Decisions -> Controls -> SAP

Human boundary

No autonomous payment release

Replay target

100% AP decision replay coverage

Differentiator

Evidence, policy, learning, impact, collaboration

Executable AP runtime

This panel calls the AP Processing Mission Runtime API. Deterministic invoice matching, duplicate detection, policy, evidence, value attribution, and replay are authoritative. LLM is used only for evidence-grounded explanation and drafts.

Runtime case

Not loaded yet.

Recommendation

Run the case to see the governed recommendation.

LLM assist

LLM status will appear here.

Replay

Replay hash and deterministic parity will appear here.

Runtime proof	Value
Status	Waiting for runtime call.

LLM narrative/draft will appear after the case loads. If the configured model gateway is unavailable, the runtime returns deterministic fallback text and discloses the LLM status.

Local Ariba-compatible procurement SOR

This panel reads the deployed local Ariba SOR through the same connector contract used by the platform. It proves supplier, purchase order, and invoice payloads are populated and disclosed as local SOR evidence rather than hidden sample text.

Connector status

Not loaded yet.

Suppliers

Load supplier payloads from local Ariba SOR.

Purchase orders

Load PO lifecycle records from local Ariba SOR.

Invoices

Load invoice match records from local Ariba SOR.

Credibility envelope will show source system, truth tier, fallback status, connector mode, and latency after load.

Object	Status	Value	Source detail
Waiting	Not loaded	-	Click Load local Ariba SOR.

Executive vision

This is not an invoice workflow engine. It is a Sphere mission runtime that treats every invoice as a finance decision requiring context, evidence, controls, human authority, replay, outcome capture, and learning.

Positioning shift

Traditional APInvoice -> OCR -> Workflow -> SAP

Primary optimizationDocument throughput and routing

elevate

Sphere APInvoice -> Evidence -> Decisions -> Controls -> SAP

Primary optimizationGoverned decision quality and auditability

Business objectives

Targets must be tied to governed decision outcomes, not only document automation throughput.

Objective	Target	Runtime interpretation
Touchless PO invoices	85%+	Straight-through only when evidence, match, policy, and authority gates pass.
Overall automation	65%+	Automation includes evidence preparation, routing, drafting, and controlled execution.
Duplicate payment reduction	95%	Duplicate detection is treated as a payment leakage control, not a simple fuzzy match.
AP processing cost reduction	60%	Reduce manual coordination, evidence chasing, rework, and low-value review.
Human effort reduction	50%	Humans focus on exceptions, approvals, supplier disputes, and policy overrides.
Early payment discount capture	85%+	Payment optimization balances discount capture, cash, and control posture.
Payment leakage reduction	90%	Leakage includes duplicates, overpayment, off-contract, fraud, and timing exposure.
Decision replay coverage	100%	Every recommendation, human decision, and write-back must be reconstructable.
Evidence-backed decisions	100%	No material AP action should be presented without evidence, source, and freshness.

Mission architecture and roles

The same AP runtime exposes different decision surfaces for analysts, supervisors, tower leaders, and the CFO.

Role	Focus	Required experience
Analyst	Exceptions, duplicate risks, missing evidence, approvals, supplier disputes	Clear queue, evidence pack, recommendation, action drawer, Ask Sphere support.
Supervisor	SLA, workload, escalations, policy exceptions, quality review	Team queue, escalation path, authority gates, override and learning review.
Tower Leader	Process performance, automation, leakage, working capital	AP performance cockpit, root-cause trends, value protection, automation candidates.
CFO	Cash, controls, risk, financial impact	Control confidence, leakage exposure, discount capture, replay coverage, value attribution.

AP agent fabric: core processing agents

Do not implement one generic AP agent. Use specialized agents with explicit outputs and governance boundaries.

