SAP for Utilities, presented by ASUG · San Antonio · Oct 6–9
SAP Business AI for utilities
Meter-to-cash, run by agents.
Trusted by leading utilities
Watch an agent run
A billing-exception agent, from the first implausible read to a correction a person approves.
Meter-to-cash
Implausible reads, estimated bills and EMMA/BPEM cases are classified, root-caused and resolved by policy; only true exceptions reach a billing analyst.
Target outcome40–60% fewer manual billing exceptions
Put an agent on your meter-to-cashWhy now
The building blocks are now production-grade — and most of them work on ECC.
SAP Business Data Cloud and Datasphere expose ERP, EAM and customer data as governed products with SAP semantics intact, with a Databricks lakehouse alongside for the heavy models. You can start on ECC.
Joule agents, Joule Studio and Build Process Automation let AI draft notifications, orders and postings with policy checks and approval steps. Draft-first is the default, not a compromise.
APM, drone and LiDAR capture, and vision models turn inspection into a continuous signal — the difference between a reactive and a predictive asset strategy.
What it takes to capture this: a model-agnostic stack, asset-first identity, evidence on every finding, and governance that decides who approves what.
26 use cases
SAP-native where it matters, open where it wins. Start with meter-to-cash, or pick the part of the value chain you own — every use case writes back to SAP, and 24 of the 26 run on ECC today.
Aging poles, transformers and conductors; storm and wildfire exposure; inspection backlogs that outrun crews.
Vision models read drone and LiDAR captures, match findings to the SAP equipment record, score risk with explainable factors and draft the PM notification for engineer approval.
Target outcomeDays, not weeks, from finding to notification
Also Databricks Mosaic AI model serving, specialist CV vendors, Esri GIS
Failure-risk scoring on substation assets from SCADA trends, DGA and thermal readings, load and SAP maintenance history, with a ranked inspection and replacement list.
Target outcome10–20% fewer unplanned outages on scored asset classes
Also Databricks Mosaic AI (MLflow, model serving), Azure AI Foundry
Satellite and LiDAR growth models rank spans by contact and fall-in risk; the highest-value spans become bundled vegetation work packages in SAP.
Target outcomeTreat the 15% of trees driving most vegetation-caused outages first
Also Satellite and LiDAR analytics partner, Databricks Mosaic AI
Post-event aerial imagery, OMS outages and crew locations roll up per feeder into damage counts, ETR estimates and material needs, then into SAP orders and reservations.
Target outcomeFaster, more credible ETRs; storm stock reserved before crews arrive
Also Computer vision on aerial imagery, weather feeds
Span-level ignition risk from equipment condition, vegetation, fuels and forecast wind, with mitigation options and a documented decision trail.
Target outcomeAuditable mitigation choices and targeted de-energization
Also Weather and fire-behaviour models on Databricks, Palantir AIP
A Joule-style copilot over failure history, manuals and standards proposes maintenance plans and task lists per asset class, and explains its reasoning.
Target outcomeWeeks saved per asset-class strategy review
Also Claude / GPT via Gen AI Hub, Databricks Agent Bricks
A retiring workforce, hard-to-find hotline skills, and knowledge that lives in people's heads rather than the system.
Crews ask in plain language for procedures, prior repairs on this pole, parts and safety notes; voice close-out drafts the confirmation and long text in SAP.
Target outcome20–30 minutes saved per order on lookups and close-out
Also Azure AI Foundry or AWS Bedrock agents, speech-to-text
An optimizer balances skills, hotline qualification, territory, outage windows and material availability; dispatchers approve the plan rather than build it.
Target outcome8–12% more orders completed per crew-day
Also Databricks optimizer jobs, Google OR-Tools
Language models classify incident and observation reports, spot recurring precursors by crew, task and location, and surface them before the next similar job.
Target outcomeLeading indicators instead of lagging TRIR
Also Azure OpenAI, Databricks AI/BI Genie
A Joule agent assembles the full order from a notification: task list, parts, permits, crew skills and outage coordination. Draft-first; the engineer approves.
Target outcomeNotification to order in hours
AMI data volumes, billing exceptions that need analysts, and customers who expect answers on the first contact.
Implausible reads, estimated bills and EMMA/BPEM cases are classified, root-caused and resolved by policy; only true exceptions reach a billing analyst.
Target outcome40–60% fewer manual billing exceptions
Also Databricks Agent Bricks, UiPath for legacy screens
A conversational assistant explains the bill line by line, sets payment arrangements and reports outages, grounded in the customer's IS-U data and tariff.
Target outcome15–25% lower handle time; higher first-contact resolution
Also Google Vertex AI / Gemini, Azure AI, IVR
Anomaly detection on AMI interval data, transformer-level balancing and account history flags tampering, bypass and meter faults, creating IS-U service orders with evidence.
Target outcomeRecovered revenue with a ranked, evidence-backed field list
Also Databricks Mosaic AI on AMI data, Snowflake Cortex, MDM feed
Propensity models predict arrears risk and match customers to the right assistance program, payment plan and channel before disconnect notices go out.
Target outcomeLower write-offs; fewer disconnect and reconnect cycles
Also Databricks Mosaic AI, AWS SageMaker, marketing automation
Short- and long-term demand forecasts from AMI, weather and DER adoption feed settlement, procurement and rate design as one governed data product.
Target outcomeOne forecast for planning, trading and regulatory
Also Databricks time-series models, Google Vertex AI forecasting
Storm stock and critical spares, contractor paperwork, FERC accounting, and a close that still takes too long.
Extracts and validates invoices, contractor timesheets and lien waivers, matches them to POs and service entry sheets, and posts — with exceptions routed for approval.
