BP Readiness Dashboard
Readiness is intentionally framed as dependency transparency, not overclaiming. The workshop environment proves runtime patterns with representative adapters; production requires BP-owned onboarding for data, identity, governance, and integration.
| Domain | Status | Needed From BP | Mitigation / Start Path |
|---|
| Data | Amber | UDP, Databricks, OneData definitions, finance master data, historical transactions, audit findings | Start with read-only certified data products plus representative adapters; close source gaps through data contracts. |
| Integration | Amber | ECC, S/4, CFIN, Ariba, BlackLine, ServiceNow, Treasury, Identity | Use event/API adapter contracts first; production wiring swaps representative adapters for BP-approved endpoints. |
| Governance | Green/Amber | AI Governance, Cyber, Identity, Data Governance, Control Owners | Agent registry, policy runtime, human approval, evidence packs, replay, and TDR readiness are visible now. |
| Process | Amber | SMEs, process owners, control owners, exception handlers | Use operating-model RACI and current-state workshops to define ownership and intervention paths. |
| Identity | Amber | Entra app registrations, service principals, SCIM/JIT, conditional access, audit sink | Identity simulation demonstrates the pattern; BP tenant onboarding remains required. |
Agent Dependency Matrix
| Agent | Data Required | Integration Path | Real-Time Need | Action Boundary |
|---|
| Duplicate Invoice Agent | SAP ECC/S4 invoices, Ariba invoices, vendor master, PO/GR, payment history | SAP MCP, Ariba MCP, Evidence MCP | Event-driven for new invoices | PaymentHold.Draft; no autonomous payment release |
| Journal Risk Agent | Universal Journal, BKPF/BSEG/ACDOCA, BlackLine close tasks, approval history | SAP MCP, BlackLine MCP | Batch/event hybrid | Draft escalation; no direct posting |
| Credit Decision Agent | AR exposure, credit limits, ratings, commodity signals, payment behavior | SAP AR, Treasury, Market Feed MCP | Near real time | Recommend limit action; human approval required |
| Treasury Liquidity Agent | Cash positions, FX exposure, forecasts, covenants, sanctions | Treasury MCP, FX Feed MCP | Near real time | Recommend hedge/liquidity action; no transfer execution |
| Forecast Intelligence Agent | Databricks gold datasets, market feeds, production, retail, project data | Databricks MCP, Market MCP | Scenario/on demand | Generate forecast and narrative; finance owner approves |
AI Necessity Map
| Use Case | Class | Why | Governance |
|---|
| Invoice 3-way match | Rules first | Deterministic match logic is cheaper, explainable, and auditable. | Only use AI for exception narrative or ambiguous evidence extraction. |
| Journal anomaly detection | ML + policy + agent | Statistical anomaly flags plus policy/materiality rules; agent gathers evidence and routes. | Human controller approves action. |
| Forecast scenario narrative | LLM assisted | Forecast math should be deterministic/model based; LLM explains drivers and options. | Narrative must cite assumptions and data products. |
| Root-cause investigation | Agentic | Requires tool calls across context, evidence, policy, history, and ownership. | Evidence contract blocks unsupported recommendations. |
| SLA routing | Rules first | Queue routing and escalation should remain deterministic. | Agent can summarize context, not bypass rules. |
Operating Model / RACI
| Capability | Responsible | Accountable | Consulted | Informed |
|---|
| Policies and controls | BP control owner | BP business/control owner | Service provider, AI governance | Audit, cyber, finance leadership |
| Agent implementation | Service provider | Joint product owner | BP SMEs, cyber, data owners | Process teams |
| Runtime operations | Service provider/platform ops | Joint platform owner | BP technology, cyber | Business owners |
| Data quality remediation | BP data steward | BP data owner | Service provider analytics | Process owners |
| Value realization | Joint team | Finance transformation lead | Process owners, service provider | CFO/FB&T leadership |
Risk Register
| Risk | Severity | Mitigation |
|---|
| Poor data quality | High | Data trust scoring, source lineage, missing-field detection, low-confidence action block |
| Missing APIs/events | Medium | Adapter contracts and event abstraction; start with Databricks/UDP extracts where APIs lag |
| Identity delays | High | Early Entra onboarding; representative identity simulation until BP tenant consent is complete |
| Model hallucination | High | Evidence contract, policy gate, confidence threshold, human approval |
| Agent failure | Medium | Runtime observability, graceful degradation, human workflow fallback, replay-assisted recovery |
| Cost escalation | Medium | Rules-first routing, retrieval before generation, caching, model routing, cost dashboard |
| Change resistance | Medium | Parallel run, role evolution, RACI, training and hypercare readiness |
Value Commitment Map
| Area | Workshop Commitment | How bp Sphere Proves It |
|---|
| Journal review | Reduce investigation effort by 40-50% | Continuous close mission; anomaly detection + evidence packs + controller approval |
| Invoice validation | 70%+ touchless target for bounded scope | P2P mission; 2/3-way match, duplicate control, Ariba/SAP evidence |
| Decision traceability | 100% for governed decisions | Decision lineage, evidence pack, replay, policy result, human approval |
| Control coverage | Move from sample-based to event-level monitoring | Control runtime + policy registry + audit evidence |
| Agent observability | 100% named agent inventory for active estate | Agent inventory with owner, policies, SOR access, cost, and provenance |
BP question
What do we need from BP to make this real?
