I'm heading back to Bentonville next week for the Fuse Ventures supply chain accelerator, so supply chain is what has been on my mind.
Spend time in those rooms and you hear a version of the same question over and over. It comes from carriers, from retailers, from food companies, from anyone who moves physical goods at scale. It sounds something like this:
A storm is forming over the Southwest. Which replenishment cycles break in the next seventy-two hours, and which suppliers behind those cycles are already running thin?
The people asking are not short on data. Every piece of it exists inside their four walls or one API call away. Weather feeds. Lane capacity. Facility locations. Inventory positions. Supplier records. Carrier schedules.
The data is there. The answer is four hours of engineering away, and by then the storm has landed.
That gap is the whole business we are in.
The last decade of supply chain technology solved tracking. You can see where the truck is. You can see the container, the pallet, the carton, the SKU. Dashboards everywhere.
Tracking tells you where something is. It does not tell you what happens next, or what else is exposed, or who upstream is the real cause.
Answering those questions means crossing systems. The weather feed does not know what a lane is. The TMS does not know what a supplier is. The ERP has never heard of a storm. Each system holds a fragment of the world and models it in a private language.
So every cross-system question becomes a project. Somebody writes a pipeline. Somebody stages a copy. Somebody reconciles two identifiers that were never meant to meet. Weeks later you get an answer to a question that stopped mattering on day three.
Pointing at silos is the easy part of this conversation. Everyone in the industry can already draw that picture on a whiteboard, and the serious operators stopped drawing it years ago and started building.
FedEx Dataworks is the clearest public example. They have said plainly that the goal is to fuse physical and digital signals into coordinated action across the value chain. That is the right ambition, stated better than most people state it. Other large carriers and retailers are funding versions of the same effort right now, with real money and real engineering talent behind them.
So the interesting question stopped being what to build. It is why this stays hard for organizations that are neither confused nor underfunded.
I think there are two possible explanations, and both of them sit at the level of architecture rather than effort or budget. Architecture is the hardest thing to change once a system is carrying real traffic, which may be why capable teams keep arriving at the same wall holding different tools.
A supply chain is a set of things in specific places changing over time. A shipment has a position and a clock. A lane has a geometry and a capacity that varies by hour. A storm has a footprint that moves. A port has a queue that grows. Every meaningful supply chain question is a question about space and time together.
Warehouse platforms are built around tables and bolt geometry on afterward. Graph databases model relationships beautifully and treat both space and time as attributes you manage yourself. GIS tools handle space natively and treat time as a column.
We spent five years building Geodesic native to space and time from the first line of code. When both are in the model from the beginning, a question about a storm and a question about an inventory threshold turn out to be the same kind of question, and the system can answer them in one pass.
The second answer is quieter and matters just as much.
In Geodesic there are no privileged tables. A storm, a pallet, a distribution center, a tier-three supplier, a lithium mine, and a replenishment threshold are all things. What connects them is a statement about two things and the relationship between them. Subject, predicate, object. That is the entire vocabulary.
It sounds almost too plain to be a differentiator. The consequence is large. When everything is represented the same way, adding a new kind of entity does not mean a schema migration and a quarter of integration work. It means new nodes and new statements in a model that already knows how to hold them.
That is what lets a weather signal sit in the same world as an inventory threshold and a supplier's parent company, with no translation layer in between. And it is why a question can cross domains that were never designed to meet.
Three ideas, and they are simple to state.
One ontology that speaks supply chain. We model the physical world with a small set of entity types operators recognize immediately: facilities, parties, products, logistics units, lanes, events, processes, and the outside signals that act on all of them. A storm is a first-class citizen in that model. So is a distribution center's replenishment threshold.
Federation instead of ingestion. Geodesic connects data where it lives. Internal systems, partner APIs, outside feeds like weather and congestion and business hierarchies, all queryable as one coherent model with no migration and no copy. Adding a data partner becomes a connection instead of a project. This matters most in multi-party ecosystems, where your suppliers and merchants will never agree to move their data into your lake.
Traversal. Once the world is modeled coherently, a question walks the graph. Weather signal to lane to facility to inventory threshold to store-level risk. Or start at a mine and walk forward: raw input to cell manufacturer to component to finished device to carrier lane to shelf. Then run the same walk backward and ask what else in the portfolio shares that upstream exposure. That second answer is usually the one that changes a decision, and it is the one nobody has time to look for today.
Four hours becomes ninety seconds. And you can run ten variants of the scenario at once, because that is what an operator actually wants at 6am with a storm coming.
Every large logistics organization is now building agents. Agents that reroute, reallocate, reprice, and reschedule. That is the right ambition.
An agent is only as good as the picture of the world it reasons against. Point an agent at a fragmented, stale, semantically inconsistent data estate and it will act with total confidence on a world that does not exist. At network scale, a confidently wrong answer is not a dashboard error. It is trucks in the wrong place and product on the wrong shelf.
Coherent context has to exist before any model sees it. That layer is the work.
Geodesic is a living model of how your physical operation actually works. It gets richer with every connection, every question, every day it runs. Provenance travels with it, so you can always ask where a fact came from and who touched it.
Two years in, an organization has built something a competitor cannot reconstruct quickly, because the relationships and the history are the asset.
We are not asking anyone to rip out their stack. We connect to what you have and make it answer questions it has never been able to answer.
If your team can name a question that takes four hours today, that is where we start.