Target outcome70–80% touchless invoice rate
Also UiPath, Azure Document Intelligence
Cleans and de-duplicates MRO masters, then recommends storm-stock and critical-spare levels from outage history, lead times and seasonal risk.
Target outcomeLower inventory with higher critical-part availability
Also Databricks Lakehouse Monitoring, MDG data-quality rules
Reads W-9s, insurance certificates and safety credentials from PDFs and images, populates the business partner record and tracks certificate expiry.
Target outcomeDays to hours for vendor and contractor setup
Also Azure Document Intelligence
Scenario planning for grid-modernization portfolios: cost curves, contractor capacity, supply-chain lead times and rate impact modeled together.
Target outcomeDefensible multi-year plans tied to actuals
Also Databricks
Close-cockpit automation with anomaly checks on FERC-account postings, plant-in-service and depreciation, plus drafted variance commentary.
Target outcome2–3 days off the monthly close
Also Databricks anomaly models, BlackLine
More filings, more data requests, NERC CIP audits, and decisions that must be defensible in front of a commission.
Assembles EPA, state and NERC/FERC submissions from SAP source data with lineage, flags gaps, and keeps an evidence pack for every filing.
Target outcomeWeeks to days per filing cycle
Retrieval over tariffs, prior commission orders, testimony and cost data drafts responses to data requests and checks consistency across the filing.
Target outcomeFaster data-request turnaround, with cited sources
Also Claude / GPT via Gen AI Hub, Azure AI Search
Correlates OT asset inventory, patch state and access logs with NERC CIP requirements; drafts evidence and flags drift before the audit.
Target outcomeContinuous CIP readiness instead of audit fire drills
Also OT security platform, Microsoft Sentinel / Palantir AIP
Interconnection queues, renewables and storage to operate, and EV load arriving feeder by feeder.
Document AI reads solar, storage and EV-charger applications, runs the initial technical screens against GIS and load data, and routes only the cases that need engineering review.
Target outcomeInterconnection SLA met with a fraction of the manual screening
Also GIS, hosting-capacity model on Databricks
IoT-to-SAP condition monitoring for solar, wind and storage assets: underperformance detection, degradation curves and maintenance triggers.
Target outcomeHigher availability; warranty claims backed by data
Also AVEVA PI, Databricks Mosaic AI, AWS IoT
Forecasts EV charging and managed-load impact by feeder, targets program enrollment, and settles incentives through IS-U.
Target outcomeDeferred upgrades on constrained feeders
Also Databricks Mosaic AI, Google Vertex AI, DERMS
Target outcomes are typical targets, sized against each utility's own data during the readiness assessment — not guarantees.
How it works
Every use case runs on the same four layers. The last one is what makes it safe to let AI near your system of record.
AMI, SCADA and IoT, GIS, drone and LiDAR imagery, weather and work history — harmonized in SAP Business Data Cloud, Datasphere and the Databricks lakehouse.
Joule agents, Gen AI Hub and AI Core, Document AI and the HANA vector engine; Databricks Mosaic AI and Agent Bricks; Azure AI Foundry, AWS Bedrock and Google Vertex AI; plus specialist vision models.
Draft-first writes into S/4HANA Utilities (IS-U), PM/EAM, APM, Field Service Management, Ariba and Finance, through BTP Integration Suite and Build Process Automation. On ECC or S/4HANA.
Explainable scores, a full evidence trail, policy checks before any write, and approval by risk tier. Outcomes from closed orders feed retraining.
Platforms
SAP owns the transaction. The model can run wherever it performs best — and the data stays governed either way.
Owns the transaction
Joule and Joule agents, Joule Studio, Gen AI Hub and AI Core, Document AI, the HANA Cloud vector engine, SAC predictive and Build Process Automation.
Anything that has to act inside SAP — notifications, orders, billing cases, approvals and postings — and copilots grounded in SAP data.
SAP Databricks inside Business Data Cloud; Mosaic AI, Agent Bricks, AI/BI Genie, Unity Catalog
Best forHeavy ML on AMI, SCADA and imagery — asset health, non-technical loss, load forecasting, vegetation and storm models — with zero-copy access to SAP data products.
Azure AI Foundry, Azure OpenAI, AI Search, Document Intelligence, Copilot Studio, Sentinel
Best forCopilots in Microsoft 365 and Teams, contractor document extraction and OT security correlation. A natural fit for Microsoft-first utilities.
Amazon Bedrock and Bedrock Agents, SageMaker, AWS IoT, Textract
Best forIoT ingestion from renewables and substations, multi-model agents, and collections and propensity models on an AWS estate.
Vertex AI and Gemini, Vertex Agent Builder, Vertex forecasting, Contact Center AI
Best forCustomer-facing conversation and IVR, demand forecasting, and geospatial and imagery analytics.
Also in the toolbox: Snowflake Cortex AI, Palantir AIP, UiPath agentic automation, NVIDIA and specialist vision vendors for grid defects, and Esri GIS analytics. Every platform connects through SAP BTP Integration Suite, and every model call is logged with lineage and an approval step before anything is written to SAP.
Getting started
Fixed scope, tied to outcomes, on ECC or S/4HANA.
An InFocus review of data, integration and AI readiness on your landscape.
Value sized against your own AMI, EAM and IS-U data. You pick the first two.
Draft-first SAP writes, an approval workflow, a full evidence trail and the first verified outcomes.
An agent fabric, MLOps gates and managed services across value chains.
Meet us in San Antonio · October 6–9, 2026
43 agents. In production.
A working session sizes the value against your own AMI, EAM and IS-U data — so the first pilot is the one worth running.
Book a working session