Where AI actually delivers in SAP analytics: real business scenarios

Monthly close takes 12-15 business days. Commentary is manual. Variance explanations are inconsistent.

Your CFO team spends 3-4 days per month on manual variance commentary. Explanations vary in depth and tone. Junior analysts write one way, senior analysts another. Actual variance drivers are documented in spreadsheets, not tied to the official numbers. Stakeholders wait for final packs because commentary is always last. Finance doesn't trust automated numbers and insists on manual review at each step.

Smart Insights and just ask find the drivers. A custom Joule Studio agent drafts commentary. People review and approve.

Smart Insights shows the top contributors to each material variance and how the value changed over time, and the CFO's team asks follow-up questions with just ask on the governed Datasphere model—including variance questions—without waiting for a new report. A custom agent built in Joule Studio can then assemble the variance facts and draft first-pass commentary for each material item, with a human-in-the-loop step before anything is published; AI-assisted features can also summarize comment threads. In SAC, a calendar process routes the commentary to reviewers who approve or reject it, and data audit records changes to public planning data. The AI does not decide what is material—set the variance thresholds yourself.

SAC Planning + Datasphere + Joule

SAP Analytics Cloud Planning: Home for actuals, plans, and the close review cycle. Datasphere: Single source of truth for KPI definitions, dimensions, and scoped access rules. Joule: analytical insights in Joule and just ask for follow-up questions. Joule Studio: the custom commentary agent—free to design, with its runtime billed through BTPEA, CPEA, or pay-as-you-go under SAP's AI commercial model.

Example: 3 analyst-days of commentary drafting a month become 1 day of review.

Assumptions (replace with yours): commentary drafting takes 3 analyst-days a month today; AI drafts plus review cut that to 1 day; 8-hour days; a fully loaded cost of $150 an hour. Time saved: 2 days × 12 months = 24 analyst-days, or 192 hours, a year. Value: 192 × $150 = $28,800 a year of analyst time redirected to challenging assumptions. Cost side: the agent's runtime consumption plus build and governance effort; SAP's pricing page lists no price per AI Unit, so get a quote before you compute ROI. Measure: days from period end to approved commentary, rework rounds, and reviewer edits per item.

Governed financial semantic model. Standardized KPI definitions. Clean historical data.

• Datasphere financial data model is published and scoped to appropriate roles
• Actuals and plan data are normalized and free of breaks across 12+ months
• Variance tolerance thresholds are defined
• SAC Planning is live and used for monthly reviews
• Finance team has agreed on variance commentary language patterns
• AI governance process is documented and review authority is named

Manual forecasts based on intuition. Siloed by region. 15-25% average forecast error.

Your sales teams forecast demand region by region. Each region has its own seasonal patterns. Results are entered into spreadsheets and consolidated weekly. Forecast accuracy runs 15-25% error. Planning cycle is 3-4 weeks. New scenarios take days to model because they require spreadsheet rework. Supply chain waits for final demand before committing production.

Smart Predict time-series forecasts per product and region. Predictive planning writes them into a version. Planners adjust.

Smart Predict builds segmented time-series models on the demand history—one per product-region combination—breaking each series into trend, cycles, and fluctuation and using influencers such as promotions or price where you also have their future values. Predictive planning writes the forecasts into a planning version as the baseline; planners adjust it with business knowledge, and value driver trees and data actions recalculate downstream figures. A scheduled multi action retrains the models after each actuals load. If demand planning runs in SAP Integrated Business Planning rather than SAC, SAP-RPT-TS—generally available since TechEd 2026—forecasts time-indexed metrics there without custom machine learning.

SAC Planning + Datasphere + BDC for external data

SAP Analytics Cloud Planning: Demand planning interface with Smart Predict forecasting. Datasphere: Unified demand history, product dimensions, and customer hierarchy. Business Data Cloud: Optional—external signals you license (weather, market data) can be shared zero-copy through BDC Connect or loaded into Datasphere.

Example: forecast error falls from 20% to 15% MAPE—a 25% relative improvement.

