Most large companies' Scope 3 number is, in large part, an estimate built like this: take how much money you paid a supplier, and multiply it by an average figure for the sector that supplier sits in. The result is added to your inventory, reported, and set against your targets.
It is worth being clear about what that calculation actually produces. It does not produce a measurement of your supplier. It produces a statement about the average carbon per pound of revenue across everyone in their sector - applied, wholesale, to a specific company that may look nothing like that average.
Which is to say: a spend-based factor is sector-averaged revenue intensity. The weakest of the carbon ratios, averaged across a whole industry, and then used as the foundation for the largest part of most companies' footprint.
What we found when we checked
We ran an exercise with a large multinational client to test how well that assumption holds. We took their spend-based Scope 3 calculation, then went and looked at what their suppliers were actually doing - and specifically at the suppliers who could provide their own direct data: service-level or product carbon footprints, backed by life cycle assessment. Then we compared the two.
The gap was not a rounding error. At the good end, spend-based estimates were out by a factor of two or three. At the extreme end, one supplier's spend-based figure came out at roughly a thousand times the footprint that supplier's own life cycle assessment showed. Not thirty per cent high. Three orders of magnitude.
We are not alone in finding this. A UK certification body compared more than fifty verified clients across twenty-eight sector codes against national spend-based factors and found that in around three-quarters of cases, actual emissions were materially lower than the spend factor implied - by an average of roughly eighty per cent, with several sector codes showing variance beyond a hundred per cent. Different method, different sample, same direction of travel: the factor is systematically wrong, and usually wrong in the direction of overstating.
Why it breaks: how unlike the average is this supplier?
Two mechanisms drive that error, and understanding them tells you exactly where to look.
The first is internal variety. A spend factor is only as good as the population it averages, and how close your supplier sits to the middle of it. A supplier delivering a single service from a single location - a contact centre in one country, say - sits fairly close to a predictable position. The activity is uniform, the grid factor is knowable, the physical footprint is legible. The average has a decent chance of being roughly right.
Now take a conglomerate. Multiple businesses, multiple services, delivered from sites that differ enormously in energy mix, process intensity and local environmental context. The sector average is now an average of averages, spanning a population with vast internal variance. It tells you almost nothing about the specific entity you are buying from. Error scales with the diversity of what sits inside the supplier.
Why it breaks: the library assumes you are typical
The second mechanism is the factor library itself. These factors are derived from national input-output tables and sector classifications, which means they are built on an assumption: that a company classified in a sector is doing the ordinary thing for that sector.
Plenty of companies are not. A business can sit squarely inside a sector classification while delivering something structurally unlike the sector norm - lighter, more digital, differently configured, or simply operating a model the classification was never designed to describe. When that happens, the factor is not slightly off. It is describing a different business entirely. That is where the extreme cases live, and it is exactly where our thousand-fold gap came from: a supplier whose classification implied a heavy, industrial pattern of activity, and whose actual operations bore no resemblance to it.
The inversion at the heart of it
There is a consequence to all this that ought to stop procurement teams in their tracks.
If your supplier's emissions are calculated as their sector's average multiplied by your spend, then nothing that supplier actually does can change your number. They can move to renewable power, electrify their fleet, redesign their product, cut their real footprint in half - and your Scope 3 figure will not move by a single tonne. The factor is fixed by their sector, not their behaviour.
The only lever the method leaves you is spend. You reduce your calculated emissions by paying your suppliers less. This is not a subtle flaw or an edge case - the carbon accounting platforms say it plainly in their own documentation: with a spend-based factor, the only way to reduce emissions is to reduce expenditure. So the metric is structurally incapable of rewarding the exact behaviour every supplier engagement programme is trying to create, while quietly rewarding you for beating your suppliers down on price. Cost reduction registers as decarbonisation. Genuine decarbonisation registers as nothing at all.
The data was never missing
Here is the part that changes how I think about this.
The usual defence of spend-based accounting is that it is what you do when the data does not exist. And for parts of a supply chain that is true - there are chains where nobody reports and nothing is measured, and modelling is the only honest option.
But that was not what we found - and the first surprise came from inside the organisation, not outside it.
When we began asking which suppliers did what, and where they delivered it from, the answer from the central sustainability and data teams was that this was not known. Supplier locations were not known. The precise nature of the services being bought was not known. The information, we were told, did not exist.
