SAP Community Utility Voice: Agents on the Slides, Migrations in the Room

What SAP for Utilities 2026 actually showed us

The following Utility Voice was authored by Marc Rosson, Community Connector at Utility.Community.

On the main stage in San Antonio, PPL’s CIO and SAP’s J. Michael Sullivan walked through a Utility Canvas that put safety, grid operations, meter-to-cash and something called “Zero Ops” on a single page. It was the most complete picture of an autonomous utility we have seen a utility present.

An afternoon later, two rooms down, PPL’s own customer-technology team showed a different slide: three releases to consolidate three legacy CIS platforms onto one SAP platform, going live in August 2027, February 2028 and July 2028.

Both slides are true, and together they are the conference in miniature. The vision is autonomous. The calendar is a migration.

I went through the agenda after the fact, all 79 sessions and the 46 decks posted so far. SAP presented what is new with agents that now have dates and a commercial model for building more of them. Utilities presented their journey on migrations, with clean-core discipline that requires executive governance, and with AI they built themselves. For those of us with extensive SAP landscapes, the vision is clear from SAP. Now we need to be able to talk to the utilities that are moving forward and learn from them.

What the agenda tells you before you open a deck

ASUG’s own agenda framework tags every session with a Utility Role. The largest group, 22 sessions, carried the “Autonomous Enterprise / AI / BTP / BDC” tag. That fits the headlines.

Of those 22, about ten were about SAP’s agents and assistants. Six were about data platforms, four were utilities describing AI they built themselves, and two were about operating the platform. Across the whole agenda, the largest single block of content was not AI at all. Twenty sessions were about running the platform: RISE, upgrades, support models, cutovers, cost and security. Sixteen more were about data foundations.

Utilities are working to move to the new solutions in 2026. What follows is organized into six layers.

Layer One: The Calendar Behind Every Slide

One date sat under almost every customer story, and SAP put it on a slide: mainstream maintenance for SAP CRM 7.0 and SAP ERP 6.0 ends December 31, 2027, with optional extended maintenance to 2030. For those with a signed RISE contract, dates for some functions move to 2033.

Utilities located themselves against that line without being asked:

Nexus Water’s conversion deadline is 12/21/27.
Consumers Energy targets S/4HANA on RISE by end of 2027.
SCE’s regulatory accounting redesign goes live in July 2027.
PPL’s CIS releases run from August 2027 to July 2028.
Avista’s greenfield program runs into 2028.
PG&E’s Propel program completes its move off ECC in 2030.

The AI roadmap assumes S/4HANA. A meaningful share of the largest utilities will spend the next four years getting there.

The test they are all preparing for is billing.

Central Hudson upgraded from S/4HANA 1909 to 2023 and compared roughly 79,000 bills between old and new systems. They found a 99.6% match and 31 defects “that would otherwise have surfaced only in production.” They also moved the cutover start by 17 hours because a storm was forecast.
Exelon’s PHI team validated 31 billing cycles before go-live and held the core cutover to 47 hours.

That is the standard any agent touching meter-to-cash will be held to. Billing is where a utility proves itself, and AI-generated test cases and test data will keep making billing tests better.

Cost moved from analyst commentary to utility slides:

SMUD listed, among its concerns, “uncertainty around moving to consumption-based pricing model.”
SCE, explaining its RISE and managed-service model, named “evolving pricing around AI, Data,” fungibility across separate SAP accounts, and the need for “a second pair of eyes” on FinOps.
City of Tacoma advised peers to bring their integrator in before negotiating the bill of materials, and to use an outside advisor for the RISE contract.
SAP’s own developer-tooling session confirmed that AI Units are purchased annually and expire after 12 months if unused.

If you read The Meter Is On, this is the same story from the buyer’s side of the table.

One more calendar note, from the federal side. TVA, now live on SAP’s sovereign NS2 cloud, was candid that security review means “new features can take six months to a year to become available.” For utilities on sovereign or government clouds, the AI timeline is not just later. It is later by design.

Layer Two: Clean Core Is an Executive Decision

If you still think of clean core as a technical guideline, the customer sessions should change your mind. The utilities that talked about it treated it as a decision right held at the top.