Assumptions (replace with yours): today's MAPE is 20% at the planning grain; the Smart Predict baseline plus planner adjustments reaches 15%. Accuracy: (20 − 15) ÷ 20 = a 25% relative reduction in forecast error; what that is worth depends on how your safety stock and service targets respond, so model it with supply chain. Speed: the baseline is generated rather than built by hand; measure cycle time from actuals load to an approved forecast. Measure: MAPE per segment against your current method (Smart Predict reports forecast accuracy per entity), forecast bias, and planning cycle time.

History about five times the forecast horizon. Clean demand signals. Defined planning hierarchy.

• Demand history is clean and continuous, free of large unexplained breaks
• Demand is organized by a clear product-region-customer hierarchy
• Major promotions and price changes are tagged in historical data
• Smart Predict is enabled in SAC Planning and configured correctly
• Planning team has agreed on time bucket (daily, weekly, monthly)

Pipeline reviews are manual. No predictive win probability. Deal scoring is inconsistent.

Your sales teams manage deals in Salesforce, but pipeline reviews rely on manual exports and spreadsheet analysis. Deal scoring varies by rep and region. Executives spend review meetings asking "what's really likely to close?" instead of driving action. Forecast accuracy is 60-70% in Q1-Q2 but drops to 40-50% in Q4.

Salesforce data loaded into Datasphere. Smart Predict scores win probability. Drafted summaries, reviewed by people.

Datasphere loads Salesforce opportunity data on a schedule through a data flow over the SAP Integration Suite Open Connectors connection—Open Connectors supports data flows only, not remote-table federation—or SAC imports it through its Salesforce import connection. Smart Predict trains a classification model on won and lost deals and scores open opportunities; influencer contributions show which deal characteristics drive the score. SAC dashboards compare stage-based and predicted close probability, flagging deals that look better or worse than their stage suggests. A custom Joule Studio agent can draft the weekly executive pipeline summary for a sales leader to review. Pipeline reviews focus on exceptions and action, not data hunting.

SAC + Datasphere (Salesforce via Open Connectors) + Smart Predict + optional AI Core

SAP Analytics Cloud: Pipeline dashboards and deal probability visualization. Datasphere: Scheduled loads of Salesforce opportunities and reps. Smart Predict: Win-probability classification. Joule Studio (optional): An agent that drafts the weekly summary. SAP AI Core (optional): Custom win-probability models.

Example: 2.5 hours of weekly summary writing becomes 30 minutes of review.

Assumptions (replace with yours): a sales-operations analyst spends 2.5 hours a week on the executive pipeline summary; a drafted summary needs 30 minutes of review. Time saved: 2 hours × 52 weeks = 104 hours a year. Accuracy: whether predicted probabilities beat stage-based ones is the real test, so backtest the model on last year's closed deals before you trust the score. Measure: forecast accuracy (committed versus closed revenue) by quarter, slipped deals flagged in advance, and review time.

Salesforce integration. Opportunity data quality. Defined sales stages.

• Salesforce is the system of record and kept current by sales teams
• Opportunity data has clean stage, created date, and deal value fields
• Sales stages are clearly defined and followed consistently
• Historical close data (won/lost deals) is available for 12-18 months
• Open Connectors connection to Salesforce set up in Datasphere (or an SAC import connection)

Headcount planning is spreadsheet-based. No skills gap analysis. Planning is reactive, not strategic.

Your HR and finance teams build annual headcount plans in Excel. Attrition is guessed at a flat rate. Talent managers have no visibility into which roles are at risk. Skills gaps are managed in separate HR tools and never connected to planning. You find out about critical departures only after quit notices.

Attrition risk by segment. Workforce plans on governed HR data. Drafted commentary, reviewed by people.

Smart Predict trains a classification model on historical SuccessFactors data to estimate attrition risk by employee segment—role family, location, tenure band—over the next six months. SAC Planning layers growth plans on top of predicted attrition to show net headcount changes and skills gaps, and AI-assisted features can draft commentary for workforce reviews. SAP SuccessFactors People Intelligence, delivered in SAP Business Data Cloud, adds prebuilt people analytics on SuccessFactors data. Keep scores at segment level unless you have a clear legal basis: if person-level scores feed decisions about individual employees—retention offers, promotions, terminations—or monitor their behavior, the use likely falls under the EU AI Act's high-risk employment category (Annex III, point 4), with obligations from December 2, 2027, and works-council and data protection rules may also apply.

