Digital Twins:
What They Are,
Where They Deliver,
How to Scale.
A digital twin is a connected model of a physical asset, process, or system — updated by real data, used to make decisions you couldn't make from the physical world alone. The word that matters is connected. A static 3D model isn't a twin. A CAD file with no live data feed isn't a twin.
PepsiCo already has a number on this — the Gatorade pilot settled the 'does it work?' question. What's open now is how it scales, and where it goes next.
Vendors apply "digital twin" to everything from a live sensor dashboard to a full physics simulation. That range matters — the investment, the data requirements, and the realistic benefit are very different at each end of the spectrum. Being clear about where a given implementation sits is the first honest conversation any programme needs to have.
A proven foundation — and the scaling question ahead.
The CES 2026 pilot with Siemens and NVIDIA produced real numbers, not a concept. It also happened at one well-instrumented facility. Turning one pilot into a global programme across beverages, snacks, and multiple geographies is a different problem.
Gatorade Plant Twin
Physics-accurate plant twin built with Siemens and NVIDIA, creating a real-time virtual replica of the entire facility — equipment layout, conveyor routes, operator paths, and material flow. Used to test changes before physical implementation.
How widely has this model been replicated across other lines and sites since the CES announcement?
Beverages, snacks, Quaker, Frito-Lay — these aren't the same type of manufacturing, and they're not at the same digital maturity. What was achievable at a Gatorade filling line in the US in three months is a different problem to an older snack plant in an emerging market. A scaling plan that treats the estate as uniform will stall at the pilot stage. The decisions that matter now are data standards, platform choices, and what the replication model actually looks like — not the technology in the flagship site.
Four value areas — all grounded in F&B evidence.
The numbers that exist come from specific implementations, not industry averages. Below is where F&B manufacturers have actually demonstrated returns — with PepsiCo's own programme as the reference where available.
Throughput & OEE
20% throughput increase at the Gatorade plant.
Physics-based simulation enables manufacturers to test layout changes, reconfigurations, and scheduling decisions before physical implementation — identifying problems before commissioning rather than discovering them on a live line. PepsiCo reported that up to 90% of potential issues were caught before implementation.
Where does unplanned downtime sit as a cost and risk concern — and is the current approach reactive or moving toward predictive?
20% throughput increase, up to 90% pre-implementation issue identification (PepsiCo reported, CES 2026)
Asset-level twins with anomaly detection consistently reduce unscheduled downtime across available F&B case studies
Process twins model line throughput — revealing where capacity is constrained before layout decisions are made
Where does most of industry actually sit? Closer to Level 1 than the case studies suggest.
Most of what gets published as a digital twin case study in F&B is a research project, not a production deployment. Most manufacturers at scale are at Level 1 or 2. The vendor case studies represent Level 3 at specific assets or lines — not plant-wide programmes. The Gatorade pilot is ahead of most of industry.
Level 1: Monitoring
Sensor data into dashboards. One-way flow, no model. Useful — but calling it a twin is a stretch. Most plants have some version of this already, often without knowing it.
Many mid-size F&B manufacturers are here. They have historians, SCADA dashboards, some IoT connectivity. Most vendor conversations that start with 'we want a digital twin' are actually starting from Level 1.
MFGPro+ and the Manufacturing Control Tower sit here — real-time production visibility across the network, connected to fleet and distribution. Solid foundation.
Existing instrumentation, a historian, basic IoT connectivity. Most plants have some of this.
Why most programmes stall: Most programmes stall on data, not technology. A twin fed unreliable or siloed data gives unreliable outputs. Fixing that before you start modelling isn't interesting work — but it's where the programme usually lives or dies.
An advisory role — helping you make the right calls before you commit.
DTG's role here is advisory. We work client-side — helping you think through scope, priorities, technology choices, and business cases, independent of any vendor. The decisions about what to build and who builds it stay with you.
The questions that matter now — what to tackle next, which platform, which sites, how to make the business case for scale — are programme design questions, not technology questions. That's where an independent advisory view is most useful: before positions harden and vendor commitments are made.
Prepared for discussion · Confidential · 2026