PPL set a fit-to-standard target above 85% and ran a joint review board, with PPL, SAP and its integrator, for every enhancement. Anything outside SAP’s top extension tier needed a documented exception escalated to executive sponsors. Their phrase: “Common design was treated as an executive commitment, not a design preference.”
Salt River Project, with its owner’s engineer Validos, scored every candidate extension on customer experience, risk, cost and long-term sustainability, rolled the score into a single number, and let a weekly Design Authority Board hear challenges. The principle on their slide: “Partners advise, SAP advises… SRP makes the final decision, every time.”
Consumers Energy held net-new development objects to about 120. Fewer than five are non-compliant with clean core, and each one is visible to the steering committee.

The counter-examples make the same point from the other direction.

SCE initially selected Microsoft Dynamics 365 for field scheduling, began design, and paused “due to high level of customization required.” It landed at 91% fit-to-standard on SAP’s field products.
Dominion Energy missed three go-lives on SAP’s mobile field application. Its own words were “our customizations were killing us.” It recovered only when a partner layer turned 80% of those customizations into configuration. The program’s north star came from its distribution VP: “Do no harm to the field worker.”

Here is why this matters for AI. The decision rights utilities are exercising over custom code are the same decision rights they will need over agents. SRP’s Design Authority Board and PPL’s enhancement review board are agent-approval boards waiting to happen. Readers of The Accountability Gate will recognize the structure.

Layer Three: The Agents Have Dates Now

The most useful thing SAP brought to San Antonio was dates. The utilities AI team showed a roadmap view drawn from SAP’s published Roadmap Explorer, and for the first time the utility-specific agents carry quarters.

Status	Agents
Available now	Utilities Customer Self-Service Agent; Case Management Assistant in Service Cloud
Q4 2026	Outsorted Billing Document Resolution Agent; reliability, risk and asset-health agents in Asset Performance Management
Q2 2027	BPEM Exception Resolution Agent; Exception Resolution Orchestration Agent; Tariff Analyzer
Ideation	Billing Error Resolution; Bill Analysis; Meter Reading Analysis; every distributed-energy agent (forecasting, flexibility, congestion, portfolio, prosumer onboarding)

Field service and asset management is further along than meter-to-cash:

Dispatching and technician-briefing agents are dated Q3 2026, and crew planning Q1 2027.
Several of those are marked as available on both S/4HANA Private Cloud and ECC.
In that domain, at least, ECC customers are not shut out.

Read the footnotes.

The headline figure for the self-service agent, a 45% reduction in contact-center FTE cost, is labeled preliminary. It is modeled on a hypothetical 2.4-million-customer utility, sourced to SAP Value Management.
The claim of 94% roadmap delivery counts items “delivered or under preparation for delivery.”
I did not find a utility named as running the self-service agent in production.

None of that makes the roadmap less real. It makes it a roadmap.

The commercial model for Industry AI is now visible, too.

SAP described a ten-week proof of concept under a collaboration or test-and-evaluation agreement. A flat fee is credited against a “north star” contract, described as a value-based priced annual cloud subscription.
On SAP’s agent-to-agent diagram, “Industry AI / FDE Agents,” built by forward-deployed engineers, sit alongside SAP’s packaged agents and customers’ custom agents.
The productization path on SAP’s slide is explicit: the first engagement’s proof of concept becomes “additional domain use cases from other customers delivered in productive systems.”

What was not on a slide, but was the most consequential news in the hallway, is who is doing that building.

A new delivery team. SAP has stood up a delivery team for Energy & Natural Resources (utilities, oil and gas, mining). It is led by Ye Li and reports directly into SAP’s AI leadership.
What it builds. Its scope is the parts a general-purpose model cannot supply for us: a utility-specific ontology, standard templates, engineered grounding, and knowledge-graph mappings of our SAP utility business data.
Where it starts. The first implementation is across PPL’s three operating areas: LG&E/KU, Rhode Island and Allentown. It is delivered by an SAP forward-deployed engineering team, and its explicit purpose is both to deliver for PPL quickly and to lay the foundation for a new set of industry products.
When. Work on the first 30 agents kicks off in January.