It was an honest answer, and it was wrong in a specific and instructive way. What did not exist was a recorded field in the procurement system - which was, reasonably enough, where the data team had gone looking. Procurement systems are built to record what was bought and what it cost, not to describe what a supplier physically does or where they do it. The query came back empty, and empty was read as absent.
But somebody in that business knew. Somebody was receiving the service. Somebody was managing the contract, sitting in the review meetings, signing off the invoices, holding the budget. Ask them and they could tell you without hesitation what that supplier actually did and which sites it was delivered from. The knowledge was in the organisation. It was simply not in the system anyone had thought to search - and it was precisely the knowledge the method needs, because you cannot judge how unlike its sector a supplier is without knowing what it does and where it operates.
The second surprise came from the suppliers. A meaningful share of them already had the answer: product carbon footprints, life cycle assessments, in some cases figures published openly on their own websites. The number existed, calculated to a standard, often verified - and the buyer was using a sector average instead, because the two systems had never been connected.
That is dark data in both its faces. Not absent - unseen. Inside the business, the context sat with people rather than in a database. Outside it, the evidence sat with suppliers who had already done the work. In neither case was the organisation blind because the information did not exist. It was blind because nothing joined what was already known to the place the calculation was being made. And the cost of leaving it dark was not abstract. It was, in the worst case, a number three orders of magnitude away from reality, sitting in a reported inventory and underneath a public target.
What to do about it
None of this makes the spend-based method illegitimate. The GHG Protocol positions it as the fallback when better methods are not feasible, and as a first screen it does something valuable: it covers everything, quickly, and shows you roughly where your footprint sits. As a starting line it is fine. The failure is treating it as a finish line - and then setting targets against it.
Four things follow.
Target the effort where the error is largest. You cannot replace every factor, and you do not need to. The expected error is highest where high spend meets high internal variety meets an awkward sector fit - the diversified suppliers and the ones doing something atypical for their classification. That is a shortlist, not a boil-the-ocean programme.
Never let a proxy and a measurement share a column. A spend-based estimate and a verified product carbon footprint are not the same kind of object, and an inventory that presents them identically is hiding the single most important thing about itself. Grade the provenance, and let the grade travel with the number.
Keep two numbers, and be clear about which is which. One has to survive audit and assurance: it goes in the external report, and it is built only from evidence that meets that bar. The other is the number you believe to be true - the one that folds in what you have learned from suppliers even where the paper trail is not yet audit-grade, the site you know runs on renewable power, the footprint a supplier has shared but not had verified. That second number is often the more accurate of the two, and it is the one to manage the business on. Reporting the assured figure externally while tracking the truer figure internally is not double-counting or bad faith; it is the difference between what you can prove to an auditor and what you know to be the case. It gives you honest internal communication, a real read on how far your measurement has matured, and a point of truth to steer by - even when it is not yet a number you can publish.
And treat a spend-based line as an open question rather than an answer. It is not a measurement of a supplier. It is a marker saying: we have not looked here yet.
A placeholder is not a measurement
The deeper problem is the one that runs through all of this: we extract a number and act as though it were read. A spend-based figure carries no information about the specific company it is attached to, and yet it flows into inventories, into disclosures, into targets, into supplier scorecards - stripped of every caveat that would tell you how much weight it can bear.
Connecting what suppliers already know to what buyers are forced to guess, and grading honestly what remains genuinely unknown, is not a reporting improvement. It is the difference between a number and a placeholder. That connection - and that grading - is what Reverberate is built to do.
Sources
Understanding the Accuracy of Spend-Based Emission Calculations — NQA. A comparison of 50+ ISO 14064-1 verified clients across 28 SIC codes against national spend-based factors: around 77% showed actual emissions lower than the factor implied, averaging some 79% lower.
Technical Guidance for Calculating Scope 3 Emissions — GHG Protocol. The method hierarchy: supplier-specific, average-data and hybrid methods sit above the average spend-based method, which applies where the others are not feasible.
Spend based emission factors — Carbon+Alt+Delete. Vendor documentation stating plainly that with a spend-based factor, emissions can only be reduced by reducing expenditure - changing intensity requires supplier-specific or activity-based methods.