"That's insane!" and the fish tank: Data governance is an AI accelerator, not a hurdle

10 hours ago

5 min read

ADSOTECH · NORDIC USER GROUP 2026

Espoo, Finland - 5.10.2026 Late in the afternoon at our Nordic User Group, Timo Ylikoski opened Claude, Anthropic's AI assistant, and asked it in plain language to update material master records, the core product data that SAP uses across purchasing, production and sales. Claude made the changes directly in Adsotech's own live SAP S/4HANA environment, in front of a room full of people who manage SAP master data for a living. There was a short silence. Then someone in the audience shouted "That's insane!"

Fish tank next to a tablet showing a presentation at an Adsotech and Precisely event, illustrating data governance, with the text "AI works with what you give it. Is your data clean enough to hand over?"

That reaction can mean two things. It can mean "I want this on Monday." It can also mean

"Why would I let that happen?" Our guess is that most people in the room felt a bit of both.

What Timo actually showed

The demo ran through Precisely's newly released MCP server. MCP stands for Model Context Protocol, an open standard that lets AI access other systems and engage with their feature set, like a human would. In practice, you describe what you need in plain language and AI carries it out in SAP, through the same controlled channels a person would use. In Timo's case, that meant Claude reading and updating material master records in our own SAP environment within minutes while the room watched.

For anyone who has spent years building spreadsheets, scripts and approval routines

around SAP data, seeing that happen in just a moment is a lot to take in. Work that can take

an entire day or several can now be completed with a few simple prompts.

The same day started with a fish tank

Earlier that morning, our CEO David Nelson told a story from his years as a kindergarten

teacher in his twenties. One of his jobs was cleaning the class fish tank. He could spend

hours getting it spotless, and after lunch there would be a Barbie doll or a lump of

playdough back in the water. Every time.

His point was about data. You can clean out the garbage once. If nobody owns the domain

and the process that keeps putting garbage in, it comes back.

Minna-Liisa Siltala from Orion Pharma made the same point from the pharmaceutical side.

"Data governance is not a one-time project. It is your daily responsibility." It never reaches a

finish line, and in a regulated industry, or any industry for that matter, the cost of treating it

as a one-off project shows up fast. Especially when AI is applied.

The general view Adsotech and our customers found alignment with was: before you

automate, look at the process, and look even harder at the shape of the data underneath it.

Garbage in, garbage out, only faster

Put those two moments side by side and the friction is easy to see. An AI assistant that can

update SAP in seconds will work just as quickly with a duplicate vendor, an outdated

material group or a missing field. Speed multiplies whatever you feed it. Clean data gets you

good work faster. Messy data gets you mistakes faster, and in more places at once.

Eric Kimberling of Third Stage Consulting, an independent ERP and transformation advisor, puts it bluntly in his guide to preparing for AI: "AI is only as powerful as the data feeding it." He warns that bad data leads to "costly mistakes that compound at scale," and that data quality needs governance that keeps going long after the first cleanup. A watchful eye and dedicated crew over the fish tank, in other words.

Where expected returns are lost

This is also where a lot of AI money disappears. Deloitte's 2025 survey of 1,854 senior

executives across 14 countries, including Denmark, Norway and Sweden, found that 85%

had increased their AI spending in the past year. Yet the typical payback was two to four

years, against the seven to twelve months the same executives expected from ordinary

technology. Only 6% saw a return within a year, and nearly half named inadequate

infrastructure and data as a barrier.

BCG's 2025 study of more than 1,250 companies tells a similar story. Only 5% were getting

value from AI at scale, and 60% reported minimal gains despite substantial investment.

Among the obstacles, 68% named a lack of access to high-quality data. BCG's own rule of

thumb is that about 10% of a successful transformation is algorithms, 20% is technology

and 70% is people and processes.

Most AI budgets are built the other way round. The tech purchase and the pilot get funded.

The data ownership, cleanup, change management and required process work get squeezed in as an afterthought. Kimberling's central argument about AI projects clashes with this status quo, and for a reason that should resonate with everyone. As he put it: "AI success is a human problem first, a technology problem second."

Governance IS the accelerator

It would be easy to read this as a tug of war, with speed pulling one way and governance

pulling the other. What we are actually seeing points to the two pulling in the same

direction. Governance is what makes the speed pay off.

In practice that means a few unglamorous things. Every data domain needs a named owner. The rules for what good data looks like need to be well established and integrated into systems and workflows. Data needs to be validated before it reaches a production ERP

environment, changes to master data need an approval step, and there needs to be a record

of who changed what and why. Sadat Ahmed from Precisely showed the new Precisely

platform, the Data Integrity Suite, and where this is heading: data checks and cleansing that used to happen after the fact, turned into a clearly owned, defined and governed end-to-end process that runs on repeat.

With that in place, the raw potential that made someone shout "That's insane!" becomes

something you can actually use on a Monday.

A short note on using AI safely: compiled expert advice

If you are starting to connect AI tools to your ERP environment, a few habits go a long way.

Know which data the AI can see, and keep classified data out unless you have a clear reason

and a legal basis for it. Try things in a sandbox before production. Keep a person approving

changes to master data, and keep an automated audit trail to stay compliant. Since February 2025, the EU AI Act (Article 4) has also required organizations that use AI to make sure their staff understand it well enough to use it responsibly. That is a question of change

management as much as a legal one.