“There is just not a room for error”: SAP is putting its AI agents in the back office

Earlier this year, at its Sapphire conference, SAP outlined its plans for an “Autonomous Enterprise” run in large part by AI agents. This week at Connect in Las Vegas, the company is focusing on how it is actually putting those agents in front of business users.

While Sapphire caters to IT, Connect targets the line-of-business practitioners who run finance, procurement, HR, and supply chain operations day to day.

Eric van Rossum, SAP’s head of product marketing for applications and suite, tells The New Stack, “Connect is all about making it real. We set the strategy at Sapphire. Now we need to come with the proof points, what that means for our customers, what it means when we have the products in our hands and so on.”

SAP is launching new Joule Assistants across finance, spend management, HR, customer experience, and supply chain this week, and it’s running more than 20 hands-on labs where attendees can build their own agents.

“We set the strategy at Sapphire. Now we need to come with the proof points, what that means for our customers, what it means when we have the products in our hands and so on.” — SAP’s Eric van Rossum

Most of the new assistants will roll out through early adopter programs and general availability releases between now and the first quarter of 2027.

SAP Sapphire 2026 – Orlando, Florida. Credit: SAP.

TechWolf acquisition

SAP also announced plans to acquire TechWolf, a Belgian company whose AI platform maps the work and skills inside an organization. Terms weren’t disclosed. SAP expects the deal to close in the fourth quarter, pending regulatory approval.

“Organizations everywhere are trying to figure out how AI is reshaping work, and what their workforce needs to look like because of it,” Andreas De Neve, TechWolf’s CEO and co-founder, says in the announcement. “We have spent eight years building the evidence layer to answer those questions. It’s always been our mission to connect skills, tasks and work to help employers solve AI transformation, and joining SAP will allow us to extend that mission across a broader set of business context and larger scale. In this next chapter, we plan to build a world-class AI company.”

Starting with finance

Customers tend to start wherever their most significant pain points are, van Rossum notes. “We see a lot in the finance area, like financial close, billing reconciliation, and elements like that.”

The new Revenue Recognition Assistant walks finance teams through how revenue is configured, calculated, and posted, and classifies transactions under the IFRS 15 and US GAAP revenue standards. A Disclosure Assistant automates the final steps of preparing external financial reports, from gathering and tagging the data to drafting the narrative and checking the result before it’s published. Meanwhile, an Overhead Accounting Assistant lets controllers update cost allocation rules and make bulk changes to cost and profit centers, which SAP says speeds up period-end work.

The finance assistants SAP launched at Sapphire will also get dedicated agents for the financial close, treasury, tax, billing, planning, and receivables. SAP expects to open early adopter programs for the new finance assistants in the first quarter of 2027.

There’s also an International Trade Assistant for companies that ship across borders. It classifies products, prepares compliance paperwork, and handles sanctions screening. In early deployments, SAP says, it has cut the work of classifying products for trade by as much as half.

“If you’re going to do a financial close or if you’re going to do statutory reporting or many of these items,” van Rossum says, “there is just not a room for error.”

SAP Pay

SAP is also moving into payments with SAP Pay, a service built on its newly formed Tereina subsidiary. As soon as an invoice is approved in SAP Cloud ERP, SAP Pay sends the payment and matches it against the invoice or purchase order, all without leaving the ERP.

For cross-border payments, companies can also settle in stablecoins. The service is available now in the U.S. and U.K.

How SAP governs AI agents

An agent, van Rossum says, “to a certain degree is actually just an extension of the process in the back end.”

Over five decades, SAP has built country-specific localization and regulatory requirements into its core applications, he notes, from Brazil’s Nota Fiscal electronic invoicing rules to AI regulations that differ between Europe and the U.S. “All of those things were always baked into our core applications, and that really gets inherited up to the agent as well.”

SAP calls this governing agents “from the inside out.” A new Governance Assistant, due in the fourth quarter, ties into the governance, risk, and compliance (GRC) tools a company already runs and flags risk and control issues across systems, so compliance teams can monitor continuously rather than in periodic reviews.

Starting in the first quarter of 2027, the Access Governance and Security Assistant will bring AI agents under the same identity governance and access controls as human users.

Results from the supply chain

Some of the earliest customer results SAP is talking about come from supply chain use cases. SAP is adding new Joule Agents to the six supply chain assistants it showed at Sapphire. Among other things, they can read product designs and recipes, spot disruptions on the plant floor, and kick off maintenance requests.

A global producer of natural ingredients, SAP says, is using the Logistics Assistant when something goes wrong in its logistics operations, while a large biopharmaceutical company, the Supply Chain Planning Assistant tracks down why planning problems happen and proposes fixes. Combined with automated performance tracking, SAP says, that gives the company a feedback loop for making its planning gradually more autonomous.

Across the suite

For now, most customers are starting with individual use cases, van Rossum says. He expects bigger gains once agents work together across workflows that run from finance to spend management, customer operations, and HR.

“Optimizing one LoB, I think we’re competing in a very narrow environment with best-of-breed, but our strength is really going to come from doing that from a cross-end-to-end perspective,” he says.

SAP CEO Christian Klein makes a very similar case in the company’s announcement, saying SAP is “helping organizations connect intelligence across every function to deliver trusted business outcomes.”

To connect those workflows, SAP now offers managed integrations that link SAP S/4HANA Cloud Public Edition to SAP Ariba and SAP Sales Cloud. Connections to SAP Taulia (this month), SAP Subscription Billing (November), and SAP SuccessFactors (by year’s end) come next. SAP runs and updates these integrations itself, and they’re included in the cloud ERP subscription, so customers don’t have to stand up separate integration projects.

SAP Cloud ERP Private customers can already try a managed Joule integration through early adopter care, ahead of a general release SAP expects by the end of the year.

“Spending any IT dollars on integration and stitching together products delivers very little value for our end customers,” van Rossum says. SAP would rather customers put that money toward agentic AI and other transformation projects, he adds.

SAP is also bringing the master data management technology from its Reltio acquisition into its AI platform with SAP Reltio, now generally available in SAP Business Data Cloud. When the same customer, supplier, product, or employee shows up in different forms across SAP and non-SAP systems, SAP Reltio works out which records belong together and reconciles them into one profile, so Joule and its agents aren’t reasoning over conflicting data.

Measuring AI agent ROI

Van Rossum says more customer conversations now start with business outcomes rather than the technology itself. Agents add cost, he notes, and that cost needs to be offset through efficiency, productivity, or new revenue. “This needs to lead to tangible business outcome and benefits.”

To help customers make that case, SAP worked with consulting partners and customers to validate benchmark KPIs for every assistant, according to van Rossum. The company is also launching a value calculator that benchmarks a customer’s own performance data against those KPIs to show where the assistants could help. Over time, SAP plans to connect that with SAP Signavio for process analysis and the AI Agent Hub in SAP LeanIX for managing agents.

In the first quarter of 2027, SAP Signavio will also get improved agent mining, which applies process mining to workflows where multiple agents hand work off to each other. And new AI insights in WalkMe let business leaders ask, in everyday language, where employees get stuck, including with AI assistants, and WalkMe builds a dashboard from live usage data to answer.

By van Rossum’s own account, most customers are still working through individual use cases. Over the next year, SAP wants to get those assistants handing work to each other across finance, procurement, HR, and the supply chain, where it believes its suite gives it an advantage over best-of-breed vendors.

The post “There is just not a room for error”: SAP is putting its AI agents in the back office appeared first on The New Stack.