Nine articles. Roughly fifty questions. Not much has come back — and most of that silence is ours, not SAP’s.
The following Utility Voice was authored by Marc Rosson, Community Connector at Utility.Community.
Since May I have ended nine articles with a question. Very few have come back with an answer.
I want to spend this one being honest about that, because in October we gather in San Antonio, and a community that only asks questions is not a community. It is an audience.
This is not a complaint about response rates. Something more useful came out of putting five months of questions in one place: they are not one kind of question. They sort by who can actually answer them — and a good many of mine went unanswered because I asked them in a room that had no way to answer.
That sort is the whole of this article. It is also, I think, the most practical thing I can hand you before the conference.
Layer One: Four Rooms, Not One List
Stack the nine articles and roughly fifty questions fall out. They divide four ways.
There are questions only you can answer, inside your own utility, with nobody’s permission. There are questions only SAP can answer, because the information does not exist outside the vendor. There are questions only we can answer for each other, because they are about lived experience and no vendor or analyst has it. And there are questions nobody has answered, which are a different animal entirely.
This is not tidiness. Aim a question at the wrong room and it returns nothing, every time. Ask a peer what only your account team knows, and you get a shrug. Ask SAP what only your own integration inventory can tell you, and you get a meeting with a slide in it. I have done both, in print, and I got exactly what I deserved.
So before you ask anything in October, know which room you are standing in.
Every question the series has asked, sorted into those four rooms with the article it came from, is on its own page: The Questions We Need Answered: the full list. Take it to San Antonio.
Layer Two: The Ones Nobody Is Stopping You From Answering
Start here, because this is the largest group and the least excusable.
Have you run the three-pile sort on your integration landscape — published APIs, non-published interfaces, and the ones SAP has explicitly closed? What fraction landed in the middle pile, and did any mission-critical flow turn out to be standing on ODP-RFC?
Have you drawn the line between the read tier and the act tier in writing, before an incident forces it?
Can you tell your operators that the data your agents read is governed, current, and traceable to a source you trust?
If a customer-facing agent went live tomorrow, would it see one accurate, consented, current view of each ratepayer — and would it know where its authority stopped?
And the one this series keeps returning to, because a regulator will ask it first. If your utility put an AI agent into production tomorrow, could you say who authorized it, what rules govern it, who watches it, where it sits in your organization, and how you would shut it down?
Not one of those requires SAP’s cooperation. Not one requires a peer’s disclosure. They cost staff time and management attention, and nothing else. If they are unanswered at your utility, it is because nobody has run the exercise — and running it is the single highest-return thing you can do between now and San Antonio, because it converts you from someone with questions into someone with findings.
The difference matters more than it sounds. A room full of people with questions produces a panel. A room with four people carrying findings produces a working group.
Layer Three: The Three With No Published Answer
Now the vendor room, and I want to narrow it hard, because a long list of asks is easy to defer and a short list is not.
Last month I put six questions on paper for account teams, on the economics of agent consumption. Three of them have no published answer today:
How many actions will a given business process generate, in our configuration?
What precisely constitutes one action in each agent implementation we are licensing?
How do retries, branches and tool invocations map onto actions?
SAP has defined the unit and published the rate, and deserves credit for both. What does not exist is the mapping from a real utility process to a billable action count — and that count is the only number a budget actually needs. Everything else in the pricing conversation is arithmetic once you have it.
Two more belong in the same room. Ask a vendor waving a certificate whether their standard is cited in the Official Journal, because that single question separates a legal presumption from a marketing asset. And ask whether agent inventory reaches tool-level dependencies, or stops at the agent — because one agent can hold many tools, and that boundary determines whether an inventory answers an auditor or merely gestures at one.
The four purchases, and what has shipped
This year SAP also bought most of the architecture it has been describing to us. Reltio for master data, closed May 7. Dremio for the open lakehouse under Business Data Cloud, closed July 6. Prior Labs for tabular foundation models, closed July 17. And a stake in n8n, the workflow tool it is embedding in Joule Studio as the layer that turns an agent’s analysis into an action. Read together, they are a coherent stack: trusted data, reachable data, a model that can read a table, and the glue that executes. I have not seen another ERP vendor assemble that natively.
But this series has tried to hold one discipline throughout — announced, delivered, adopted are three different words — and as of September 7 the honest reading of all four is announced and closed, not delivered. SAP’s completion notices for Reltio and Dremio carry no integration plan and no dates. Prior Labs operates as an independent lab, and its TabPFN-3 model is published under an evaluation-only license; production use needs a separate commercial agreement. n8n’s general availability inside Joule Studio was targeted for the third quarter, and with three weeks left in it, SAP’s own developer tutorial still labels the environment pre-release and not for production. On the July earnings call SAP moved the year-end promise from 224 agents to more than 400, cut its operating profit guidance to absorb the deals, and said the count of what is actually live would be reported with the third-quarter results — which land in late October, a fortnight after we leave San Antonio.
None of that is a criticism. Integrating three companies in one summer is slow because it should be. It is a reason to add three questions to the vendor room, because they are answerable by nobody else and every one of them lands on a utility system.
