Categories
Last week, I spent time in Nashville attending and speaking at NextGen Supply Chain Conference. NextGen is a conference about moving material and data. The people in the room run warehouses, fleets, ports, plants and distribution networks, and their technology stacks reflect that. ERP, warehouse management, transportation management and order management are categories that have been in production for decades, and most of these buyers default to buying them off the shelf. When speakers talked about specific versions of Manhattan, most of the room followed the shorthand without explanation. That maturity matters, because it means new value enters at the edges of the core systems: truckload optimization, inventory accuracy, traceability, and now AI.
Both days of the conference were packed with insights. A few standout sessions that really hit the mark:
- Wayfair's compelling keynote on the future of home delivery,
- Tractor Supply’s masterclass on leveraging supply chain as a growth engine,
- Penske’s perspective on what changes and what endures,
- Fanatics on agentic AI in demand forecasting,
- DP World’s practical blueprint on scaling transformation,
- A retail panel with a powerhouse lineup from Target, GXO, and Americold,
- Lastly, a deep dive led by NOAA on weather data for strategic long-term supply chain planning.
What I want to share are the conversations that stayed with me afterwards. Five of them, in particular, point in the same direction.
1. DP World: The data center as a supply chain
Most people I spoke with knew DP World as a logistics name. Fewer knew that DP stands for Dubai Ports, or how far the company has moved into US logistics by acquiring operators here and growing them. The move that caught my attention is data centers. DP World has gone all out on this market. On October 20th, the week of the conference, it announced an expanded US data center logistics network that now supports close to 50 data center sites, with services running from inbound receiving and inspection through kitting, server configuration and testing, rack integration, white-glove transport, spare parts and reverse logistics.
The playbook is worth studying. Their head of data center strategy for the Americas put the logic plainly: a delay in a single critical component can push back the date new computing capacity goes live. ]That means DP World sells control of the whole path from origin to installation, as one program with one critical path. At the conference, Carey Boone, who leads transformation for DP World in the Americas, gave his fireside chat on aligning operating models, process discipline and talent capability. A port operator that becomes a data center logistics provider must change all three at once.
The data center buildout also explains a cluster of topics I heard across the two days. It creates demand for diesel generator sets, grid capacity, water filtration for cooling, HVAC, and land suitable for siting. Each of those is a supply chain of its own, and each one must land on the same construction schedule.
2. Cummins: Selling into large programs of work
Cummins sits right in the middle of that demand, since large installations such as data centers need generator capacity. My conversation with Pritam Chowdhury was about how they go-to-market on large programs of work tied to big installations, and how they see the data that runs through those programs. A program of that size involves the owner, the builder, the engineers, the equipment makers and the logistics providers, each running its own systems and each holding part of the picture.
What links Cummins and DP World is that both are selling into programs where many companies have to coordinate on one timeline. In programs like that, the scarce resource is shared information across company boundaries: which component is late, which site it affects, and who decides what happens next. Data sharing between ecosystem partners came up in more conversations than I expected, and I think this is why.
3. NOAA: The data nobody is using
The session I keep returning to was Jenny Dissen's. She leads retail for NOAA's Industry Partnerships program, which the Department of Commerce funded in 2024 to push NOAA data into three industries: retail, insurance, and architecture and engineering. NOAA's archive of environmental observations, which she puts at close to 100 petabytes, sits in Asheville, North Carolina, and is moving to the commercial clouds.
Hurricane Helene flooded that same city, and she watched it from her home by the river. The storm destroyed about 60 miles of CSX track, bridges and roadbed on the Blue Ridge line, one of the railroad's four north-south routes, carrying more than 14 million tons of freight a year. It took about a year and roughly $450 million to rebuild. CSX plans in 200-year horizons, and it had identified no climate risk for that corridor before the event. Her question to the room was how a company goes from no risk to blown away in a matter of months. Her answer was uncomfortable: the observations were never there, so the engineers had no signal to pick up.
NOAA has rebuilt its tools for people outside the science community. The Storm Events Database is the authoritative record that a storm did or did not hit a place on a given day, and it is what a company has to submit in an insurance claim or in litigation. The Climate Atlas lets anyone pick a region and see its temperature and rainfall history. The new Event Footprint Catalog draws the actual boundary of every tornado path and hail swath, refreshed daily. Retailers asked for it because they want to know how next week's forecast compares with every time something similar happened before, and what it cost them. She also passed on a line from retailers that stayed with me: they would take information that is 60 percent accurate if it arrived a month early.
Then she said the most useful thing I heard all week. NOAA cannot put its data into business context for anyone, and it cannot measure its own return on investment. It asked the room for partners to co-build products and write case studies. NOAA knows where the storm went. Only the company knows what was standing there: which distribution center, which three suppliers, which truck lane, and how much inventory was on hand.
