Route Optimisation Saves Emissions You Cannot Report: The Scope 3 Counterfactual Problem in 2026

The number you want and the number you are allowed to report

The request usually arrives from the sustainability team in the second week of the reporting cycle. They need the emissions reduction attributable to routing, for the CSRD submission and for three customer scorecards. Operations has excellent data: every route, planned and actual distance, vehicle assignments, telematics, two years of history.

The answer that comes back cannot be used. Not because it is wrong, but because it is the wrong kind of number.

Emissions disclosure under the GHG Protocol Scope 3 Standard is an inventory. It reports what you emitted in a period, as an absolute figure. “Routing reduced our emissions by 400 tonnes” is not an inventory entry. It is an avoided-emissions claim, which is a comparison against a scenario that never occurred, and it sits outside the inventory rather than inside it.

So the two numbers answer different questions and belong in different documents. The inventory answers what you emitted. The avoided figure answers what you would have emitted otherwise. An operation can hold immaculate routing data and still be unable to produce the second one defensibly, because the constraint is methodological rather than technical.

This is the actual reason nobody can prove it. Almost every discussion of this problem assumes better measurement is the fix, and better measurement does not touch it.

Also Read: ESG Reporting Requirements for Logistics Companies (NA & EU)

Why the inventory number moves too slowly to credit routing

An inventory is honest and unhelpful to operations at the same time.

Absolute logistics emissions are driven by volume, network shape, fleet composition, and energy source. Routing quality sits underneath all of those. So an operation that improved sequencing materially, in a year when parcel volume grew 12% and two depots opened, will report emissions going up. The disclosure is accurate. It also gives no credit whatsoever to the work that was done, and it will not do so in any growth year.

Intensity metrics are the partial answer: grams of CO2e per parcel, or per tonne-kilometre. They strip out volume growth and are the right thing to trend internally. Their limitation is that they are not attributable either. Intensity improves when demand happens to get denser, when a larger vehicle class is introduced, when a new depot shortens stem distance, or when the mix shifts towards urban drops. None of that is routing, and an intensity improvement cannot be split between them after the fact.

Which leaves the gap that operations tries to fill with an avoided-emissions figure, and where the methodology problem starts.

The five methodology forks that decide whether a number survives assurance

Most guidance on logistics emissions stops after listing the inputs: distance, vehicle type, load factor, fuel. Those are necessary and they are not the hard part. The hard part is the choices made when converting them, because each fork changes the answer by more than routing itself does, and an assurance provider will ask about all five.

Distance basis. Planned or actual, and measured how. Great-circle distance understates road distance materially. Road-network distance without telematics misses diversions and dead running. Using planned distance is the most common error and it flatters the result twice, because it credits a plan that was not executed and quietly excludes the extra kilometres from failed attempts and re-attempts.

Fuel and energy model. Actual consumption from telematics or fuel cards is primary data and always preferred. Modelling consumption from distance and vehicle class is acceptable but blind to load factor, gradient, and stop frequency, which is where much of the real variance lives. A model that treats a half-empty van and a full one identically will not detect the consolidation gains that are the main mechanism by which routing reduces emissions.

Well-to-wheel or tank-to-wheel. Tank-to-wheel counts only combustion. Well-to-wheel includes producing and distributing the fuel or electricity. This is not a technicality for European fleets: under tank-to-wheel an electric van reports zero, so every kilometre electrified looks like a total elimination rather than a shift upstream. ISO 14083 and the GLEC Framework both require well-to-wheel, and a tank-to-wheel figure presented as a footprint will not pass review.

Allocation across shipments. A shared vehicle’s emissions must be divided among consignments, by weight, by volume, or by tonne-kilometre. For mixed-density freight the three methods produce very different per-shipment numbers. This matters commercially as well as technically, because the customer receiving your figure is comparing it against numbers from their other suppliers, who may have allocated differently.

Emission factor source and vintage. Which database, which year, and whether prior periods were restated when factors changed. Switching factor sets between reporting periods creates movement in the number that has nothing to do with anything the operation did, and an auditor who spots an unexplained factor change will discount the whole series.

