Saltar al contenido

Tecnología · Artículo

Your old ERP is about to become more valuable

The usual story says AI replaces legacy systems. For industrial companies the opposite is closer to the truth. An agent is only as good as the record it can check itself against, and that record usually lives in the ERP.

Por Redacción de demi5 min de lectura

Todas las publicaciones

Esta publicación aún no está en español. Aquí está la versión original.

Many strategy slides show the same order: first the new ERP, then AI. It sounds tidy. Clean up first, then modernise. In practice it often means three years of migration while AI waits, and then a new system in which the old questions come back.

This piece argues the opposite. AI does not make the existing ERP worthless. It makes it more valuable. Not because it is modern, but because it has something no language model brings along: a record the company has agreed on.

What an agent really needs

Picture an agent preparing a quote. It reads the request, finds similar past orders and suggests a price. How does it know the part number is right? That the customer has framework terms? That the lead time for this material can't be met right now? It can only check if there is one place where these things are binding.

Most medium-sized and large companies already have that place. In 2025, 46.45 per cent of EU enterprises with at least ten employees used ERP software, 69.93 per cent of medium-sized and 88.71 per cent of large ones (Eurostat). Order, item, price, customer, supplier: in the ERP they are not just stored, they are agreed. A model can phrase things. It can only check against something that is right.

The why is not in the ERP

At the same time, the ERP holds only the what. Which tolerance applies, which supplier is approved, which price was agreed. The why lives elsewhere: in drawings, test reports, change notices, emails with the customer, notes from the 2019 complaint.

That is exactly where AI works today. In 2025, 20.0 per cent of EU enterprises used AI, and the most common use was analysing written language, at 11.8 per cent (Eurostat). So AI mostly reads text. Its value for industry appears the moment that text is tied to records: this test note belongs to this item, this email to this order, this change to this drawing revision.

An everyday example: a customer asks why a part has been packed differently since the last delivery. The ERP shows the new packing instruction and the date of the change. The reason, a transport complaint about damaged sealing surfaces, sits in an email to the carrier and in the complaint's closing report. If you want to answer the customer in ten minutes instead of two days, you need both halves, and the link between them.

An agent that only reads documents guesses. An agent that only reads the ERP doesn't know why. Only the two together give an answer you can check.

The fear of incompatibility points the wrong way

Companies' concern is real. Of German firms without AI that had already considered using it, 45 per cent named incompatibility with existing devices, software and systems as a reason, and 44 per cent named problems with the availability or quality of data (Destatis).

Many conclude that the old system has to go first. But the numbers say something else. They say AI has to reach the existing systems. That is a question of reading, not replacing. An ERP that has run for fifteen years often has ugly screens and odd fields. But it holds fifteen years of real transactions. A migration rarely carries all of that history across.

Two projects, the same people

For many, there is also a calendar question. Mainstream maintenance for SAP Business Suite 7 runs until 31 December 2027, optional extended maintenance until 31 December 2030 (SAP). So many companies face two decisions in the same window: what happens to the ERP, and how do we start with AI?

Both projects need the same people: the few who know what each field in the old system means and why each exception was built. Start both at once and two projects fight over the same experts. A smarter order secures the knowledge first: make data and documents readable, record how they connect, and then decide what to migrate. That work is due in any migration anyway, so it pays twice.

This does not make an ERP change wrong. But it gets easier once it is clear which data is actually used, which fields are only historical and which documents hang on which records. Then you migrate knowledge, not just tables.

The investment no balance sheet shows

Economists describe a pattern that comes with every general purpose technology. It requires large complementary investments, for example in new processes and skills, which are mostly intangible and poorly measured in the national accounts (NBER).

The link between documents and records is exactly such an investment. Which test reports belong to which item, which drawing applies to which order, which email explains which price deviation: none of it is on a balance sheet. But it decides whether an agent gives an answer with a source or a plausible guess.

The legal frame points in a similar direction, at least for machine data and cloud services. The Data Act has applied since 12 September 2025. It is meant to give users more control over the data their connected devices generate, industrial machinery explicitly included, and to make it easier to switch between cloud providers (European Commission). It says nothing about ERP systems. We still read it as a signal: data should be able to move without the system having to be replaced.

Our assessment

An assessment, not a fact: in 2027, the question in ERP projects will shift from "which new system?" to "how fast can agents read our existing records?". Read access and clean master data will then be valued above new modules.

The practical conclusion is unspectacular. Keep the ERP boring and correct. Make everything around it readable. And connect the two so that every answer can say where it comes from.

With demi, your systems stay as they are. If an adapter for yours is missing, we build it with you.

Fuentes

  1. Eurostat, ene 2026: Digital economy and society statistics - enterprises
  2. Eurostat, 11/12/2025: 20% of EU enterprises use AI technologies
  3. Statistisches Bundesamt (Destatis), 2025: Nutzung von IKT in Unternehmen (Erhebung 2025)
  4. SAP News Center, 07/10/2026: Planning Beyond 2030: What SAP Business Suite 7 and SAP NetWeaver Customers Using Third-Party Databases or Java Need to Know
  5. National Bureau of Economic Research (Brynjolfsson, Rock, Syverson), oct 2018: The Productivity J-Curve: How Intangibles Complement General Purpose Technologies (NBER Working Paper 25148)
  6. European Commission, 05/11/2026: Data Act

Tu configurador

Aún no hay casos de uso

Tu configurador todavía está vacío

Añade los casos de uso que necesitas: desde cualquier tarjeta o página de caso. Aquí verás lo que aporta cada uno al año, con el cálculo.

Ver casos de uso