In short: ESRS E5 wants resource inflows in mass with a secondary content share attached, and most purchasing systems carry cases, litres and each, with a gross shipping weight on maybe two thirds of items. The item ranking by mass is close to the inverse of the ranking by spend in several industries, so a data programme scoped off the spend Pareto collects the wrong two hundred items. Limited assurance tests whether the number can be re-derived rather than whether it is right, which makes a versioned factor library and a stored mapping from transaction lines to datapoints the two artefacts worth building first. The value chain phase-in in ESRS 1 is a three year countdown, and it usually gets spent defending an estimate instead of building the collection route.
The request arrives in October, from a sustainability lead who has been given the resource use section and has read the standard properly. She needs the mass of materials that came into the business last year, split by material category, with the share that was secondary rather than virgin. She would like it by the end of the month.
The planning team can produce every purchase order line from the last three years in about four minutes. Forty thousand lines, item code, supplier, quantity, unit of measure, value. The units are cases, pallets, litres, rolls and each. There is no mass anywhere in that extract. The conversion from a case to kilograms exists, in a specification held in the product lifecycle system, occasionally as a scanned PDF, and it covers the items somebody once needed a shipping weight for.
That gap is the whole of CSRD supply chain reporting in one exchange, because the accounting method for emissions is a solved and well documented problem while the question of where the number comes from has barely been started.
The datapoint that has no unit in your systems
ESRS E5, resource use and circular economy, asks for resource inflows expressed as mass, with the share of secondary reused or recycled materials and components, and for outflows including waste split by treatment route. Mass is the natural unit for a material question and it is the unit almost nobody's transactional data is denominated in.
Take a manufacturer with 6,200 purchased items. The item master carries a gross weight on 4,100 of them, because those either ship to customers or move on a pallet somebody had to book. Coverage looks like sixty six percent, which sounds workable until you notice two things. The weight recorded is gross, including transit packaging, which is the wrong quantity for a material inflow figure and right for a freight one. And the missing third is not random: it is the items that arrive in bulk, in tankers, in reels, or in units of measure the warehouse never had to palletise.
The instinct at this point is to sort the item list by spend and work down until coverage is good enough. That instinct comes from scope 3 category 1 work, where sorting by spend is a reasonable start, and it produces the wrong list here.
Consider a pharmaceutical manufacturer buying an active ingredient at 900 per kilo and buying excipients, solvent and packaging at around 2 per kilo. The active might be sixty percent of purchase value and three percent of purchased mass. Rank by spend and the top two hundred items get you most of the money and a rounding error of the tonnage. Rank by mass and you get a different two hundred items, mostly packaging, bulk chemicals and commodity inputs, which is where the recycled content question actually lives.
Run both rankings before scoping anything. In a business with a wide value range across the bill of materials, the overlap between the top hundred by spend and the top hundred by mass is often under a quarter, and the size of that overlap tells you whether the emissions data programme and the resource use data programme are one project or two.
What limited assurance actually tests
CSRD requires the sustainability statement to sit inside the management report and to carry an assurance opinion, initially at limited assurance, with the Commission mandated to assess a move to reasonable assurance later. Companies that have been through a first cycle describe the assurance conversation the same way, and it is worth knowing what it consists of before you design anything.
The assurance provider samples transactions and traces them back to a source record. They re-perform the calculation on that sample. They ask who can change the inputs and whether there is a record of changes. They ask whether last year's figure can be produced again today from the same inputs.
That last question defeats more first-time reporters than any other, and the cause is nearly always the factor library. The UK government publishes a new conversion factor set every year. The US EPA updates its factors. Ecoinvent ships numbered releases. If your process refreshes those in place, overwriting last year's values, then last year's number cannot be reproduced and the year on year movement you disclose is a mixture of two effects nobody has separated.
The fix is the same discipline a finance team applies to a price and volume variance. Hold the factor set as a versioned artefact with an effective date, keep the version that was used against each reported period, and decompose the movement into an activity effect and a factor effect. Report the activity effect as the operational change, because it is the only part of the movement your supply chain caused. A year where tonnage fell four percent and the factor set moved seven percent produces a headline improvement that no decision in your business created, and an assurance provider who asks the right question will find that out before your board does.
The second artefact worth building early is the mapping. One row per disclosed datapoint, naming the source system, the query or extract that produces it, the unit, the conversion applied, the owner, and the treatment for missing values. It is roughly a week of work for a first pass, it is the document the assurance provider will ask for, and it survives every reorganisation of the reporting process. Spreadsheet formulas dragged across a range do not.
The value chain phase-in is a countdown
ESRS 1 includes a transitional provision for value chain information. For the first three years, an undertaking that cannot obtain information from its upstream or downstream value chain after reasonable effort may explain those efforts, why the information was not available, and how it plans to obtain it in future.
