Primary question
What must bp provide?
Realizing bp Sphere at bp
Introduction · what bp must make available to enable enterprise-scale agentic operations
Validation readiness · enterprise enablement
bp Sphere becomes real when enterprise context, data, policy, and governance are made usable by agents.
The validation question is practical: what is the minimum bp must make available so we can start building credible agents now, while the broader context, data, event, security, and governance foundations mature over time?
Back to bp Sphere intro
Open transformation command center
Open agent inventory
Open context engineering
Enablement model
Context + data + events + policies + governance
Operating principle
Trust and control before autonomy
Validation use
Answer technology, data, cyber, governance, and finance leaders
What bp must make available
This is the practical answer to the likely validation question. bp Sphere can orchestrate intelligence, but bp must provide the enterprise foundation it is allowed to reason over and act through.
| Domain | Required from bp |
|---|---|
| Business Context | Process definitions, policies, decision ownership, exception ownership, operating model, and dimensional finance meaning. |
| Data | Databricks, SAP, Ariba, Endur, Murex, trading, maintenance, documents, historical transactions. |
| Technology | APIs, events, identity, monitoring, data lineage, observability, integration gateways. |
| Governance | AI controls, approval boundaries, auditability, model/agent certification, replay expectations. |
| Security | Access management, cyber controls, data classification, secrets, encryption, residency constraints. |
| Operations | Human supervision model, escalation procedures, service ownership, runbooks, support SLAs. |
Most organizations start with agents. Successful organizations start with context.
Enterprise context: the dimensions agents need to reason like finance operators
Context is not one ontology diagram. For finance it is the combination of legal, management, process, commercial, policy, time, evidence, and value dimensions that give the same transaction its business meaning.
| Context dimension | Finance example | Why it matters |
|---|---|---|
| Legal entity | Company code, legal entity, statutory reporting unit, intercompany relationship. | Journal policy, close accountability, tax and audit treatment. |
| Management hierarchy | Business unit, function, region, asset, project, cost center, profit center. | Who owns the decision and where financial impact lands. |
| Process context | P2P, O2C, R2R, FP&A, Treasury, ST&S; process stage and exception state. | Whether the signal is a normal transaction, exception, control break, or approval boundary. |
| Commercial context | Supplier, customer, counterparty, contract, commodity, route, trading book. | Whether the same number has different risk meaning in procurement, credit, treasury, or trading. |
| Policy context | Authority matrix, SOX control, accounting policy, procurement policy, credit limit, contract clause. | Which decision path is allowed and when human approval is mandatory. |
| Temporal context | Close day, payment run, forecast cycle, month-end cutoff, SLA, market timestamp. | Why a decision is urgent now and what changes if delayed. |
| Evidence context | Invoice, PO, GR/SES, journal support, contract, approval, market feed, model run, email. | What proof supports the recommendation and what is still missing. |
| Value context | Working capital, EBITDA, cash, risk exposure, leakage avoided, effort saved. | How bp Sphere ties work reduction to measurable business impact. |
Minimum viable start: do not wait for the full enterprise foundation
Getting everything will take time. The agentic build should start with a bounded, high-value slice and expand as bp’s context, events, policies, and data products mature.
| Starting point | Minimum required | Why this is enough to begin |
|---|---|---|
| Minimum viable context | One function, one or two high-value processes, canonical object names, owner map, policy thresholds, and 12-24 months of representative history. | Lets agents reason credibly without waiting for a full enterprise ontology. |
| Minimum viable data | Curated invoice/PO/GR/payment or journal/close datasets, supplier/customer master slice, evidence documents, and approval records. | Enough to build useful P2P, R2R, credit, or FP&A agents. |
| Minimum viable events | A small set of triggers such as Invoice Blocked, PO Released, GR Posted, Journal Submitted, Forecast Changed, Credit Limit Breached. | Lets agents operate in near-real-time for selected workflows. |
| Minimum viable policy | Authority thresholds, approval matrix, SOX/control rules, evidence requirements, escalation rules. | Keeps recommendations governed from day one. |
| Minimum viable governance | Named business owner, data owner, policy owner, human approver, exception owner, and audit owner. | Prevents agents from becoming unowned automation. |
| What can come later | Full enterprise ontology, all historical systems, all regions, full event catalog, every policy document, autonomous write-back. | Do not delay agentic building while these mature. |
Recommended first build pattern: pick one workflow, one region or business unit slice, one evidence pack, one policy family, and one governed action path. Prove the full chain before scaling breadth.
