ERP in the Age of AI: How Enterprise Systems Are Changing

When enterprise software first appeared, it was a big step forward: it brought structure to business processes and made them faster and easier to track. Even so, the systems themselves stayed complicated. They needed training, customization, and constant human involvement. With the arrival of generative AI, that’s started to change.

How exactly will AI change the role of enterprise systems, and which companies will be able to bring it into their infrastructure without losing control of the business? That’s what Digital Business discussed with Andrey Treshchuk, CEO of Impace Group.

For more than 20 years, Andrey was a senior executive at the Canadian company TerraLink, one of the major players in the global IT and business consulting market. After selling his stake in TerraLink, he launched a new international group of companies that works with SAP and other makers of widely used business IT platforms across Europe, Central Asia, and the Middle East.

“The complexity of large systems hasn’t gone away, but the path to results is getting much shorter”

— You’ve been working with ERP implementation for many years. Over that time, what has changed the most in these systems?

— In a little over three decades, ERP has grown from “a system for record-keeping” into the digital backbone of an organization. It holds together processes, control, the history of actions, and the business logic behind company rules. And that backbone is not going anywhere. What will change is the way people interact with it. How they use it, and how they develop it.

AI is already changing these processes. Requirements analysis, project documentation, scenario write-ups, interface setup, test cases, user guides, and first-line support organization — a large share of this routine work is already being done faster. The complexity of large systems has not disappeared. But the path to results is becoming much shorter.

— What are the limits of traditional ERP systems?

— ERP works best in a world where everything is laid out in fields, roles, and workflows. But business rarely exists in such a neat environment. In real life, there are always exceptions, overlapping responsibilities, conflicting interpretations, and pieces of knowledge that live not in the system, but in people’s heads.

— How ready are today’s ERP systems, including the new generation of SAP, to work with AI?

— The market leaders already offer assistants, prompt generation, natural language search, and AI-based scenarios.

If we take SAP as an example, this is no longer just theory. You can already see it in real products: SAP Joule shows how AI is becoming part of the user’s day-to-day workspace, while SAP Ariba has long shown that the value of an enterprise system today is created not only inside a local ERP core, but also across shared cloud-based networks of processes and data.

To be honest, when we talk about AI readiness, the real question should be aimed less at software vendors and more at leaders in big business and government. The market loves talking about AI models. It is much less keen to talk about master data, who actually owns the data, the patchwork of integrations, and all the old internal compromises still holding things together. But that is exactly where the real complexity shows up.

There’s one more point that many people still underestimate. A company that avoids cloud solutions is making its own path to AI much harder. In theory, you can build on-premise setups. In practice, that almost always means a slower pace, heavier integration work, higher support costs, and less access to AI services that are evolving fast. If a company is not using the cloud, adapting AI quickly becomes much more difficult.

How AI is changing ERP systems

— What has AI already changed in enterprise systems?

— The first and most obvious shift is that ERP is no longer the only interface. In the past, users had to know exactly where to go, what to open, which report to run, and who to ask for context. Now they are increasingly starting with a simple, natural question: what is happening with inventory, why margins are down, where procurement is off track, and what contract risks need attention.

Second, the market is flooded with assistants. They are already built into almost everything: office suites, ERP, CRM, and industry-specific systems. And they are changing user habits faster than companies can rewrite their internal rules.

Third, the whole logic of implementation is changing. Until recently, projects would spend months gathering requirements, mapping out scenarios, agreeing on screens, and preparing user documentation. Now a big part of that routine work is moving much faster.

— So individual pilots are not yet proof that the technology really works at company scale?

— Almost every large company already has a story along the lines of, “We gave the team an assistant, and it saved them ten hours a week.” Great. Then comes the less exciting part, and that is where you find out whether the project has a real future or ends up in the demo graveyard: where the data comes from, who is responsible for its quality, how logging is handled, what happens when exceptions come up, who signs off on the result, and who carries the risk when something goes wrong.

— Can AI turn ERP from a record-keeping system into a decision-making system?

— The market is already moving in that direction. The only thing is, the word “decision” needs to be used carefully here. AI can explain, recommend, trigger an action within a narrow set of rules, and in some cases try to manage a business-critical process.

Over the next few years, a different scenario is likely to become standard: AI will get faster at pulling together context, suggesting options, and carrying out limited actions where the cost of a mistake is manageable. The key word here is “limited.”

— Where is the line between AI recommendations and real involvement in decision-making? When can a company start trusting it with more critical processes?

— The worst approach is to go to either extreme: keeping AI away from everything, or expecting it to run the whole show right away. Mature companies take it step by step. First comes observation and explanation. Then recommendations. Then simple actions. After that, limited autonomy in clearly defined scenarios.

When big money is involved, two things matter most: the ability to reverse an action, and the ability to understand why it happened in the first place. Without those, no “smart agent” should be let loose in a real business.

— Where, in practice, are companies seeing the biggest economic gains from AI adoption?

— The first major gains come where AI removes friction between functions. Finance, procurement, logistics, manufacturing, and sales all operate by their own logic. A huge amount of time and money goes not into the work itself, but into getting those functions to coordinate.