PPL is leading, and the rest of us benefit from a utility willing to go first. PPL’s own enterprise architects have offered to share progress with peers. That is exactly the kind of sounding board this community exists for. It also means the ontology at the heart of the next generation of SAP utility products is being shaped now, inside one investor-owned holding company with three operating companies and four jurisdictions. How that ontology is reviewed, contributed to and exported is a governance question for all of us. Co-ops, municipals, public power, water and gas-only utilities run on different processes and should be asking it before January, not after.

For now, the agents are designed to be asked, not to act alone. The exception agent starts “upon user request” and offers recommendations. SAP’s own framing is the right one: “Go from assist to autonomy. Start by making people faster and more accurate. As trust grows, enable touchless execution.”

Layer Four: The AI Utilities Actually Run Is Their Own

The production AI stories I found in the decks were built by utilities. They show up first in project delivery and IT support, not in customer-facing autonomy.

OG&E used generative AI during its S/4HANA program to grade specifications, write and refactor ABAP, generate test scripts and triage defects. Afterward it turned those tools into a platform:
- an incident tool that has analyzed more than 15,000 incidents;
- a delivery workspace with more than 80 projects in flight, grounded in its enterprise architecture repository.
Their most interesting use is governance: “GenAI is here to augment your internal organization to work with and validate the work your SIs are doing for you.”
PG&E’s Propel team moved specification-to-sign-off for custom objects from two to four months down to two to four weeks.
- What they did not change matters more: “Zero SDLC gates removed, 100% of approvals human-owned.”
- The engine is mostly a hyperscaler’s agentic coding tool rather than SAP’s.
- The most honest line of the conference was theirs: “Effort saved is not schedule saved.”
Nexus Water is using Joule across its migration and published the cleanest AI governance language I saw:
- confirm guiding principles and “no-fly zones”;
- “keep humans accountable — place approvals around financial, customer and policy impacts”;
- “start with standard — validate standard Joule capabilities before building agents.”
Consumers Energy reported automated remediation of 93% of its custom code and AI-generated specifications and test scripts from its integrator.
On the operations side:
- UGI runs AI-assisted scheduling with operator-qualification checks built into dispatch.
- Summit Utilities reported more than 80% auto-scheduling and a pilot of SAP’s Dispatcher Agent.
- Southwest Gas built a custom high-bill investigation tool on SAP Business AI for its contact center.
- The last two are agenda abstracts; I have not yet seen the decks.

The pattern is consistent:

bounded use;
a named owner;
human sign-off;
value measured first in project and support work.

PG&E described an agent lifecycle of register, certify, supervise and adjudicate, with a rule that fits on a sticky note: “No named SME, no plugin.” That is what utilities are answering on their own while the packaged agents arrive.

Layer Five: You Can’t Automate What You Can’t See

Avista’s enterprise architect titled her session with the problem: 1,600 applications and no process documentation. Her slide was blunter: “1,500 applications. Almost none of it documented in processes. Link to business capabilities unknown.” Her lesson for anyone starting out was “Toolchain first, then ERP — build the map before you redraw the territory.”

Hold that against New York’s new requirement that utilities file a complete inventory of their AI use cases. You cannot inventory what you cannot see.

On data platforms, utilities said plainly that SAP’s Business Data Cloud is part of the answer, not all of it.

Avista, evaluating BDC alongside Databricks for a greenfield S/4HANA program, concluded “complementary, not competing.” GIS, outage and ADMS data land in the lakehouse.
NRG feeds two lakes from one governed SAP layer.
SAP’s own BDC slide lists SCADA inputs, sensor telemetry, DER performance and wildfire risk maps as “customer-managed data products.”
On SAP’s roadmap, utility master-data products are dated Q4 2026 and billing and time-series products Q2 2027. The BPEM-exception and AMI-event data products are still “product vision.” Those are the data the Q2 2027 exception agents will need.

SAP’s Q3 results, two weeks after San Antonio, showed how central this layer has become to SAP’s own selling.

Nine in ten of SAP’s 50 largest deals in the quarter included AI and Business Data Cloud.
SAP described a knowledge graph anchored in BDC covering more than 7 million fields, about 500,000 tables and 400 data products.
AI-based master-data matching from Reltio, for customer, material and supplier records, is “coming very soon.” From Q1 2027, data products for non-SAP sources are to be generated automatically into that knowledge graph.
What SAP has not said is which system owns the utility master data an agent will reason over: business partner, premise, device and connection object. Reltio or MDG is a question anyone planning an IS-U migration should be asking now.