Layer 1 — Executive answer
Early access to governed data, integration paths, identity, governance owners, and SMEs is the success driver; AI model quality is not enough.
Layer 2 — Detailed technical answer
Start read-only with certified/curated datasets and representative adapters, then replace adapters with BP-approved APIs/events as identity and governance gates complete.
Layer 3 — Live demonstration
BP question
How does ownership work if AI moves to a service provider model?
Layer 1 — Executive answer
Technology ownership can move; accountability cannot. BP owns policy, data, controls, risk, and business outcomes.
Layer 2 — Detailed technical answer
RACI is enforced through agent owners, business owners, policy/control owners, human approval paths, and replayable audit records.
Layer 3 — Live demonstration
BP question
What data and integrations do agents need?
Layer 1 — Executive answer
Every agent has explicit source, API/event, policy, evidence, latency, and action-boundary requirements.
Layer 2 — Detailed technical answer
Agents consume canonical business events and data products rather than binding directly to SAP table semantics when possible.
Layer 3 — Live demonstration
BP question
What happens when data is inconsistent?
Layer 1 — Executive answer
We assume inconsistency exists. Low-confidence data blocks or escalates action instead of pretending certainty.
Layer 2 — Detailed technical answer
Source lineage, data quality, confidence, missing fields, conflict detection, and evidence completeness are evaluated before recommendation or action.
Layer 3 — Live demonstration
BP question
How does this handle ECC, S/4, CFIN, and Quantum changes?
Layer 1 — Executive answer
bp Sphere should not depend on a final SAP architecture decision; it consumes business events and canonical objects.
Layer 2 — Detailed technical answer
ECC/S4/CFIN adapters map to canonical events such as InvoiceReceived, JournalPosted, ReplicationFailed, and CloseRiskDetected.
Layer 3 — Live demonstration
BP question
What is bp Sphere relative to SAP, Databricks, Foundry, Yalla, and UDP?
Layer 1 — Executive answer
bp Sphere is the governed enterprise decision runtime. It does not replace BP systems of record, data platforms, or hosting strategy.
Layer 2 — Detailed technical answer
UDP/Databricks provide governed data products; bp Sphere adds context, policy, evidence, agents, decisions, human oversight, actions, replay, and learning.
Layer 3 — Live demonstration
BP question
How are agent identities managed?
Layer 1 — Executive answer
Every agent has its own digital identity; agents do not share human credentials.
Layer 2 — Detailed technical answer
Agent identity maps to roles, entitlements, policy boundaries, temporary access, audit events, and human accountability.
Layer 3 — Live demonstration
BP question
How does this align to Nexus, MCP, A2A, Azure, and AWS?
Layer 1 — Executive answer
Use BP-preferred platforms and keep the runtime portable. Register agents/tools, classify use cases, and avoid unnecessary AI.
Layer 2 — Detailed technical answer
Nexus/LaunchPad/MCP/A2A status appears in the agent registry; cloud access uses workload identity patterns and short-lived scopes.
Layer 3 — Live demonstration
BP question
How is cybersecurity handled?
Layer 1 — Executive answer
Zero trust, least privilege, identity-first access, policy enforcement, and audit-everything are the baseline.
Layer 2 — Detailed technical answer
Every action is tied to human/agent/service identity, policy result, evidence, source access, and replay record.
Layer 3 — Live demonstration
BP question
What happens when an agent, model, SAP, Databricks, or identity provider fails?
Layer 1 — Executive answer
Failure is expected; the architecture degrades safely to policy, cached context, or human workflow.
Layer 2 — Detailed technical answer
Runtime health, kill switch, model fallback, adapter failure detection, DLQ/retry, and replay-assisted recovery are visible.
Layer 3 — Live demonstration
BP question
How is AI cost controlled?
Layer 1 — Executive answer
Use deterministic logic first, retrieve before generation, cache aggressively, route to the cheapest sufficient model, and approve expensive loops.
Layer 2 — Detailed technical answer
Cost is tracked by agent/model; model routing and policy gates prevent uncontrolled autonomous spend.
Layer 3 — Live demonstration
BP question
How do we address upstream process dependencies?
Layer 1 — Executive answer
Finance complexity is often created upstream. bp Sphere exposes causality, owner, system, and recommended structural fix.
Layer 2 — Detailed technical answer
Signals link to process variants, source systems, context graph objects, policy/control failures, evidence gaps, and transformation opportunities.
Layer 3 — Live demonstration
BP question
What measurable commitments can BP track?
Layer 1 — Executive answer
Commitments should be KPI-backed: cycle time, manual effort, control coverage, traceability, cost, risk, and working capital.
Layer 2 — Detailed technical answer
Each value claim links to decisions, evidence, policy, source records, action, outcome, and replay.
Layer 3 — Live demonstration
Final 30-Second Answer
UDP and Databricks govern trusted data. Nexus, LaunchPad, MCP/A2A, identity, cyber, and TDR govern how AI is registered and controlled. bp Sphere sits above those capabilities as the enterprise decision runtime that turns governed data, policies, controls, events, evidence, and human accountability into explainable decisions, actions, replay, learning, and transformation outcomes.
Source -> Data product -> Context -> Policy -> Evidence -> Agent -> Decision -> Action -> ReplayUpdated 2026-07-19 19:22 UTC