The honest take on AI in SAP

A snapshot of what runs in SAP today, and what is coming.

AuthorAuthorAuthor

Ellen Köhler, Simone Rüschenberg & Dr. Wolfgang Faisst ·
September 10, 2026 • Estimated Reading Time: 6 minutes

Read of the week

SAP has promised 200 AI agents by the end of this year. In August, 17 were generally available. An agent is software that finishes a task on its own, for example matching open invoices. A chatbot only answers questions. An agent acts. That makes 200 a big promise.

We interviewed Dr. Wolfgang Faisst, Co-Founder and CEO of ValueWorks.ai. He spent 14 years in SAP product development, co-started S/4HANA and co-led SAP’s early AI program. Unless otherwise cited, all information in this episode comes from that interview.

Three terms to keep in mind:

  • ECC: the old SAP, running on your own servers. Standard support ends in December 2027, so everyone has to migrate.

  • S/4HANA: the current SAP. Despite the name, not automatically cloud. Three flavours: public cloud (SAP runs it, one standard version), private cloud (your own SAP, hosted, custom code comes along), on-premise (your own servers).

  • Joule: SAP's AI, see section 2.

00· · In this issue

01· · Why SAP is under pressure
02· · Joule, SAP's AI: what it does, and what it does not
03· · S/4HANA is probably not the finish line
04· · The bill: from seats to consumption
05· · Your data: how to get it out

01· · Why SAP is under pressure

Not from its customers. Cloud revenue still grows at more than 20% a year. The pressure comes from three sides:

  • The stock market. The share nearly halved in the twelve months to June 2026, and in August UBS cut the stock from Buy to Neutral over the slow agent rollout.

  • Its own late start on AI. Integrating bought products like Ariba and Concur took the focus away. CEO Christian Klein now runs product development himself.

  • A new generation of ERP competitors, built for AI from day one.

Three waves of enterprise software

  • First wave: SAP, Oracle, Sage. Born on-premises.

    Second wave: NetSuite, Workday. Born in the cloud.

  • Third wave: Rillet, Campfire, DualEntry, Lensing. Born with AI.

💡 SAP is not short of customers. It is short of shipped AI. The share price shows it.

02· · Joule, SAP's AI: what it does, and what it does not

Joule is SAP's AI: an assistant inside the SAP screens, and the place where SAP's agents will run.

It already does

  • Answer questions on your own data, summarise, draft, turn a question into a chart

It does not yet do

  • Reliable answers. Independent tests show Joule answering specialist questions inconsistently, sometimes wrongly. Useful entry point, not a source of truth.

  • Your whole landscape. It sees your SAP data, and little of your non-SAP systems.

  • Work with outside tools. That door opens in Q4 2026 at the earliest.

💡 Before you believe a Joule demo, ask: which edition, which release, generally available or early adopter?

03· · S/4HANA is probably not the finish line

Many companies are moving from ECC to S/4HANA right now, because ECC's standard support ends in December 2027. What does that buy you?

  • SAP's AI features assume a clean core: standard SAP, custom code moved out. Only the public cloud flavour is clean-core by definition.

  • Public and private share the name, but run on two different code bases.

  • Whoever migrates to S/4 private today will likely face a second migration to public cloud.

And the target keeps moving. In 2026 SAP spent over a billion euros on a new data layer:

  • Dremio, a data platform (lakehouse)

  • Reltio, master data

  • Prior Labs, forecasting models

SAP also put money into n8n, the automation tool, in May. Amount and stake are not public; the deal values n8n at 5.2 billion dollars, and n8n gets built into Joule Studio. SAP Databricks is not a purchase but a partner product. Its future is open, because SAP now owns Dremio and Prior Labs, which cover similar ground.

SAP is knitting all of this into one platform. Once done, customers will most likely have to move again, onto the new stack.

One weak spot stays: non-SAP data. Half your landscape or more is not SAP. Many customers therefore build their data layer on another stack, often Microsoft Fabric.

💡  Plan for a series of migrations, not one.

04· · The bill: from seats to consumption

Until now you paid SAP per user (seats). Agents break that logic: software does the work, not users. So SAP moves to consumption pricing: you pay per use. Think of a phone plan:

  • Included in your cloud subscription: chat and the simple assistants.

  • Paid on top, by usage: the advanced features, above all agents. One agent run costs a multiple of a chat.

