Planning Intelligence
Continuous Forecast Intelligence
Move bp from monthly and quarterly forecast cycles to a continuously updated enterprise forecast that reacts to operational, market, capital, and financial events.
Core decision question
What is most likely to happen next?
Every signal, context package, evidence pack, policy result, recommendation, human approval, action, replay, and learning artifact should serve this question.
North Star KPIs
Forecast accuracyForecast cycle reductionDecision response timeScenario coverageCapital allocation effectivenessAgent workforce
Forecast Intelligence AgentDriver Intelligence AgentScenario Intelligence AgentCommodity Intelligence AgentCapital Planning AgentNarrative Generation AgentVariance Intelligence AgentEnterprise integrations
SAP S/4HANADatabricksPlanning systemsProject systemsTrading platformsRetail systemsProduction systemsTreasury systemsFuture UDP context layerMonitored domains
RevenueEBITDACash FlowWorking CapitalCapexProductionTrading MarginRefining MarginRetail MarginLow Carbon InvestmentsReference signal ticker
Signal visibility must be more than the word signal. Each mission shows reference signal counts, severity, scenario REF-TICK ID, and current blocked/recommended state. These are scenario proof shapes until connected to live traversal.
Reference signal generated from the scenario library; REF-TICK IDs show the expected production trace shape and are not attested live events.
4747 reference signals / scenario
99 high-severity driver shifts
ScenarioScenario event REF-TICK-10421 at 14:21
33 forecast assumptions stale
Reference signals and recommended responses
| Signal | Source | Impact | Recommended response |
|---|---|---|---|
| Brent moved to $68/bbl | Commodity feed | Downstream EBITDA forecast -$180M | Run downside and capex-protection scenario |
| Refinery utilization dropped 4.2 pts | Production system | Cash flow -$74M | Recalculate margin and inventory assumptions |
| Project delay on low-carbon asset | Project system | Capex phasing +$220M | Adjust capital allocation forecast |
Demo sequence
- Change Brent crude price
- Context updates across trading, refining, production, treasury, and planning
- Forecast recalculates
- Business units impacted
- Cash flow changes
- Capital allocation actions generated
Source-system traceability for major numbers
Every major value is tied to a source system, API object, data product, or source transaction. This answers where the number came from before BP asks.
Context pack: business unit: Group Finance · cost center: CC-1042 · entity code: BP-GROUP · owner: mission runtime.
| Number / metric | Source system or data product | Record / API object | Business context |
|---|---|---|---|
| Revenue forecast -$420M | Reference Databricks finance forecast feature store shape | scenario_forecast_run=REF-FCST-2026-06-06-014 | Business unit: Customers & Products |
| Brent $68/bbl | Reference commodity price feed shape | scenario_market_event=REF-TICK-10421 | Steward: Trading Finance |
| Capex phasing +$220M | SAP PS + project system | project=WIND-H2-884 | Owner: Capital Planning |
Evidence, policy, replay, learning, and action artifacts
Decision memory
| Runtime artifact | Identifier | Credibility proof |
|---|---|---|
| Evidence pack | EP-FC-10421 | data-evidence="pack" · source records and supporting documents |
| Policy ID | FIN-FC-2026-01 | Policy runtime gate evaluated before recommendation |
| Replay ID | REPLAY-FC-10421 | Replay available from signal through human judgment and action |
| Learning | patterns learned: 4 | Decision memory updates after accepted/rejected outcome |
| Action payload | Draft hedge/capex workflow | Next action: governed execution after approval limit $500K and human override check |
Alternative actions table
Alternative actions table. This is the concrete recommendation surface: bp Sphere shows what it recommends, what it rejected, the expected value, the risk, and the action path.
| Alternative action | Confidence | Expected value / impact | Risk / tradeoff | Action path |
|---|---|---|---|---|
| Hedge downside margin exposure | 91% | +$120M EBITDA protected | Basis risk remains | Human-approved hedge workflow |
| Delay discretionary capex | 84% | +$310M cash preservation | Project delivery risk | CFO review |
| Maintain plan and monitor | 63% | No immediate disruption | $180M downside exposure | Watchlist only |
Runtime injection console
Use these controls during the workshop to show the mission changing state. Injection creates a named runtime event, updates mission KPI/readiness, adds a replay row, and opens the trace from signal to learning.
Mission KPI
Forecast accuracy
Awaiting scenario injection.
Mission readiness
Baseline
No injected blockers or opportunities yet.