SAC Planning + Datasphere (SuccessFactors data) + Smart Predict

SAP Analytics Cloud Planning: Workforce planning interface with headcount budgets and org structure visualization. Datasphere: SuccessFactors data through remote tables or data flows (role, tenure, compensation, performance—bring in only what the model needs). Smart Predict: Attrition probability by role and segment. People Intelligence (in BDC): Prebuilt people analytics.

Example: earlier hiring for segments where attrition is forecast to exceed plan.

Assumptions (replace with yours): a 400-person engineering segment is planned at 8% annual attrition; the model forecasts 12%. Gap: 400 × (12% − 8%) = 16 additional leavers a year that the plan does not cover—about 4 a quarter to start recruiting for earlier. Validate first: backtest the model on last year's actual leavers by segment before you plan on it. Measure: forecast versus actual attrition per segment, vacancy days in critical roles, and planning cycle time.

SuccessFactors integration. Clean org structure. Defined competency framework.

• SuccessFactors kept current with employee data, tenure, role, and manager
• Historical employee data available for at least 3 years
• Org structure clean and reflects actual reporting relationships
• Critical roles and key-person dependencies identified
• SuccessFactors connection set up in Datasphere
• Legal, data protection, and works-council review completed for the use case

Supplier concentrations and single points of failure go undetected until disruption occurs.

Your procurement team manages 1,000+ suppliers across plants and regions. Risk is fragmented: supplier financial health tracked in one tool, delivery performance in another, SKU concentrations nowhere. When a key supplier goes bankrupt or a logistics hub closes, supply chain finds out reactively. Mitigation is expensive and rushed.

Late-delivery risk scored per order. Concentration made visible. Alerts reviewed by buyers.

Datasphere unifies procurement, inventory, and supplier master data. Smart Predict trains a classification model on past purchase-order lines to estimate the probability that an open order arrives late, using supplier, material, route, and lead-time features, and Smart Insights shows which suppliers drive a change in on-time delivery. Licensed third-party risk data, such as credit ratings, can be loaded alongside. SAC dashboards visualize concentration risk (% of spend on the top three suppliers, single-source SKUs, geographic concentration), and a custom Joule Studio agent can draft the weekly risk summary for buyers to review.

Datasphere + SAC + Business Data Cloud (optional)

SAP Analytics Cloud: Supply chain dashboards with risk scoring and concentration visualizations. Datasphere: Unified procurement, inventory, and supplier data. Smart Predict: Late-delivery classification. Business Data Cloud (optional): SAP-managed procurement and supply chain data products, and intelligent content such as Spend Intelligence and Supply Chain Intelligence.

Example: concentration risk quantified before a disruption, not after.

Assumptions (replace with yours): 1,000 suppliers; in one critical category the top three hold 35% of spend, and 120 SKUs are single-sourced. Exposure: those 120 SKUs and that 35% of category spend are where a supplier failure would hit first, so size safety stock or second-source effort there. Lead time: how much earlier the late-delivery model warns you is an empirical question; backtest it on last year's late orders before you rely on it. Measure: on-time delivery, share of late orders flagged in advance, and days to activate an alternate source.

Procurement data quality. Inventory tracking. Supplier master unified.

• Purchase orders and receipts tracked in ERP with complete delivery and quality data
• Inventory levels by location and SKU updated daily
• Supplier master consolidated (no duplicates for same vendor under different names)
• Spend and SKU data normalized to common GL accounts and product hierarchies
• Historical supplier performance data available for 12+ months

Customer churn detected too late. Retention reactive, not proactive. High-value customers lost.

Your sales team reacts to lost deals but doesn't predict which customers are at risk before they leave. Account managers manage hundreds of accounts and cannot spot warning signs (declining purchases, support tickets spiking, executive changes). When a big customer leaves, post-mortems reveal they were deteriorating for 6 months.

Churn risk scored per account. CRM and ERP data combined in Datasphere. Alerts reviewed by account owners.