The practical rule underneath all five is that assurance rewards traceability and consistency rather than precision. A conservative method, documented, applied identically across periods, with lineage back to source records, will pass. A more sophisticated method that cannot be reproduced from the records will not.

Also Read: What Is Carbon Neutral Shipping? A Guide to Better Logistics

The counterfactual problem, and three ways to handle it honestly

To claim routing saved something, you need what would have happened without it. That route never ran, so the baseline has to be constructed, and how it is constructed is the whole claim.

Three constructions are common and two of them will not survive scrutiny.

Year-on-year comparison is confounded by volume, network changes, fleet renewal, and mix. It measures everything at once and attributes it to whatever the report is about.

Pre-deployment versus post-deployment is better and still confounded, because organisations rarely change one thing. A platform deployment usually arrives alongside process change, new KPIs, management attention, and often a fleet refresh. Attributing the whole delta to the software is the classic overstatement.

Solver-generated baseline is the most common in vendor reporting and the weakest of the three. The system computes what an unoptimised plan would have cost and reports the difference. The problem is not arithmetic, it is that the unoptimised baseline is a strawman: no dispatcher would have run the naive sequence the solver constructs to compare against. The system is marking its own homework against an opponent it invented.

Three approaches hold up.

Held-out comparison. Keep a subset of depots or regions on the prior method for a defined period. This is a genuine control group and the cleanest evidence available. It is also operationally expensive, because you are deliberately running part of the network worse.

Staggered rollout as a natural experiment. This is the practical answer for most enterprises and it is usually free. If deployment reached sites in waves over several months, the sites not yet live during any given window are a control group, and the comparison is between concurrent periods rather than across years. Most large operations already hold this data and have never analysed it this way. If you are mid-rollout now, the window is open and it closes when the last site goes live.

Documented conservative counterfactual. Where no control exists, state the assumed baseline explicitly, select the least favourable defensible value rather than the most likely one, and publish the assumption next to the claim. A conservative number with a visible method is worth more than an optimistic one without.

Whichever is used, the same three disciplines apply. Label it avoided emissions rather than a reduction. Keep it outside the inventory. Never net it against reported emissions, which is the single fastest way to turn a defensible operational achievement into a challengeable environmental claim.

Also Read: EU Green Deal & Logistics: Key Regulations, Timelines & How to Prepare

Three ways operations reports routing emissions

The row that decides the rest is baseline. Everything else in an emissions claim is arithmetic that can be checked. The baseline is the only part that is a judgement, which is why it is the part an assurance provider will spend its time on, and the part most operations have never written down.

What to hold in the system so an auditor can follow it

This is a retention and versioning requirement rather than a calculation requirement, and it is the reason operations has to be involved. Nobody can reconstruct these records after the reporting period closes.

Per shipment and per route, retain planned and actual distance with the measurement source named, vehicle and energy type, load factor at each leg, the allocation basis used, the emission factor identifier and its vintage, and execution timestamps including failed attempts and re-attempts.

One detail is routinely missed. If a route was re-planned during execution, which version is the planned distance? An operation that re-optimises continuously and does not version its plans cannot answer that, which means the distance basis in its emissions calculation is undefined. Plan versioning is usually treated as a dispatch concern. It is also an emissions accounting dependency.

Also Read: Reduction of Carbon Emissions by Minimizing Fuel Miles

Who owns which part

The instinct to hand this to operations is half right and worth stating precisely, because getting the split wrong is how these programmes stall.

Operations owns the primary data and its lineage: the activity records, the versioning, the retention, and the discipline of capturing actuals rather than plans. Nobody else can produce this, and no methodology survives without it.

Sustainability owns the methodology and the disclosure: the five forks above, the factor sets, consistency across periods, and what is claimed publicly. Operations should not be selecting allocation methods or authoring environmental claims.

Finance and internal audit own the assurance relationship and the control environment, because a CSRD figure is a reported number subject to external review rather than an operational KPI.

