Turning Plant Data Into a Product Carbon Footprint

Most plants can tell you their Scope 1 and Scope 2 numbers to two decimals. Ask the same plant for a cradle-to-gate carbon figure on the pallet of product leaving the gate, and the answer usually comes from a spend table: dollars purchased times an industry-average emission factor. That number is defensible for a screening inventory. It is close to useless for a customer who wants to decarbonize a specific supply chain, because it cannot tell two suppliers apart and it does not move when you actually change the process.

The pressure to replace that estimate with a measured one is now coming from downstream. A buyer's purchased goods sit in their Scope 3 inventory, and for most companies Scope 3 dwarfs everything they emit directly. CDP, which runs the largest corporate supply-chain disclosure program, reported that members' value-chain emissions run on average around 26 times their combined operational emissions (CDP, 2024). So a manufacturer chasing a credible target has no choice but to push into its suppliers' data. If you run a processing plant, you are that supplier, and the request landing in your inbox is for a product-level number grounded in what your meters actually saw.

This piece is about the data engineering between the historian and that number. Not the policy case for measuring Scope 3, which is settled, but the specific question of what plant process data you need, how to allocate it to a product, and how to score it so the figure survives a customer's auditor. I'll assume you already know the basics of the greenhouse gas accounting framework.

What a product footprint actually needs from the plant floor

A product carbon footprint is a life-cycle calculation, and the relevant standard is ISO 14067:2018, which sets requirements for quantifying the carbon footprint of a product consistent with the broader LCA standards ISO 14040 and 14044 (ISO, 2018). For the part of the life cycle that happens inside your fence — the gate-to-gate slice — the calculation is only as good as the activity data you can pull from the floor.

The GHG Protocol's product and Scope 3 guidance is explicit that primary activity data can come from meter readings, purchase records, utility bills, engineering models, direct monitoring, mass balance, or stoichiometry (GHG Protocol, 2011). In a plant, that maps onto signals you already have or could have:

• Energy by carrier. Electricity at the line or feeder, not just the main intake. Natural gas, fuel oil, or biomass to each boiler and furnace. Purchased steam, chilled water, and compressed air, metered at the point of use rather than estimated from a single site bill.
• Throughput. Mass or volume of saleable product per line, per shift, ideally from the same flow and weigh instruments that already feed the historian.
• Process gases and fugitives. Refrigerant top-ups from maintenance logs, CO2 used in carbonation or inerting, process vents. These are small in mass and large in warming potential, and they are the figures most often guessed.
• Materials in. Quantities and identities of feedstock, ingredients, and packaging, tied to the lot that became a given product run.

The resolution question matters more than people expect. A plant that meters energy only at the main breaker can produce a site total, but it cannot answer "how much of that belonged to product A versus product B." Sub-metering at the line, pulled over OPC-UA or Modbus into a historian, is what turns a site number into a product number. You don't need second-by-second data for a carbon figure — a footprint is usually reported annually — but you do need the energy and throughput streams aligned on the same time base so they can be summed over the same production window. A monthly gas bill against per-shift production records will not reconcile, and the gap is where the estimate creeps back in.

Alignment is also a sanity check, not just bookkeeping. If you sum metered electricity across every line and it doesn't land near the site intake, you have a missing load or a miscalibrated instrument, and you want to find that before the number reaches a customer. The same closure logic applies to materials: feedstock in, product out, and losses should balance within a tolerance you can defend. A plant that can close its mass and energy balance to a few percent has a footprint built on arithmetic; a plant that can't is publishing a guess with extra decimal places. Treat the reconciliation as a gate the data has to pass before it's allowed into the calculation.

Allocation is where plant data earns its keep

Single-product plants are rare. Most lines make several SKUs, many sites run shared utilities, and process industries routinely produce co-products from one operation. So the central problem is allocation: dividing the plant's measured emissions among the things it makes. This is also where a spend-based estimate quietly fails, because money spent tells you nothing about how a shared boiler's gas should split across three products.