Agent	Purpose	Output	Boundary
Invoice Discovery Agent	Discovers SAP, Ariba, email, and EDI invoices	Invoice Intake Event	Starts AP case creation without claiming source ownership.
Document Intelligence Agent	Extracts header, lines, tax, PO references, and payment terms	Evidence Package	Supports GPT-4o, Claude, local VLM, OCR, and confidence scoring.
Supplier Intelligence Agent	Enriches supplier profile, payment history, risk, contract terms	Supplier Context	Finds supplier status, disputes, bank changes, and criticality.
PO Validation Agent	Checks PO existence, status, budget availability, and tolerance	PO Validation Result	Separates PO facts from matching recommendation.
GR Validation Agent	Checks goods receipt, service entry sheet, delivery confirmation	Receipt Evidence	Identifies missing GR/SES and owner for follow-up.
Matching Agent	Performs 2-way, 3-way, and 4-way matching	Match Decision	Explains pass, tolerance breach, quantity issue, price issue, or missing evidence.
Coding Agent	Suggests GL, cost center, WBS, project	Coding Recommendation	Uses learning memory and requires human review for sensitive coding.
Compliance Agent	Checks policy, SoD, approval limits, authority matrix	Control Result	Enforces policy before recommendation or write-back.
Duplicate Invoice Detection Agent	Checks number, amount, supplier, timing, semantic similarity	Duplicate Risk Finding	Showcase control for payment leakage prevention.
Fraud Detection Agent	Checks vendor changes, bank changes, unusual patterns, split invoices	Fraud Risk Finding	Routes high-risk cases to human approval and controls.
Payment Optimization Agent	Optimizes payment timing, discounts, and working capital	Payment Recommendation	Never releases payment autonomously.

Sphere differentiator agents

These agents are the difference between AP automation and an enterprise finance decision runtime.

Agent	Purpose	Output	Why BP should care
Evidence Intelligence Agent	Builds evidence packs, proof chains, and source lineage	Audit-ready Evidence Pack	Turns AP decisions into inspectable proof.
Decision Replay Agent	Captures context, evidence, policies, agent output, human actions	Replay Record	Supports audit, recovery, and learning.
Learning Runtime Agent	Learns from human corrections, exception outcomes, coding changes, approvals	Learning Artifact	Creates organizational memory under governance.
Impact Attribution Agent	Tracks leakage avoided, discounts captured, working capital, productivity	Value Attribution	Connects AP actions to measurable business outcomes.
Policy Intelligence Agent	Converts SOPs, FIM documents, tolerances, and approval matrices into executable policy	Policy Runtime Asset	Critical BP-owned control intelligence.

Human-in-control framework

AI recommends. Humans decide. No autonomous payment release.

Gate	Trigger	Required behavior
High value	$100K authority gate	Human approval required before payment release, hold override, or supplier-sensitive action.
New vendor	Supplier not established or low confidence	Route to supplier master / controls review.
Vendor bank change	Bank details changed or unverified	Block autonomous release and require evidence.
Policy exception	Tolerance, SoD, tax, authority, or country rule breached	Escalate with policy trace.
Duplicate risk	Duplicate likelihood or prior payment candidate found	Hold recommendation; human decides.
Fraud risk	Unusual payment pattern, split invoice, supplier anomaly	Route to controls / supervisor.
SOX sensitive	Control-sensitive transaction or audit scope	Require evidence pack and replay record.

Sphere AP case model

Every invoice becomes a decision case carrying evidence, policy, AI findings, human action options, replay, and outcome capture.

Case component	Contains	Purpose
Evidence	Invoice, PO, GR/SES, contract, supplier, payment history	Every AP case starts as an evidence-backed decision package.
Policies	Relevant controls, tolerance rules, approval rules, authority gates	Policy trace is visible in the drawer and replay.
AI findings	Match status, duplicate risk, fraud score, coding recommendation	Findings show confidence, evidence, uncertainty, and alternatives.
Actions	Approve, reject, escalate, request evidence, request supplier response, open SAP, create replay	Actions are role-aware and controlled by authority.

Analyst workbench enhancements

Priority queue

Critical exceptions first: duplicate risk, fraud risk, missing evidence, SLA, and payment leakage.

Live threats

Duplicate risk, fraud risk, payment leakage, supplier disputes, and policy breaches stay visible.

Ask Sphere

Analyst can ask why blocked, show similar invoices, find policy references, draft supplier response, and research prior decisions.