ASUG’s own research, cited in one partner session, reported that data migration as a top S/4HANA risk rose from 27% to 44% of respondents in a year. NiSource offered a counter-pattern worth studying. It runs GIS and EAM on a common platform with live CDS views in both directions and no middleware, and pushes back “on attempts to re-host GIS data in data lakes.” Two credible architectures, opposite directions, same goal: data an agent can trust.

Layer Six: The Hard Part Is Still People and Process

For all the agent roadmaps, the sessions that drew the most practical nods were about people and process.

PG&E gave a full session to change communications on a program that runs to 2030. Its premise fits on one line: “Technology has changed. Human behavior hasn’t.”
Dominion’s program had one non-negotiable: “Do no harm to the field worker.”
Puget Sound Energy assessed change impact across 61 departments and trained more than 2,000 staff in the ten weeks before its Service Cloud go-live.
TVA built a change-champion network across the business for a general-ledger cutover.
Avista’s most transferable lesson was not about a tool at all: “Connect the people, before the tools.”

This is where the enterprise architect’s job gets harder, not easier. Agents add a third participant to every process design conversation. Someone has to decide which steps the agent takes, which a person approves, and how the people whose work changes find out before go-live rather than after. EA is the discipline of managing complex change. The change just got more complex.

Two Weeks Later: What SAP Told Investors

On October 21, SAP reported its third quarter. The numbers say the business behind all those agent slides is working for SAP.

Total revenue passed €10 billion in a quarter for the first time: €10.27 billion, up 13%.
Cloud revenue was €6.67 billion, up 26%. Cloud ERP Suite, the RISE-anchored bundle most utility migrations buy into, was €5.92 billion, up 29%, and is now 89% of SAP’s cloud revenue.
Current cloud backlog, contracted revenue still to be recognized, reached €24.1 billion, up 28%.
Cloud gross margin slipped 0.7 points to 74.3%, which SAP attributes to AI and infrastructure investment. Services revenue fell 5% as SAP deliberately shifts from billable hours toward AI adoption.
SAP now expects 2026 cloud revenue toward the upper end of its €25.8 to 26.2 billion range.

Read against San Antonio, four things stand out.

Agents: target versus live. SAP’s stated goal was more than 400 agents by year-end. On the call it cited 40 assistants and 300 agents across the autonomous suite, up from 224 agents and 51 assistants at Sapphire in May, with one quarter left. The framing was “quality over quantity.” That is a reasonable position, and it is also worth dating. Count again in January.
No utilities in the examples. The customer stories on the call came from AstraZeneca, Decathlon, Salling Group, Accenture, Danfoss and Reinhausen. No utility was named in any win or go-live, and SAP’s commitment to Industry AI for energy and other asset-intensive industries by the end of 2026 did not come up.
Pricing is moving slowly. No tiered agent pricing was announced on the call, though SAP’s commercial guide, republished the same week, now prices agent actions in two tiers: 0.02 AI Units for a Standard action and 0.05 for an Advanced one. (Updated October 22, 2026.) SAP’s CFO described the move to consumption-based pricing as a “very gradual” ramp, with the aim of selling outcomes rather than tokens. The per-action billing questions from The Meter Is On remain open.
Migration is the revenue. An analyst asked directly about the ECC maintenance deadline. SAP said RISE deals are “richer” with AI and Business Data Cloud, but gave no ECC numbers. That is Layer One seen from the seller’s side: the calendar is what is filling SAP’s backlog.

What Wasn’t on the Agenda, but Was at the SAP Booth

Distributed energy and grid operations were not in the sessions, but SAP did have a very impressive demo of their new SAP DER solution on the vendor floor.

While the AI components for SAP DER are just starting, SAP’s end-to-end DER solution for electric utilities was shown to many utilities at the conference, both at the SAP booth and in one-on-one meetings.
SAP’s product managers told me they will be working with those utilities this quarter to take the next step in confirming the SAP DER roadmap.
The demo which I saw was complete and had all the components that a demand-side management program needs in order to deploy and manage distributed energy. I would encourage any utility looking at different demand-side management solutions to get with SAP and review its DER solution to see where it fits in your environment.