The failure mode is operations attempting to own the claim, which produces numbers that get discounted, or sustainability attempting to own the data, which produces methodology applied to activity records that do not support it.

What to measure

Share of emissions from primary data. The proportion calculated from actual fuel or telematics rather than modelled from distance. This is the single best indicator of how much scrutiny your figure can withstand.

Distance basis coverage. The percentage of routes where actual distance is captured, not planned distance substituted.

Intensity trend, per parcel and per tonne-kilometre. Both, because the two diverge when vehicle mix or drop density changes, and the divergence is informative.

Factor set stability. Whether emission factors changed between periods and whether prior periods were restated. An unexplained change invalidates the trend.

Counterfactual coverage. The share of claimed savings backed by a control group or a documented conservative baseline rather than a solver comparison. For most operations this starts at zero.

Also Read: Sustainable Last-Mile Delivery: 2026 Enterprise Guide

How Locus makes routing emissions defensible rather than merely favourable

Locus, the world’s first Decision-Intelligent, Agentic TMS, treats emissions as an operational output of the decision record rather than a calculation performed afterwards on exported data. Its agents run a continuous Sense-Decide-Execute-Learn loop against a model of more than 250 real-world constraints, and because emissions can enter that model as a constraint alongside cost and service, the routing decision and the emissions record derive from the same source of truth rather than from two systems that have to be reconciled.

Three properties matter for assurance specifically. Traceability means each decision retains its inputs and the plan version it produced, which is what makes a distance basis definable in an operation that re-plans through the day. Explainability means a route arrives with the constraints it honoured, so an assurance provider can see why a vehicle was assigned rather than inferring it. And because deployment is typically staged across sites, the rollout sequence itself is retained, which is the raw material for the natural-experiment counterfactual described above.

Across its customer base Locus has contributed to 17M+ kg of CO2 avoided and 800M+ miles reduced over 1.5B+ deliveries. The wording there is deliberate and worth noting as an example of the discipline this piece argues for: avoided, not reduced from an inventory, because that is what a figure of this kind actually is.

Locus is recognized by Gartner for seven consecutive years, featured in the 2026 Hype Cycle for Supply Chain Execution and Logistics Technologies, named a Leader in TMS by QKS Group (SPARK Matrix), and ranked #1 in Route Planning on G2’s 2026 Best Software Awards. In October 2025, Ingka Investments, the investment arm of Ingka Group, the world’s largest IKEA retailer, acquired Locus. Locus continues to operate independently. Further analyst recognition is published in full.

One deployment illustrates the counterfactual point directly. A Fortune 50 parcel and logistics provider centralised dispatch across 51 sites in a 120-country network, running more than a million freight shipments a year. The rollout was staged site by site, which is precisely the structure that yields a control group: at any point during the programme, live sites and not-yet-live sites were operating the same network under the same conditions with different planning methods. Weekly execution rose from 75% to 92% and more than $14 million in unused capacity was surfaced, including $565,000 at a single site. Capacity surfaced at that scale is distance and fuel not spent, and because the rollout was staggered, the comparison supporting it is concurrent rather than year-on-year.

A global FMCG operation across 10 countries, 1,000+ distributors, and 5,000+ riders reports 12,000+ trips saved per month. That figure is a good illustration of the problem this piece describes: trips saved is a genuine operational result and it is still a comparative claim, so putting it into an emissions disclosure requires stating what those trips would otherwise have been and how that was established.

Request a Locus logistics emissions assurance readiness assessment to review your distance basis, primary data coverage, factor governance, and whether your rollout history can still support a counterfactual.

Start with the baseline, not the calculation

Most operations approach this by improving the calculation, because that is the part that looks like an engineering problem. It is the wrong end.

Establish what share of your emissions figure comes from actual rather than modelled data, and write down the baseline you are implicitly comparing against when you claim a saving. If the honest answer is that the baseline is whatever the optimiser generated, you do not have a claim yet, and no amount of additional precision will create one.

Then check whether you are still mid-rollout somewhere. If you are, you are holding a control group that will not exist in six months.