  • Building your own assistants stays free until 31 December 2026. 2027 prices: not published.

Read that again as a CFO. You pay per use to analyse data you already own, inside a system you already pay for. Well, this has caused real anger among customers.

The €210–325k are one worked example, built on public list prices and volume assumptions. Not a quote, not a benchmark; take it as a direction.

Four lines belong in the order form: an AI-unit allocation in numbers, a consumption cap, a benchmarked overage rate, one owner of the invoice.

05· · Your data: how to get it out

Tools copy data out of SAP over official interfaces. Many also used an internal shortcut called ODP-RFC, because it was fast. SAP never officially allowed that.

In June, SAP closed the shortcut with a security patch. Pipelines on it stopped overnight. Reopening is possible, at your own risk, until December 2026.

The good news: getting data out works today, no migration needed. Copy it over an official route into a data layer outside SAP; any AI can work with it there. Wolfgang's advice for every ECC customer.

Before you plan, run the SAP Readiness Check: a free tool your IT team starts inside your system. The report lists what a migration would break.

💡 Get your data into a layer you control. Then you can start with AI now and migrate SAP on your own schedule.

One thought from Ellen and Simone

We are not SAP experts, and this is something we keep debating in our network.

Most of the new AI-native ERP players are being built in the US. In Europe, SAP still has something they do not: enormous trust, an installed base that is hard to match, and decades of knowledge about how companies actually run.

So here is the question: at what point does it make more sense to build something new rather than keep modernising what already exists? There is a point where enough has been invested in the old system that starting again becomes unthinkable, even when a clean sheet might end up better.

With its scale and its customer base, why not build a genuinely AI-native ERP alongside the old one, and give customers a gradual path across?

A big thank you to Wolfgang for the candid conversation. If you have questions, you can reach him via his LinkedIn profile.

Bottom Line

  • Pricing changes: from 2027 the advanced AI is consumption-based, paid per use. 

  • S/4HANA is probably not the finish line. Expect another move onto SAP's new data stack.

  • On ECC: check whether you can migrate straight to public cloud, skipping private cloud. That avoids a second migration.

  • You can start with AI now, outside SAP, on your own data in a data layer. 

  • First, run the SAP Readiness Check, free for SAP customers. It shows what a migration would break.

  • And SAP's timeline is under pressure. The next months show whether the catch-up succeeds and the trust holds.

CFO Watchlist

60% of CFOs want to shape value creation. Only 25% lead the decisions.

  • EY surveyed 1,610 CFOs. 47% of their capacity still goes to reporting and controls.

  • Only 21% rate their AI readiness as advanced. Data quality is the top barrier.

  • Just 11% call their team highly adaptable. Those teams beat targets 42% of the time, the rest 3%.

Surveyed February and March 2026, publis

Mistral raises €3bn, the largest tech round in European history

  • Mistral builds LLMs, the European answer to Anthropic and OpenAI. €3bn Series D at a €20bn valuation, double a year ago. Samsung led, with Nvidia, ASML and the EQT-managed Scaleup Europe Fund.

  • With EU data rules on every CFO’s risk list, a well-funded European alternative is worth watching.

  • We interview Mistral’s FP&A Lead at our Finance Collective Launch Night in Vienna on 22 September. Join us.

Announced 8 September 2026

German business mood turns, despite energy prices

  • ifo business climate up to 88.8 in August from 86.7 in July.

  • Companies rate current situation and outlook better. Uncertainty fell.

  • Production set to rise over the next three months.

Published 25 August 2026

FROM SIMONE

CFO Summit: The New Era of Finance

11 November 2026 · Munich

The CFO Summit is where the Finance Collective community meets in person, open to any finance leader interested in the new era of finance. On the agenda: how the finance function, the tech stack and the CFO role are being reshaped.

WHO WRITES THIS

Two CFOs, two vantage points.

Ellen sits in the operating team of a PE growth fund and works closely with scale-up companies. Simone spent 20 years running finance functions, then switched sides to build Finance Collective, a network of CFOs. Every second Thursday we write about what's on finance leaders' desks right now, and talk to people from our networks who have already solved it.

Feedback, suggestions, or something you want us to cover? Write us on [email protected].

CFO Playbook reflects our personal opinions, not those of our employers and is not professional advice.