SAP has just reported its first quarter with more than €10 billion in total revenue, with its current cloud backlog up 26% at constant currencies. Christian Klein's verdict: the Autonomous Enterprise strategy "is not just resonating, it is being adopted."
Customers are signing. That moves the question from the keynote to the contract: what has to be true inside an organisation before an agent can run a piece of its business? Agents don't arrive on a clean slate. They inherit the core system, the data, the processes and the controls they are given. Those are the building blocks, and none of them come with a licence.
What SAP means by the Autonomous Enterprise
At Sapphire in May, SAP set out the stack. At the bottom sits the SAP Business AI Platform: Business Technology Platform, Business Data Cloud and a knowledge graph that gives agents the context of SAP's processes and data. Above it, the Autonomous Suite reorganises the applications into finance, spend, supply chain, HR and customer experience, with more than 200 agents and over 50 Joule Assistants. Joule Work is the front door, and the AI Agent Hub is where agents are registered and governed. SAP says each assistant comes with defined ROI KPIs, and that RISE with SAP customers get a contractual commitment to activate three Joule Assistants in their first year.
Analysts add a fair caveat. Constellation Research notes that the vision is not unique (ServiceNow, Salesforce, Microsoft, Workday and the hyperscalers are chasing the same idea), that much of it is still rolling out, and that the value case is clearest for organisations already heavily invested in SAP, with a clean core and well-governed data. In other words, the advantage goes to whoever executes.
Five building blocks that decide whether it works
From the deal side, these are the five things that decide whether an autonomous process works in production or only in a demo.
1. A clean core. SAP's agents are built against standard processes and released APIs. Heavily modified systems are where they fail quietly, or need expensive custom work to reach. The useful number isn't how much custom code a programme migrates. It's how much it retires.
2. Trusted master data. An agent that pays a supplier, re-plans stock or chases a customer acts on whatever record it is given. If two customer records disagree, it won't ask which is correct. Every key data object needs a named business owner, a quality measure and enough full data rehearsals to prove it before go-live.
3. Processes you can describe. Agents run processes that can be written down: steps, rules, exceptions and who decides. Many organisations only know their happy path. The rules-based, high-volume steps are the first candidates for autonomy; the rest need process owners with the time to decide.
4. Guardrails and sign-off. Autonomy is a set of decisions, not a switch. Which actions can an agent take alone, which need a human, and who answers when it gets one wrong? An agent register, access limited to what each agent needs, and a re-test whenever a model, prompt or data source changes are what make that auditable.
5. A business case per assistant. If an assistant has no baseline and no KPI before go-live, nobody will be able to say afterwards whether it worked. The business case also has to carry the AI licence and consumption costs, and someone has to own the benefit once the programme team has gone.
Check your readiness
Fifteen questions, three per building block. Answer for your own organisation if you are buying, or for the client you are bidding to if you are a partner. It takes about five minutes, and nothing you enter leaves your browser.
How ready are you for the Autonomous Enterprise?
I am answering as:
Your strongest and weakest building blocks, and what to do next, will appear here.
Five questions for your next SAP proposal
Whatever the score, the same five questions will tell you whether a proposal plans for autonomy or only promises it. Buyers can put them into the evaluation; partners can use them as a red-team checklist before they submit.
- 1How much custom code will be retired, not just migrated, and where does the plan show it?
- 2Who owns each master data object, and how many full data rehearsals are planned, with what exit criteria?
- 3Which process steps will an agent run in the first release, and which stay with people?
- 4Where is the agent register, who signs off agent output, and what triggers a re-test when a model changes?
- 5What is the KPI and baseline for each assistant, and are AI licence and consumption costs in the price?
A note for the partners bidding
Every SAP proposal this year will mention Joule and agents. The ones that win will be the ones that show the building blocks in the plan and the price: a data workstream sized for real rehearsals, process owners named and given time, a governance model for agents from the first sprint, and a business case the client's CFO can follow. That is a better story than a vision slide, and an evaluator can score it.
Buying SAP, or bidding for it? For buyers, our buy-side tender support writes these building blocks into your requirement and scoring, so every bidder has to answer them the same way. For partners, our Red team reviews test your proposal against them before the evaluators do. Related reading: AI-First ERP and How Many Data Runs?
Sources
- SAP News, SAP Announces Q3 2026 Results, 21 October 2026.
- SAP News, 2026 SAP Sapphire Keynote: Powering the Autonomous Enterprise, May 2026. Customer figures and assistant ROI KPIs quoted there are SAP's own.
- Constellation Research, SAP Sapphire 2026: SAP makes its case that it should be your autonomous enterprise platform, May 2026.
- The readiness check and the five building blocks are Dealcraft Advisory's practitioner framework, not an SAP methodology.