Read as relief, that provision buys three annual cycles. Read as a schedule, it means the collection route has to be operating by year four, and year four is a fixed date the moment you file your first statement.
What tends to happen in the interim is that the three years get spent on the estimate. Teams refine the modelling, argue about factor selection, improve the narrative around the gap, and produce a defensible figure each year. None of that work builds the route by which primary data arrives. When the phase-in expires, the estimate is better and the pipeline still does not exist.
The work that does compound is unglamorous. Getting mass and material composition onto the item master as maintained fields with an owner. Getting supplier and origin recorded at the receipt rather than inferred from the purchase order. Getting the unit of measure conversions out of PDFs and into a table. All of it is item master hygiene, all of it is useful for other reasons, and it is what the collection route is built on top of. Approaching the same suppliers for primary data is a separate discipline with its own failure modes, and GG3 covers the mechanics of asking without drowning them.
The Omnibus moved the deadline and left the data question intact
Anyone scoping this work since early 2025 has been aiming at a moving target. The Commission published its simplification proposal on 26 February 2025. A stop the clock directive followed in April 2025, pushing the reporting dates for the second and third waves back by two years and delaying the related transposition deadlines. The substantive amendments, including the employee threshold for who reports at all, went through several rounds of negotiation after that, and EFRAG ran a parallel revision aimed at cutting the number of mandatory datapoints substantially from the first set.
Check the current text before you scope anything, because the thresholds have been recut more than once. Two things have held across every version.
The first is a cap on what large reporters may demand from smaller companies in their value chain. In the proposals, information requests to companies below the employee threshold are limited to a defined voluntary standard for smaller undertakings. Whatever the final number, the direction is settled: the questionnaire you were planning to send your long tail of suppliers is not going to be enforceable, and the marginal supplier data you were counting on will not arrive that way.
The second is that nothing in any version reduces what you must report about your own operations and your own purchases. Quantities bought, mass received, where it came from, how it moved, what was wasted and by which route. Those are your records. Every simplification round has narrowed who reports and trimmed the datapoint list, and none of them has made your own transactional data somebody else's problem.
Which numbers a planning system should own
There is a clean split in the ESRS datapoint list, and drawing it early saves an argument later.
A planning platform can reasonably own the physical flow datapoints, because it already holds them at the grain the disclosure needs: purchased quantities and their mass, production volumes and yields, transport activity in tonne kilometres by mode, inventory written off, waste arising by treatment route where it comes off a plant or a distribution centre, and packaging placed on the market.
It should not own workforce datapoints, governance disclosures, policy existence, or incident registers. Those live in HR, legal and EHS systems, and routing them through a planning platform adds a hop and an owner without adding anything.
The datapoint that gets fought over is the factor library, because finance, sustainability and planning all have a claim on it. The test that settles it is grain. A factor is useful for a disclosure at the level of an annual total, and it is useful for a decision only if it can be applied at item and lane level at the moment the decision is taken. S1 works through what that requirement does to factor selection, and the conclusion for this post is narrow: put the library where it can be applied at the finer grain, because the coarser use can always be aggregated up and the reverse does not work.
One consolidation trap is worth flagging. The report is drawn at legal entity level over a calendar year. The plan is drawn over a network of nodes and lanes on a rolling horizon, and that network includes co-manufacturers and third party warehouses that sit outside the consolidation boundary. Your reported scope 1 will exclude a co-manufacturing site that your planning model treats as a plant, which is correct under both sets of rules and guarantees the two numbers never agree. Write the boundary mapping down once. Y7 covers what co-manufacturing does to the plan itself.
The limit
Assurance tests the process rather than the physics. A spend-based category 1 figure with clean lineage, a versioned factor set and a full audit trail will pass limited assurance and can still be wrong by a wide margin, because the WRI and WBCSD Scope 3 Technical Guidance (2013) puts spend-based data at the bottom of its own quality hierarchy and the assurance standard does not require you to climb it. Being auditable and being accurate are independent properties, and the reporting regime currently rewards the first one much harder.
The mass data will be incomplete for years. The honest treatment is to disclose the coverage alongside the figure, stating the percentage of purchased mass that carries a verified weight and composition, and to resist the temptation to gross up the remainder using an average. Treating unknown material as zero secondary content biases the recycled share down. Treating it as the portfolio average biases it wherever the unmeasured items are unusual, which they generally are, since the reason they are unmeasured is that they arrive in an awkward unit from a supplier who is hard to reach.
And a planning system is the wrong home for a good part of this. The qualitative disclosures, the double materiality assessment, and the transition plan narrative are documents produced by people, and no data model improves them.
Start with one query. Join last year's purchase lines to whatever mass exists on the item master, and report the percentage of purchased tonnage you can convert to kilograms with no manual lookup. That single figure tells you whether the next twelve months are a data project or a supplier project, and it takes an afternoon.