ISO 14067 inherits the allocation hierarchy from ISO 14044, and the order is not optional (ISO, 2018). First, avoid allocation by subdividing the process — if you can meter line 1 and line 2 separately, you don't have to split anything, you just measure each. Second, where subdivision isn't possible, allocate on an underlying physical relationship, typically mass or energy content. Only as a last resort do you fall back to economic value, splitting emissions by the relative revenue of the co-products.

Read that hierarchy as an instrumentation specification. Every step down it is a step away from measurement and toward assumption, and the thing that keeps you near the top is sub-metering. A shared steam header feeding two unit operations is an allocation problem on paper. Put a flow meter on each branch and it becomes a measurement. The question worth asking on any contested figure is simple: can I move this allocation up the hierarchy by adding one meter? Often the answer is yes, and a single instrument retires an argument that would otherwise recur every reporting cycle.

Physical allocation also forces a choice of basis, and the basis has to fit the process. Mass allocation is defensible for a stream where products are genuinely interchangeable by weight. It is misleading where a low-mass, high-value cut carries disproportionate processing — energy allocation, splitting by the heat or electricity each path consumed, often tracks the real driver better. The standard lets you pick the relationship that reflects the physics, and the plant data is what lets you defend the pick.

Primary versus secondary, and the score that decides which

Not every input can be metered, and the framework doesn't pretend otherwise. The working distinction is between primary data — values from direct measurement or calculation based on direct measurement, drawn from inside the relevant process — and secondary data, meaning industry averages, proxy values, and life-cycle database figures. A real footprint is a blend. The skill is knowing which inputs justify the cost of measurement and which can ride on a generic factor without distorting the result.

The GHG Protocol's product standard gives you the tool to make that call honestly: a set of five data-quality indicators applied to each input (GHG Protocol, 2011):

• Technological representativeness — does the data reflect the actual technology and process route used?
• Geographical representativeness — does it match the location, with its grid mix and fuel supply?
• Temporal representativeness — is it current, or a number from a decade-old database?
• Completeness — how statistically representative of the real operation is the sample?
• Reliability — how dependable are the source, the collection method, and the verification behind it?

Scored together, those indicators feed a pedigree-matrix approach that ties each qualitative rating to an uncertainty range, so the footprint carries an explicit quality grade rather than a false air of precision. This is the part plant data changes most. A metered electricity figure from your own line, this year, on your own technology, scores at the top of every indicator. A national-average grid factor from a three-year-old database scores well on completeness and badly on temporal and technological fit. Same number of significant figures on the page; very different confidence. The pedigree matrix makes that difference legible to a customer instead of hiding it.

The Partnership for Carbon Transparency, run by WBCSD, has built its whole program around raising the share of primary data in these footprints, and its PACT Methodology reached version 3 in April 2025 with stronger requirements for data reliability and verification pathways (WBCSD, 2025). The direction of travel is clear: secondary factors are the starting point, not the destination, and the parts of the footprint a customer cares about are the ones you can show you measured.

Climbing the methods ladder for purchased goods

When your customer slots your product into their inventory, it lands in Scope 3 Category 1, purchased goods and services. The GHG Protocol's technical guidance lays out four calculation methods for that category, and they form a ladder from crude to precise (GHG Protocol, 2013):

The jump that matters is from the bottom two rungs to the top two. Spend-based and average-data run on secondary factors and need almost nothing from you. The hybrid and supplier-specific methods require real activity data from inside the plant, and that is precisely where instrumentation pays off. The guidance frames supplier-specific as the most accurate route because it relates directly to the actual good purchased and avoids allocation guesswork on the buyer's side. But "supplier-specific" is only as strong as the allocation and data quality underneath it. A product footprint built on a single site-level meter and an economic split is supplier-specific in name and barely better than average-data in substance.

This is the honest reason to instrument before you report. The methods ladder isn't a menu you pick from once; it's a description of how far your data lets you climb. Sub-meter the lines, capture the process gases, align the time bases, and you can offer a hybrid or supplier-specific number your customer can actually use to compare you against a competitor. Stay at the main breaker and you're stuck handing over a figure that says little more than the industry average already did.