Accessing SAP, Ariba, and other SORs vs using curated Databricks data
bp Sphere does not need to choose only one pattern. Direct SOR access and curated Databricks products solve different parts of the problem and should be combined deliberately.
| Pattern | How it works | Best use |
|---|---|---|
| Direct SOR access | SAP / Ariba APIs, OData, CDS views, BAPIs, events, read replicas, or approved integration services. | Best for live transaction state, workflow status, payment run timing, approval state, and evidence freshness. |
| Curated Databricks access | Unity Catalog governed tables, data products, gold/silver models, lineage, quality scores, historical snapshots. | Best for history, analytics, feature generation, trend detection, simulations, and cross-domain joins. |
| Hybrid pattern | Use Databricks for curated history and feature context; use SAP/Ariba APIs/events for current state and action readiness. | Recommended starting architecture because it is credible and practical. |
| Evidence access | Documents from contract repositories, SharePoint, invoice images, approval memos, emails, and policy stores. | Agents need source proof, not only structured tables. |
| Write-back posture | Read first; draft actions second; write back only through approved workflow/API after human approval. | Reduces cyber, control, and audit risk while pilots mature. |
The five pillars required for enterprise-scale agentic operations
Pillar 1
Enterprise Context
- Business glossary
- Process hierarchy
- Organization hierarchy
- Cost centers
- Asset, supplier, counterparty, contract, trading, and finance hierarchies
- Minimum viable context for first agents
Pillar 2
Enterprise Data
- Databricks and data products
- SAP and Ariba data
- Workday, Endur, Murex, ServiceNow
- Document repositories
- Historical transactions
Pillar 3
Enterprise Events
- Invoice Created
- PO Released
- Goods Receipt Posted
- Journal Submitted
- Counterparty Updated
- Contract Changed
- Maintenance Work Order Created
- Price Exposure Changed
Pillar 4
Enterprise Policies
- Approval policies
- Authority matrices
- Credit and accounting policies
- Procurement and contract policies
- Risk, trading, and SOX controls
Pillar 5
Enterprise Governance
- Decision ownership
- Escalation ownership
- Exception ownership
- Risk and audit ownership
- Human supervision model
Enterprise foundation required before safe autonomy
Business Context
↓
Enterprise Ontology
↓
Enterprise Data
↓
Enterprise Events
↓
Enterprise Policies
↓
Agent Runtime
↓
Human Supervision
↓
Business Outcomes
Databricks integration model
bp provides
Enterprise intelligence foundation
- Unity Catalog
- Governance metadata
- Curated business datasets
- Historical transactions
- Data lineage
- Data quality scores
bp Sphere provides
Decision operating layer
- Agent orchestration
- Context resolution
- Decision intelligence
- Simulation
- Evidence generation
- Human supervision
SAP · Ariba · Endur · Murex · Workday · Documents
↓
Databricks
↓
bp Sphere
↓
Agents
↓
Users
Data quality requirements
Most AI failures are data quality failures. bp Sphere can expose data quality issues; bp must own remediation.
| Area | Examples |
|---|---|
| Completeness | Missing supplier IDs, incomplete cost-center mapping, missing approval owners. |
| Accuracy | Incorrect coding, stale master data, incorrect amount or currency normalization. |
| Consistency | Different definitions for supplier, contract, exposure, exception, or forecast driver. |
| Timeliness | Delayed updates that make real-time agent signals unreliable. |
| Duplication | Multiple supplier, customer, contract, or asset records for the same business object. |
| Traceability | Missing lineage between source record, evidence pack, policy check, decision, and action. |
AI governance requirements
Model Governance
Models
- Inventory
- Version management
- Approval
- Monitoring
- Drift monitoring
Agent Governance
Agents
- Registry
- Ownership
- Permissions
- Scopes
- Certification
- Retirement
Decision Governance
Decisions
- Evidence requirements
- Policy validation
- Human approvals
- Replay
- Auditability
Learning Governance
Learning
- What can learn
- What cannot learn
- Approval requirements
- Promotion controls
Cybersecurity requirements
Identity
Access
- bp SSO
- AAD / Entra ID
- Role-based access
- Least privilege
Data Security
Protection
- Encryption
- Tokenization
- PII controls
- Data residency
- Classification
Runtime Security
Execution
- Agent isolation
- Secrets management
- Credential vaults
- API authentication
- Network segmentation
Audit Security
Traceability
- Every action recorded
- Every recommendation recorded
- Every approval recorded
- Every execution recorded
- Replayable history
What bp provides vs what bp Sphere contributes
bp provides
Enterprise authority and operating substrate
- Systems
- Data
- Policies
- Governance
- Users
- Controls
bp Sphere provides
Enterprise decision layer
- Enterprise context layer
- Evidence intelligence fabric
- Decision intelligence
- Agent orchestration
- Policy enforcement
- Human supervision
- Autonomy management
- Learning system
- Replay system
- Observability
- Cross-functional reasoning
Capability maturity journey
| Stage | Operating model |
|---|---|
| Stage 1 · Read Only Agents | Discover, analyze, recommend, explain. Human executes. |
| Stage 2 · Governed Actions | Prepare, draft, validate, recommend. Human approves, system executes. |
| Stage 3 · Autonomous Operations | Execute, monitor, learn, escalate. Human supervises by exception. |
90-day bp foundation program
| Timing | Focus |
|---|---|
| Weeks 1-4 · Discovery | Process inventory, data inventory, policy inventory, agent inventory, system inventory. |
| Weeks 5-8 · Context Foundation | Ontology, business glossary, agent registry, evidence framework, policy framework. |
| Weeks 9-12 · Pilot Build | P2P, R2R, Credit, Contract Validation, Maintenance Verification, Spend Intelligence. |