For the wider picture of what is happening in distributed energy, our weekly DER research digest tracks the news, research and regulatory developments every Friday.

The question this series keeps asking, where agent authority stops at the operational-technology boundary, was not answered in San Antonio, but in regulated industries, we all will continue to ask these questions.

Honest Caveats

Coverage. I read 46 of 79 decks. The other 33 are agenda abstracts, and abstracts are promises. The coverage gap falls hardest on the AI track: of the 22 sessions the agenda tagged as Autonomous Enterprise, only ten have decks posted. Several of the most promising AI stories may change this picture when their decks arrive:

Summit’s Dispatcher Agent pilot;
Southwest Gas’s high-bill tool;
SAP’s live billing-exception demo;
SAP’s session on AI consumption models.

Vendor framing. Benefit numbers in SAP decks come from SAP’s value-modeling teams, and partner case figures are vendor-asserted. One utility’s two decks report different savings figures, and one of them carried a note that the figures still needed reconciling. I have quoted mechanisms and dates over percentages for that reason.

Financial figures. The Q3 figures are SAP’s non-IFRS numbers, with growth year over year, from SAP’s quarterly statement. Quotes from the earnings call come from a third-party transcript and should be checked against SAP’s own webcast before they are relied on.

Slides are not conversations. The best material at any conference happens in the hallway. This piece covers what was put on screen.

My own involvement. I served as a strategic content advisor to ASUG for this conference, moderated the clean-core development session, help convene a partner user group that appears in one deck, and have a working relationship with many of the partners, helping them engage their customers in user groups. This is something that is needed throughout the SAP ecosystem, not just for SAP. All of the products you invest in should come with a ready-made user community of existing customers to collaborate with. It’s my goal to help all the partners do this. Over the past year I have also been working with SAP and its partners on SAP DER, contributing to those product discussions, so weigh what I say about the DER demo with that in mind. I also co-hosted our 2026 EA workshop with SAP’s Martin Mysyk, and I am helping to arrange an ongoing series of webinars, starting with the reprise of NiSource’s LeanIX presentation on November 2 (see our Events page).

What We Still Need to Influence

Name the adopters. Which utilities run the self-service agent and the case-management assistant in production, and with what measured results rather than modeled ones? SAP’s Q3 call named customers in six industries and no utility.
Publish the ECC matrix. Several field and asset agents are marked as available on ECC. Publish the full list by agent, so utilities on the 2027–2030 path know what they can use before they convert.
Govern the industry ontology. The Energy & Natural Resources team’s work at PPL is meant to become the foundation of new products for all of us. Publish how other utilities review it, contribute to it and export what is theirs, and give public power, co-ops and water a seat before the first 30 agents harden into a template.
Explain “value-based” for regulated buyers. A north-star subscription priced on modeled value has to survive a prudence review in a rate case. Show us the terms.
Date the data the agents need. The BPEM and AMI-event data products are undated while the agents that depend on them are dated Q2 2027. And say whether Reltio or MDG owns the IS-U master-data domains, business partner, premise, device and connection object, that those agents will read.
Put grid operations and DER on the dated roadmap, and say plainly which pieces belong to partners and which pieces will be part of the SAP solution.
Give sovereign-cloud customers an AI release schedule.

One encouraging data point: SAP reported that 64 customer improvement requests for Service Cloud V2 and related tools were delivered in a single quarter, and its own advice to utilities was to vote together. Influence works when we use it. See our Influence page.

Your Turn

Which of these layers is your organization living in right now: the calendar, the governance, the waiting, the building, the seeing, or the people?

A few concrete ways to keep this conversation going:

Conference reprises: we are bringing San Antonio sessions back as Utility.Community webinars. First up is Evan Prince and Jeff Carbol of NiSource on November 2, on running enterprise architecture with LeanIX (“EAs need tools as well”).
If you attended: complete the conference survey and download the decks. 46 of the 79 are posted now.
Ontology: if you have a view on how a utility-industry ontology should be governed before it hardens into product, let’s start that working group now.

Need an introduction to anyone mentioned here? Ask. Find me on Utility.Community, or reply here.