Smart Predict, or a custom model on SAP AI Core, trains a classification model on historical customer behavior—order frequency and value, payment behavior, support interactions—against your churn definition, and scores current customers. Datasphere combines S/4HANA customer transactions (remote tables or replication flows) with CRM data loaded from Salesforce through an Open Connectors data flow. SAC dashboards surface at-risk high-value accounts, and a custom Joule Studio agent can draft a weekly at-risk summary with suggested next steps for account owners to decide on.

SAC + Datasphere (CRM & ERP data) + Smart Predict or AI Core + Joule Studio

SAP Analytics Cloud: At-risk account dashboards and retention tracking. Datasphere: S/4HANA transactions via remote tables or replication flows; Salesforce accounts via Open Connectors data flows. Joule Studio (optional): Weekly at-risk summaries. Smart Predict or AI Core: Churn probability models.

Example: a handful of retained high-value accounts can pay for the program.

Assumptions (replace with yours): 200 high-value accounts averaging $250,000 in annual gross margin; 10% churn a year; the model flags those accounts in time, and outreach saves 1 in 4 of them. Value: 200 × 10% = 20 accounts lost a year; saving 5 of them keeps 5 × $250,000 = $1.25 million of annual gross margin, before the cost of outreach and any discounts. Validate first: measure precision and lead time on last year's churned accounts. Measure: churn in the scored segment against a control group, and margin retained net of retention cost.

CRM system current. S/4HANA customer transactions tracked. Churn definition clear.

• Salesforce or other CRM is system of record for accounts and opportunity pipeline
• Opportunity close dates and amounts are reliably recorded
• Historical customer transaction and order data available for 24+ months
• Churn definition agreed (no orders in 12 months, explicit account closure, etc.)
• Datasphere connections to S/4HANA and, through Open Connectors, to Salesforce set up
• Account value segmentation defined (high, medium, low value)

Order release is manual. Planners check material, capacity, and scheduling one order at a time. Shortages surface on the shop floor.

Your production planners release dozens of production orders a day. For each one they check component availability, work-center capacity, and dates across several apps before releasing it. When a component is short, they find out late, chase a workaround, and re-plan; demand changes mid-week make it worse. Planner time goes into checking rather than into resolving the real exceptions.

Planners ask Joule to release orders. The agent checks prerequisites and proposes workarounds. Planners decide on exceptions.

In Joule, a planner asks in natural language to release a set of production orders. The Production Planning and Operations Agent validates material availability, capacity, and scheduling constraints for each order, releases those that pass if the planner wants it to, and recommends workarounds for shortages. Related agents cover production master data (SAP's Q3 2026 release highlights list a Production Master Data Readiness Agent) and material reservations. On the analytics side, SAC and Datasphere track release lead time, shortages found before versus after release, and on-time completion, so you can see whether the agent changes outcomes.

Joule + SAP S/4HANA Cloud Public Edition + Datasphere + SAC

Production Planning and Operations Agent: Order-release checks and workarounds, in Joule. SAP S/4HANA Cloud Public Edition: Production orders, inventory, capacity, and bills of materials. Datasphere and SAC: Release, shortage, and on-time KPIs. Commercials: SAP states that AI Units are required for the agent.

Example: order-release checking falls from 10 to 3 hours a week per planner.

Assumptions (replace with yours): 5 planners each spend 10 hours a week checking and releasing orders; with the agent doing the checks, 3 hours of exception handling remain; 46 working weeks a year; a fully loaded cost of $90 an hour. Time saved: 5 × (10 − 3) = 35 hours a week, or 35 × 46 = 1,610 hours a year. Value: 1,610 × $90 = $144,900 a year of planner time—set against the AI Units the agent consumes (SAP's model charges 0.02 AI Units per action for standard agents and 0.05 for advanced ones) and enablement effort. Measure: release lead time, shortages caught before release, and on-time production order completion.

SAP S/4HANA Cloud Public Edition. Clean master data. Joule and AI Units in place.

• Production runs on SAP S/4HANA Cloud Public Edition, the edition the agent works with
• Bills of materials, routings, and work-center capacities are current
• Lead times captured for all materials and suppliers
• Joule is deployed and AI Units are in the contract
• Release rules agreed: which orders the agent may release and which need a planner
• Datasphere connection to S/4HANA set up for the release KPIs