Latest replay event
None
Click an injection below to create a replayable runtime event.
| Time | Event | Scenario | Decision impact |
|---|---|---|---|
| --:--:-- | Waiting | No scenario injected | No mission state change |
Decision portfolio
| Decision | Impact | Owner | Status |
|---|---|---|---|
| Hedge downside margin exposure | $120M EBITDA protected | Trading Finance Lead | Human-approved |
| Delay discretionary capex | +$310M cash preservation | Capital Planning Lead | CFO review |
| Re-phase retail promotion spend | +$35M revenue recovery | FP&A Lead | Recommended |
Vertical operating-model audit
This validates that the mission is useful from frontline work through enterprise steering: Agent Runtime → Analyst → Supervisor → Function Leader/Controller → FB&T Leader → CFO.
| Role layer | What this layer can do | Executive credibility proof |
|---|---|---|
| Agent Runtime | Forecast, driver, scenario, commodity, capital planning, narrative, and variance agents recalculate forecast state. | Signal-to-forecast trace with assumptions, drivers, confidence, and scenario coverage. |
| Analyst | Review forecast drivers, explain forecast changes, and simulate alternatives. | Can answer what changed, why, and which assumption moved the forecast. |
| Supervisor | Monitor forecast quality, compare business units, and review assumption ownership. | Can see forecast confidence, stale assumptions, and review workload. |
| Function Leader / FP&A Lead | Understand forecast confidence, scenario coverage, and forecast-to-plan variance. | Can compare base, downside, upside, and management action cases. |
| FB&T Leader | View enterprise outlook, forecast risks, capital impacts, and intervention needs. | Can identify missions and teams affecting enterprise forecast quality. |
| CFO | Simulate enterprise futures and review strategic actions. | Can see cash, earnings, capital, and risk consequences of a scenario. |
Executive credibility scenarios
These are the scenarios BP can challenge live. The mission must show signal, context, evidence, policy, decision, action, replay, learning, and enterprise impact for each.
| Scenario | Inject | Expected runtime proof | Action |
|---|---|---|---|
| Oil price shock | Brent drops 20% | Forecast, cash, business-unit impact, scenario comparison, and recommended actions update. | |
| Refinery outage | Major refinery offline | Production, margin, cash impact, and forecast revision become visible. | |
| Capital project delay | 6-month delay | Capex, cash, portfolio impact, and capital allocation recommendation update. |
12+ dimension BP credibility audit
| Dimension | Visible proof requirement | This mission's answer |
|---|---|---|
| Signal | Business event or market movement is visible and linked to a mission decision. | Brent moved to $68/bbl · Scenario event REF-TICK-10421 at 14:21 |
| Context | Entity, business unit, owner, system object, policy domain, and history are resolved. | Resolved through enterprise context graph: owner, system object, business unit, history, and policy domain. |
| Evidence | Source records, documents, approvals, and source-system snapshots are inspectable. | Evidence pack includes source system record, curated data snapshot, policy reference, and supporting document where applicable. |
| Policy | Authority, control, risk, accounting, tax, treasury, or investment policies are referenced. | Policy/runtime gate is evaluated before recommendation or action. |
| Decision | The mission produces a named decision object, not just a metric or chart. | Hedge downside margin exposure |
| Recommendation | Recommended response and alternatives are shown with impact. | Hedge downside margin exposure · confidence 91% · impact +$120M EBITDA protected |
| Human Override | Human approval, rejection, override, or escalation path is explicit. | Approval, rejection, override reason, and escalation owner are retained in replay. |
| Action | Action payload is traceable to workflow, system update, notification, hold, hedge, or approval. | Human-approved |
| Replay | Event, context, evidence, policy, recommendation, human judgment, action, and outcome can be replayed. | Replay captures signal, context, evidence, policy, agent output, human judgment, action, outcome, and learning. |
| Learning | Accepted/rejected outcomes are captured as future pattern memory. | Outcome becomes pattern memory after human decision and realized impact are known. |
| Business Impact | Financial, risk, cash, control, or value impact is quantified. | $120M EBITDA protected |
| Enterprise Integration | Systems of record, data products, APIs, and future UDP context layer are named. | SAP S/4HANA · Databricks · Planning systems · Project systems · example: Reference Databricks finance forecast feature store shape |
| Source Drill-down | Users can drill from decision to source transaction, document, or curated data product. | Source drill-down points to SOR/API record, Databricks/UDP data product, document, or market feed behind the decision. |
| Confidence | The decision carries confidence and uncertainty drivers. | Confidence is shown in decision formation and decomposed into data freshness, evidence completeness, policy clarity, and model uncertainty. |
| Alternatives | At least two alternative actions or scenarios are available. | Hedge downside margin exposure | Delay discretionary capex | Maintain plan and monitor |
| Audit Trail | Every material step is retained with timestamp, actor/agent, evidence, and policy result. | Timestamped audit trail spans agent, human, policy, evidence, and action events. |
| Executive Narrative | The mission can produce an executive answer: what happened, why, what to do, and expected outcome. | Narrative answers what happened, why, recommended action, value, risk, and next owner. |
| Mission KPI | A mission-level KPI changes when the scenario changes. | Forecast accuracy |
| Readiness | Readiness or confidence updates when blockers, policies, or evidence status change. | Mission readiness/confidence updates when signals, evidence, policies, or blockers change. |
| Enterprise Impact | The mission shows enterprise-level consequence beyond one transaction. | What is most likely to happen next? |
Runtime architecture
SignalBusiness event, market movement, policy breach, or performance variance.
ContextEntity, asset, customer, project, policy, historical pattern, and owner resolution.
Evidence + PolicySource records, documents, controls, authority, and data-quality checks.
Decision + ActionAlternatives, confidence, financial impact, owner, action payload, and approval path.
Replay + LearningFull trace, outcome, accepted/rejected recommendation, and pattern memory.