Making that join is harder than it sounds, and the difficulty is in the meaning of the data. NOAA publishes storm polygons, county-level event reports, radar cells about a kilometer across, weather station readings and forecast grids. A business runs on sites, lanes, purchase orders and SKUs. Before anyone can say a supplier was exposed to a storm, someone has to decide what exposed means for a supplier, a lane and a store, and write that definition down so that planning, finance and the insurance team all use the same one. Joining forecasts with history adds another layer, because a forecast is a model's output and a past event is a recorded fact. That is a definitions problem, and it is the same one I heard in every other conversation in Nashville.
4. Apple: The impact of the data foundation on talent and culture
Piu Ghosh of Apple framed the people side of this better than anyone I heard. The talk was called From Planner to Orchestrator, and one slide set out what teams need for that transition under three headings.
The first was mindset, which she called curiosity over routine. Recognition shifts away from the person who runs a manual routine fastest and toward the person who prompts the best scenario explorations and stress-tests the results. The second was skills, which she called bilingual talent: people who speak the operations vernacular of cycle times, on-time delivery and yield, and who also understand data science concepts such as confidence intervals and latent variables. The third was culture, built on psychological safety, an environment where people are encouraged to interrogate machine recommendations and bring their lived operational nuance to them.
All three depend on something underneath them. Bilingual talent needs a shared vocabulary, which means on-time delivery has to be defined once, in one place, and mean the same thing to the planner, the data scientist and the agent. Interrogating a machine recommendation is only possible when you can see how the machine reached it, which data it used, how fresh that data was and which definition it applied. Another speaker on the main stage compared AI agents to the arrival of the personal computer and said the fastest companies will put AI in the hands of people across every operating department. I agree, and Piu's slide describes the people those departments will need. The data foundation has to give those people something they can question.
5. Toyota and Alan Amling: Data quality, semantic layers and ontology
My longest conversations were with Toyota and with Alan Amling, who spent 27 years at UPS, most recently running corporate strategy, and now teaches supply chain strategy at the University of Tennessee. Both kept returning to the same three subjects: data quality, the semantic layer, and the need for ontologies. Hearing that from a manufacturer known for operational discipline and from someone who ran strategy at UPS told me these ideas have moved out of the data team and into the operating conversation.
Everyone at the conference used the phrase data quality, and it covers far more than bad fields. It includes the same supplier recorded differently in two ERPs, two teams counting order lines with different rules, and timing mismatches, where one system updates weekly and another updates with every transaction, so any model built across them has to decide which version of the truth to use. That decision is an engineering judgment. Most supply chain organizations do not have enough engineers to make it for every question they want to ask.
This is where I find the language of ontology useful. If you draw the mind map of a business from the top, with customers, suppliers, plants, lanes, orders and regulators, it looks remarkably similar from one company to the next. When you bind that map to the data that supports real decisions, it becomes an ontology. At the bottom, every company is different: one plant runs SAP, another a homegrown system, and the tables look nothing alike. The semantic layer is what connects the two, defining measures and dimensions in business terms so that a person or an agent asking a question gets an answer grounded in agreed definitions, with a trail back to the source.
Alan sat through my own session, and his summary of it afterwards was that a use-case-first approach outperforms a data-first one. That is the argument I made. Most data programs stall because they try to fix every data quality issue across the enterprise before they deliver anything. In my session I walked through a global industrial company that had grown through dozens of acquisitions and was running several ERP systems, where bidding on large contracts meant reconciling part numbers, bills of materials and delivery milestones by hand. The team thin-sliced the problem: one large customer, one large order, one major supplier. They harmonized only the parts, milestones and bills of materials that slice needed, and the next use case reused what the first one had built.
Two points from that session are worth repeating here. Analytical decisions can start out 90 percent right and get sharper with use, while a transactional system needs every record correct from the first day, so the two deserve different standards. And data quality problems are best surfaced as you go, through the definitions in the ontology, where someone can see each one and decide whether it matters for the decision at hand. AI now helps with the slow part, suggesting descriptions for similar parts across bills of materials which cuts much of the manual effort out of harmonization.
What I would ask on Monday morning
Put these five conversations together and a single picture appears. The physical buildout is accelerating, led by data centers. The programs behind it involve many companies on one timeline. Valuable outside data such as NOAA's is free and underused. AI is moving into every department. And underneath all of it sits the same question of shared definitions.
So here is the question I would put to any supply chain team after Nashville. Pick one disruption you lived through this year: a late component, a storm, a supplier failure. Could you answer, from your own systems and in an afternoon, which orders it affected, which customers felt it, and who had the authority to act? If the answer depends on three people reconciling three spreadsheets, that is where to start.
Our whitepaper, From Sensing to Orchestration: Building a Supply Chain That Can Act on What It Sees, goes into more detail on how to build that foundation one decision at a time.