AI in ERP Systems: What's Already Being Automated Today and What's Next

Glühbirne mit Gehirn und KI-Chip auf violettem Hintergrund

October 12, 2026

 

Gartner predicts that by 2026, four out of ten enterprise applications will include task-specific AI agents. In 2025, the figure was not even one in twenty. In Germany, according to Bitkom, active AI use in companies has more than doubled within a year, from 17 to 41 percent. Given these figures, it’s worth taking a look at the state of AI integration in ERP systems. In this article, we’ll therefore examine what artificial intelligence is already capable of today and where the trend is likely to head in the coming years. We’ll also take a closer look at our partner ERP system, Haufe X360.

First things first:

  • Major international ERP providers are using initial real-world data to demonstrate just how far AI agents have come today—for example, in automated accounting.
  • According to Gartner, the data foundation is crucial to success: AI projects run significantly more reliably with clean, well-structured data.
  • Three maturity levels help with classification: assistive AI, process-integrated AI, and agent-based AI.
  • With Release 2026R1, Haufe X360 is specifically expanding its capabilities to include its own AI-powered data analysis, thereby building on automation that has long been running reliably within the system.

How Quickly AI Is Being Integrated Into ERP Systems

The jump from less than 5 percent to 40 percent in the use of AI agents in enterprise software within a year comes from a Gartner study published in August 2025. By 2035, Agentic AI could account for around 30 percent of total revenue from enterprise application software in a high-end estimate—an increase from the current 2 percent to over $450 billion. For Germany, a Bitkom survey of 604 companies conducted in April 2026 paints a similar picture: Among companies with more than 500 employees, active AI usage already exceeds 60 percent. The study explicitly identifies AI agents for accounting, scheduling, and customer service as growth areas—precisely the processes that converge in an ERP system.

What is already technically possible today

Some of the major international ERP providers rolled out their first AI agents with robust real-world metrics in 2026. For example, in the automated processing of journal entries in a widely used cloud-based ERP system, manual rework decreased by 70 to 80 percent in the first quarter following implementation. At another provider, the time required to process payment disputes was reduced from an average of four days to less than four hours. These figures show that AI agents in the ERP environment have long since moved beyond the announcement phase. 

The database makes all the difference

Gartner has examined what determines the success of AI projects in the enterprise environment: With a solid data foundation, AI projects run significantly more reliably; without one, the success rate drops to about 40 percent. The decisive factor, therefore, is not the technology itself, but rather how well data, processes, and responsibilities are prepared before an AI function goes live. Companies with a clear data model and defined responsibilities for AI-related issues are already among the pioneers today, even though this group remains a minority overall. For German small and medium-sized enterprises (SMEs), the DIHK identifies precisely these points as key levers: data quality, legal certainty, and a system landscape that can be seamlessly integrated. Those who invest in these areas gain a clear competitive edge.

An Overview of Other Change Management Models

Roboterhand und menschliche Hand greifen vor blauem Hintergrund aufeinander zu
  • McKinsey 7-S Model: Considers seven interrelated factors: Strategy, Structure, Systems, Shared Values, Skills, Style, and Staff. Suitable for transformations involving multiple companies or locations; often too broad for a single project at a medium-sized company.
  • Kübler-Ross Stages of Grief: Originally developed in grief research, this model describes emotional stages such as shock, denial, bargaining, acceptance, and integration. It helps leaders view resistance as a normal phase rather than a warning sign.
  • Burke-Litwin Model: Analyzes how external factors—such as the market or regulation—impact the organization all the way down to individual performance. Highly analytical in nature and intended primarily for corporate structures with multiple levels of management.
  • Agile Change Management: Operates in short cycles with ongoing feedback rather than a fully planned end-to-end process. Suits companies that implement ERP modules incrementally rather than planning everything around a single deadline.
  • Prosci 3-Phase Approach: The organizational framework surrounding ADKAR: Preparation, Management, Reinforcement. In practice, it is usually used in conjunction with ADKAR, not as a standalone model.
  • Management by Objectives (MbO) and OKR: These are actually leadership models, not change management models in the strict sense. They are useful after ERP implementation to embed new processes through clear objectives.

McKinsey: By 2028, one-third of software applications will use agent-based AI

McKinsey reports that early adopters of AI-integrated ERP environments are already achieving EBIT improvements of 5 percent or more. Forrester expects that about half of ERP vendors will introduce their own governance capabilities for the autonomous operation of AI agents by 2026. According to McKinsey, nearly one-third of all software applications are expected to include agent-based AI capabilities by 2028. Gartner goes even further: By 2029, at least half of knowledge workers are expected to have developed new skills in working with AI agents, similar to what happened with spreadsheets twenty years ago.

And what does that mean for a cloud-based ERP system like Haufe X360?

With Release 2026R1, Haufe X360 is specifically expanding its feature set to include its own AI-powered data analysis, which identifies patterns in existing ERP data without requiring anyone to build a query beforehand. An additional AI Assistant has been announced. However, a large part of what saves time in Haufe X360 today has been running reliably and automatically for years and is deliberately not labeled as AI—because it simply doesn’t need to be.

In the financial module, the system automatically processes e-invoices in ZUGFeRD and XRechnung formats, validates them against the standard, and archives them in compliance with GoBD. In inventory management, demand planning handles reorders based on inventory levels and consumption data. In sales, the system automatically calculates shipping costs; in the service module, maintenance processes run automatically; and in project management, Haufe X360 continuously compares target and actual costs. This automation is rule-based, not agent-based, and that is precisely why it has been in stable use for years.

A point of integration for AI functions arises when partners utilize the platform’s open API architecture themselves. Via REST and SOAP interfaces, Haufe X360 can be connected to Microsoft Teams, Power BI, HubSpot, and logistics partners such as DHL. BOLD & EPIC Transform has developed its own mailbox agent based on this architecture, which automatically captures documents from the email inbox and transfers them to the system instead of requiring manual entry. This demonstrates just how much more is already possible today based on the open architecture, regardless of the next Haufe release. Experiences from medium-sized businesses with the platform confirm this.

Will ERP soon be entirely agent-based?

In short: probably not, at least not in the next few years. Even if the tools were theoretically capable of this, the majority of companies still lack the appropriate data infrastructure, the necessary system integration, and/or the right governance model. Furthermore, bookkeeping entries, invoices, and reports submitted to tax authorities are subject to documentation requirements such as the GoBD. A system that makes decisions here completely autonomously, without a human being able to understand the decision and correct it if necessary, creates more audit risk than it saves in time. 

Therefore, AI is best applied where tasks are clearly defined, recurring, and verifiable based on rules—not where business judgment is required. Reading and categorizing documents, comparing payment behavior with historical patterns, verifying supplier data during onboarding, and reconciling inventory levels with consumption data: these are tasks where an agent can take on a significant role because the results can be clearly verified.

The Inbox Agent from BOLD & EPIC Transform falls into the same category: It handles a narrowly defined, recurring task—namely, capturing and sorting documents from the inbox—and can therefore be reliably deployed today, without having to wait for a fully agent-based future, which will take even longer to materialize for core processes that require verification.

To find out whether a comparable agent would be worthwhile for your own processes, simply schedule a brief consultation with BOLD & EPIC Transform.

Contact Us Now

FAQ on AI in ERP Systems