Which master data product is an IS-U shop supposed to standardize on? SAP now owns Master Data Governance and Reltio and has put both inside Business Data Cloud. Customer, premise, device and connection-object data spread across CIS, MDM, GIS and a DERMS is precisely the four-copies-of-one-entity problem Reltio was bought to solve. We need the answer before an agent starts reconciling those copies on its own.
Which agent actions belong on the governed backbone, and which on the agile layer? SAP’s stated split is that Integration Suite stays the deterministic, high-volume path and n8n becomes the human-in-the-loop path that reaches Teams, email and external systems. That is a governance decision dressed as a tooling one, and it runs straight into the OT boundary question in Layer Five. An n8n workflow that can notify a crew is one thing. One that can touch a switching order is another.
When does a tabular model become a product a regulated utility can license? Predicting payment delay or outage risk directly on ERP data without a data science team is the capability that would move Joule from answering to deciding. Today it is a benchmark and a research license. The date on which it becomes something we can put in a rate case is a date only SAP has.
October is the one week this year when the people who can answer these are in the same building as us. A question asked in a hallway in San Antonio is worth ten asked into a comment thread.
Layer Four: The Silence That Is Ours
Here is the group where the return has been thinnest, and where I cannot blame anyone but us.
Has anyone successfully negotiated a hard cap or an alert threshold on AI consumption — and what did it cost you elsewhere in the agreement? Has anyone built the capability registry, a living inventory of every AI tool with its owner and risk tier, and did it survive a real integration backlog? Has anyone taken the accountability question to an internal auditor and gotten a useful answer back? Is anyone treating a European conformity declaration as supply-chain evidence today, and how did the auditor react?
SAP cannot answer these. No analyst can. No consultant can, whatever they tell you. They are answerable only by a utility that has done the thing, and their entire value is that they come from someone who has.
I understand why they have been met with quiet. Utilities do not publish. Some of this is contract terms under NDA. Some is competitive caution. Some is the entirely reasonable instinct that saying “here is what our authorization model looks like” in public is a bad idea for a critical-infrastructure operator, and I would not argue otherwise.
But there is a difference between not publishing and not telling each other. The fifth article in this series argued that the agreement between SAP, DSAG, and our community was itself a form of leverage. Leverage requires that we know what one another has actually done. A community of operators who each solve the same problem privately has all the cost of solving it ten times and none of the benefit of having solved it once.
You do not have to write an article. You have to tell one person in San Antonio.
Layer Five: The One Question Nobody Has Answered
One question appears in six of the nine articles and is answered by no source I have found.
Where does agent authority stop at the OT boundary?
Not SAP’s published agent architecture. Not the AI Golden Path. Not — fairly, given its scope — the DSAG utilities survey. ADMS, SCADA and OMS sit outside SAP’s perimeter and inside many of the workflows agents will touch, and the guidance on that seam does not exist.
I could not see that from inside any single article. It only appears when you stack five months of them and notice the same gap in each. That is worth saying plainly: the silence here is not ours. We have asked, repeatedly, in the right room, and nothing has come back.
Which makes it the one to press hardest in October.
Honest Caveats
Several things cut against this piece, and they should be said plainly.
Asking is cheaper than answering, and I have been on the easy side of that trade for five months. It costs me an afternoon to pose a question and costs you a quarter to answer one. I am aware of the asymmetry and I am not going to pretend the ledger is balanced.
The denominator is mine. I counted my own questions and graded my own response rate. Some of those questions were rhetorical. Some were premature. Some were badly aimed, which is the argument of Layer One turned back on its author.
Silence may simply mean the question was early. In May, “how many actions will this process generate” was not answerable by anyone, SAP included. A question that arrives before its answer exists is not evidence of a disengaged community.
People may well be doing this work and not talking about it. I would rather that be true than the alternative, and if it is true, the fix is a conversation rather than a program.
The acquisition scorecard has a short shelf life. Everything in Layer Three about what has and has not shipped is true on the day I wrote it and may be wrong by the time you read it. That is fine. Ask the account team what is generally available today, not what was announced in May, and you will have a better answer than this article.
And the conference is not a safe venue for all of it. I am not asking anyone to breach a confidentiality term. On the commercial questions I want the shape, not the numbers — that a cap was achievable, roughly what was traded for it, not what you pay.
What I Am Actually Asking For in San Antonio
Not a panel. One answer each.
Bring one question from this series — the full list is here — that you have answered — even partially, even badly, even if the answer is “we tried and it did not work” — and tell one other person how it went. If you have run the integration sort, say what fraction landed in the middle pile. If you have drawn the read-versus-act line, say where you put it. If you have asked your account team any of the three unanswered cost questions, say what came back. If you took the accountability question to your auditor, say what they asked you that you had not thought of. If your account team has told you which master data product you are meant to be on now that there are two, say which.
Then I will write it up. That is the next article in this series, in November, and it is the one I would most like to write: what answers actually came back, named or unnamed at your discretion.
If nothing comes back, I will write that too. It would be a finding about us rather than about SAP, and it would be worth publishing.
Five months ago I started asking whether we could answer a regulator. The prior question — the one I should probably have asked first — is whether we can answer each other.
Find me in San Antonio. Bring one answer. The easiest place to do that is Wednesday evening, October 7, from 6:00 to 8:30 at the Yard House on the River Walk, a short walk from the hotel — register here, since space is limited.