Making the number exchangeable and auditable

A footprint that lives in a spreadsheet on one engineer's laptop is not an asset. Two things turn it into one: a standard way to hand it over, and a trail back to the measurements behind it.

On the handover, the industry has converged on PACT's data-exchange approach, which defines a common format and protocol so a product footprint can pass from supplier to customer system-to-system rather than as a PDF retyped by hand (WBCSD). The point is comparability: a footprint exchanged this way carries its boundary, its reference period, and its data-quality rating alongside the number, so the receiving company knows what it's integrating rather than guessing.

On the trail, the requirement is becoming legal, not just commercial. Under the EU's sustainability reporting rules, the climate standard ESRS E1 — set in Commission Delegated Regulation (EU) 2023/2772 — requires in-scope companies to disclose gross Scope 3 emissions across their value chain, reported without netting out removals or credits (EFRAG, 2023). Numbers that feed a regulated disclosure get looked at by assurance providers, and an auditor's first question is always the same: where did this come from? A footprint whose inputs trace back to specific historian tags, with the meter, the time window, and the allocation rule recorded, answers that question. One assembled from memory and a spend report does not.

That is the practical case for treating carbon as another measured quantity on the plant, governed by the same lineage discipline as quality or yield data. The same edge telemetry and analytics platform that already historizes energy and throughput is the natural place to compute and store the footprint, because the lineage from tag to reported number stays intact. It also means recalculation is cheap. When a customer asks why this quarter's figure moved, you can point at the gas meter and the production log instead of redoing a study.

Cheap recalculation is worth more than it first appears. A footprint is not a one-time deliverable; it's a quantity that should move when you switch fuel, change a recipe, or shift production between lines, and a customer running a real decarbonization program wants to see it move. If every update means re-running a manual life-cycle study, the number goes stale and the supplier relationship stalls on a yearly spreadsheet exchange. If the inputs are live historian streams with a fixed allocation rule on top, the current footprint is always one query away, and so is the answer to why it changed. That responsiveness is what turns a compliance artifact into something an engineer can use to compare two operating modes and pick the lower-carbon one. The instrumentation that gets you a defensible number in the first place is the same instrumentation that keeps it honest from one quarter to the next.

Where this breaks

None of this makes the number perfect, and overclaiming precision is its own failure. A few places where plant-data footprints go wrong:

Small categories swallowed by effort. ISO and the GHG Protocol both apply cut-off thinking — a minor input contributing well under a percent of the total may be excluded or left on a secondary factor. The mistake is spending months metering a trivial stream while a refrigerant log stays a guess. Screen first, instrument where the mass and the warming potential actually sit.

Refrigerants and fugitives. These rarely have a clean meter. They come from maintenance records, top-up logs, and mass-balance estimates, and their high global-warming potential means a sloppy figure here can outweigh a carefully metered energy stream. Honest practice is to flag them as the lower-quality inputs they are in the data-quality score, not to dress them up.

Allocation that flatters the product. Switching allocation basis until the headline number drops is the carbon equivalent of cherry-picking. Pick the physical basis that fits the process before you see the result, document it, and keep it stable across reporting periods so the trend means something.

Double counting and biogenic carbon. Grid electricity, upstream fuel production, and biogenic CO2 all have accounting rules that are easy to get subtly wrong, and a primary energy meter doesn't settle them. The meter tells you how much energy you used; the framework tells you how to characterize its emissions. You need both, and conflating them produces a precise number built on the wrong factor.

So is a metered footprint worth the instrumentation? For the inputs that dominate the result — energy and the major process streams — yes, clearly, because that is what moves you up the methods ladder and survives assurance. For the long tail, a good secondary factor honestly scored is the right answer. The discipline is knowing which is which, and the plant data is what lets you tell the difference instead of guessing across the board. That judgment, more than any single meter, is what separates a footprint a customer can act on from one they'll quietly discount.