Collaboration

Escalate, mention experts, attach evidence, share replay, and route owner asks through Teams, Outlook, HR hierarchy, and ServiceNow.

BP data and integration architecture

Read broadly for context; write narrowly through approved and replayed actions.

System	Read	Write	Role
SAP ECC	Yes	Limited	Current finance source of record; live validation and controlled write-back.
SAP S/4	Yes	Yes	Future strategic core; adapters should support coexistence.
SAP CFIN	Yes	Limited	Consolidated finance visibility and cross-entity context.
Ariba	Yes	Yes	Procurement, suppliers, contracts, PO evidence, and approved workflow updates.
UDP / Databricks	Yes	No	High-volume history, features, patterns, analytics, and model-ready data.
OneData / Foundry	Yes	No	Governed master data and operational data products.
ServiceNow	Yes	Yes	Evidence requests, escalations, tickets, workflow tasks.
Teams / Outlook	Yes	Yes	Drafted and governed collaboration asks, approvals, supplier communication.

Event layer

The AP runtime should support NATS or Kafka events so decisions can be triggered and reconstructed consistently.

Event	Meaning	Runtime behavior
InvoiceReceived	New invoice detected from SAP, Ariba, email, EDI, or WIM	Create AP case and evidence shell.
MatchCompleted	2-way, 3-way, or 4-way matching result available	Update case status, confidence, and next action.
DuplicateDetected	Duplicate risk crosses policy threshold	Block release recommendation; prepare hold and evidence pack.
ApprovalRequested	Authority or policy gate requires human decision	Route to owner with SLA and replay context.
PaymentReleased	Approved payment execution or release event observed	Capture outcome, value attribution, and replay closure.

Observability

Every AP agent must expose operational health, decision quality, and business impact.

Dimension	Metrics	Why it matters
Performance	Execution time, success rate, failure rate	Shows whether AP agents are production-operable.
Quality	Confidence, accuracy, override rate, human agreement	Shows whether recommendations are trusted and improving.
Business	Leakage prevented, automation rate, discounts captured, value delivered	Shows why the runtime matters to finance leadership.

AP learning system

Every human action can create a pattern, lesson, recommendation, or future automation opportunity under governance. The improvement loop is eliminate, simplify, standardize, automate, then autonomize.

Judgment capture

Human corrections, rationale, overrides, coding changes, approval decisions, and supplier dispute outcomes are captured.

Learning governance

Patterns require confidence, validation, approval, lifecycle state, rollback, and replay trace before activation.

Process improvement

The runtime identifies repeated exceptions, missing policies, poor source quality, automation candidates, and autonomization candidates.

North star roadmap

Phase	Scope	Outcome
Phase 1: Evidence + Controls	Duplicate payments, matching, approval routing	Create credible AP decision foundation.
Phase 2: Decision Runtime	Learning, replay, policy intelligence	Move from workflow support to governed decision intelligence.
Phase 3: Autonomous Exception Resolution	Supplier outreach, missing document collection, auto-remediation drafts	Automate coordination while preserving approval gates.
Phase 4: Finance Learning System	Process redesign, policy optimization, autonomous control improvements	Turn AP outcomes into compounding enterprise memory.

Key change from uploaded AP architecture

The uploaded architecture is an AP automation architecture. Sphere should elevate it into AP Decision Runtime.

Model	Flow	Implication
Standard AP automation	Invoice -> OCR -> workflow -> SAP	Optimizes document throughput but often hides decision proof.
Sphere AP runtime	Invoice -> evidence -> decision -> controls -> SAP	Optimizes governed decisions and creates replayable proof.
Strategic moat	Evidence Intelligence, Replay, Policy Intelligence, Learning, Impact Attribution, Collaboration	Differentiates from SAP, Ariba, Coupa, HighRadius, Tipalti, and hyperscaler reference patterns.

Implementation principle: AP capabilities must reuse Evidence Intelligence Fabric, Decision Replay, Policy Intelligence, Learning Runtime, Impact Attribution, Collaboration Runtime, Discovery Agent, and the continuous process improvement loop.