The second big gain is early problem detection. Errors, duplicates, anomalies, suspicious deviations, odd movements across the supply chain. The sooner the system brings those issues to the surface, the less it costs to fix them.

The third is speed. The gap between a question and action is often too wide. AI works best where that gap can be cut down dramatically.

— What is stopping companies from making full use of AI inside ERP?

— Companies love to say, “We have a data problem.” It sounds technical and almost never offends anyone. But what usually sits behind it is something else: the company does not have a shared version of reality. Different departments are operating in different worlds. As long as people are dealing with that, the conflict can be papered over in meetings. Once AI is layered on top, the whole setup starts to show its cracks.

There is also a second issue, and it gets talked about far less. If a company does not have an official AI setup, it already has an unofficial one. Employees have long been using external LLMs for emails, analysis, translation, summaries, and brainstorming.

— What should a business do if employees are already using AI in a scattered, unstructured way?

— Give people a legal, manageable way to use it. A company needs a single gateway to AI. That solves several problems at once: security, cost control, consistency of results, and one more thing people rarely talk about, which is respect for employees. When a company tells people, “Don’t use it,” it is pretending AI is not already part of the job. When it gives them a proper tool, it acknowledges reality and starts managing it.

— Why not build your own LLM in that case?

— That is one of the most expensive fantasies in the market. Building a general-purpose in-house model for every possible use case may sound appealing, but in practice it usually turns into a costly, hard-to-manage operation with a very shaky business case. It is expensive, complicated, demanding to maintain, and in the end it is usually weaker than the best external models.

At the same time, local models for narrow, clearly defined tasks are a perfectly workable option. Where there is a closed process, a limited set of documents, a strict format, and a high level of sensitivity, they can make sense. But only as a tool built for a specific process, not as a company-wide dream of one model that does everything at once.

“Many companies in Kazakhstan still try to keep everything on-premise and remain cautious about cloud solutions”

— In which of the regions you work in is the transformation of enterprise systems moving faster?

— Each region has its own strong side. Europe often stands out for its discipline. There is usually a stronger culture of methodology, compliance, and project delivery. The Middle East moves on the strength of faster decision-making and bigger ambition. If leaders see a clear strategic case for a major shift, they are often ready to back it quickly.

Kazakhstan is a country where a large share of the economy is made up of extractive industries and other capital-intensive sectors. These industries tend to be more conservative. They rely on long-term investment, and the cost of mistakes is high. That makes it hard to approach digital transformation in a “lab-style” way. Big business here respects people who can talk about technology without sounding naive.

Where a single company’s stability affects a meaningful share of exports, the regional budget, and social commitments, no one is going to rebuild the architecture just because a new buzzword is in fashion.

There is one more very practical constraint. Many companies in Kazakhstan still prefer to keep everything on-premise and are cautious about cloud solutions. At the previous stage, that made perfect sense: it meant control, security, and a familiar model of infrastructure ownership. But in the AI era, that mindset is starting to do less to protect a business and more to limit it. The next wave of technology will almost certainly require companies to be more open to cloud services, shared platforms, and a much faster pace of change.

— Do Kazakh corporations have a real chance of moving to an AI-first architecture?

— I don’t think there are any built-in barriers to that. But I would not recommend rushing into it.

A sudden jump to AI-first looks great on presentation slides. But I would start with a few basic questions: is the company actually ready for an AI-first architecture? Does it have a clean transactional core? Are data owners clearly defined? Does it have a proper digital identity setup, access control, and solid integration across its systems? And, most importantly, can you connect intelligence to that environment safely without opening up ten new vulnerabilities along the way?

And this brings us back to the cloud again. If a company has spent years deliberately avoiding the service model and keeping everything fully on-premise, bringing AI into the mix quickly and cost-effectively will be much harder. Not impossible, but harder, slower, and more expensive. So for many companies, the next step will be moving to a new level of architectural maturity.

What future awaits ERP systems

— Which functions will remain within ERP, and which ones are likely to move elsewhere in the coming years?

— What will stay in ERP is everything tied to the company’s source of truth: master data, transactions, access rights, workflows, action history, compliance, and the rules that govern each process. In other words, everything that ultimately feeds into audit and reporting.

What will start to fade is the friction: digging through menus, pulling standard reports by hand, some of the instructions, and some of the routine requests. Along with that, a big part of the user frustration that for years was treated as just the normal price of working with an enterprise system will begin to disappear.

People will use ERP in a different way. The question will come first, and the menu second. For many tasks, the menu may disappear from the workflow altogether.

— What will the ERP of the future look like?

— As a dependable core inside a smarter system. SAP and other ERP platforms will remain at the center of the enterprise stack for a long time. They handle the money, record obligations, manage procurement, track assets, and close financial and operational periods. This layer is far too important to turn into a testing ground.

But around that core, a new layer is growing fast. A layer of intelligence, conversational interfaces, agents, and cross-system context. People will think less and less about which system the answer lives in. What they will care about is something much simpler: what’s going on, and what do we do next?

That is where the real shift happens. ERP will stay, but the need to put users through the wringer just to get the most out of it will disappear.