There is a concept in professional kitchens called mise en place, everything in its place before service begins. The best chefs do not start cooking and hope the ingredients sort themselves out mid-order. The prep happens first, deliberately and completely, so that when the pressure is on the kitchen can move at pace without chaos.
For some reason my brain decided to align that principle to the concept of data quality and its use in agentic AI. I finally figured out what caused the spark, Zero100's AI-Ready Data Imperative report which defines the material difference between data that informs a decision, data that triggers a recommendation, and data that an agent acts on autonomously (there is also a nice podcast on the topic). A dashboard built on imperfect data gets questioned by a human who applies judgement and context. An agent built on imperfect data acts, reroutes the shipment, adjusts the replenishment signal, commits to the supplier, before anyone has noticed the input was wrong. The same ambiguity that a person absorbs in a meeting becomes operational risk the moment a system is doing the acting.
The same research found that only 15% of organisations consider their data foundations execution-ready. Let that land for a moment, because that means 85% of the AI programmes being announced, funded, and celebrated right now are being built on ground that cannot support the weight being placed on them.
The honest reality is that the foundations have not been prepared for this level of trust, and most organisations are not yet having that conversation clearly enough to know it.
The instinctive response is to treat this as a reason to delay. Clean everything first, then deploy agents. That instinct is understandable, but most organisations do not have the luxury of acting on it. Board pressure is intense, Wall Street is rewarding AI narratives and punishing silence, and vendors are racing to rebadge everything as agentic. The result is a top-down mandate to move fast colliding with a bottom-up reality that the foundations are not ready. Waiting for the data estate to be fully clean is not a strategy, it is an indefinite deferral dressed up as diligence. I have yet to meet the enterprise that finished a data quality programme and declared victory.
So the real question is not how do we clean everything but what do we actually need to clean to make a specific decision trustworthy.
Zero100's research consistently finds that the list of data inputs that actually determine the outcome of any given decision is shorter than most people expect. Work backwards from the decision, inventory rebalancing, late shipment triage, supplier payment timing, and ask which fields, if wrong or late, make the agent's action wrong or late. Typically ten to twenty fields. They call these Golden Inputs, not because they are glamorous, but because they carry disproportionate weight and deserve disproportionate attention. Govern those fields with the same rigour applied to operational KPIs, and you have a foundation an agent can genuinely act on. Leave them in the same state as everything else, and you have an agent making fast, confident, wrong decisions.
Here is where the argument takes a turn that most conversations about data readiness miss entirely. In supply chain, the triggering event is almost always external. A supplier confirmation, a carrier delay, an invoice exception, a forecast revision. The decisions agents will be asked to make are not primarily driven by data that originates inside your four walls. They are driven by signals arriving from outside them. Which means the most consequential Golden Inputs in any supply chain decision are very often external data fields, not internal ones.
This matters because it reframes the entire data cleansing challenge. Internal data cleansing is valuable but it is a continuous battle, because dirty data keeps entering the estate faster than it can be cleaned. Every new transaction, every supplier update, every manual entry is a potential source of contamination. It is the culinary equivalent of trying to keep a kitchen clean while unchecked deliveries arrive throughout service.
Business networks change that dynamic fundamentally. Data arriving over a governed network is structured, validated, and clean at the point of entry. Those fields do not join the cleansing backlog because they never needed cleansing in the first place. Think of it as the difference between a kitchen that inspects every ingredient as it arrives from the supplier, and one that waits until service to discover the fish was not fresh. Quality is established before it enters the kitchen, not discovered as a problem halfway through cooking.
This is where SAP Business Network plays a defining role. With millions of trading partners already connected and transacting through structured, governed document flows, the network is delivering AI-ready Golden Inputs at scale today, not as a future state. Purchase orders, advance ship notices, invoices, confirmations, each carrying their critical fields, all arriving clean by design. For organisations building toward agentic AI in supply chain, that is not a nice-to-have, it is the shortest path between ambition and trustworthy autonomous action.
The path to agentic AI is not a transformation programme. It is a sequence of small, governed steps, each one earning the right to the next. Start with a single decision, identify its Golden Inputs, and rather than beginning an internal cleansing project of uncertain duration, ask which of those inputs arrives via a business network and is therefore already clean, already structured, already trustworthy. In supply chain that answer is often most of them.
The problem in most kitchens is not the quality of the chefs. It is that the ingredients keep arriving without quality checks, and the team spends half its time sorting through the delivery before it can start cooking. The best kitchens do not just prep well, they have rigorous supplier standards, so the ingredients arrive ready to use.
That is what business networks do for agentic AI. Not a better mop. A cleaner kitchen.
