The Carbon Border Adjustment Mechanism is a compliance regime dressed as a climate policy, and category managers are the people it actually lands on. The in-scope goods — cement, iron and steel, aluminium, fertilisers, electricity, and hydrogen — cover a surprising share of what a UK manufacturer or public body buys. The liability is computed on embedded carbon, multiplied by quantity, adjusted for any carbon price already paid at origin, and settled against a certificate price that moves. None of those inputs sit still, and the retrospective spreadsheet approach that many teams started with in 2024 is now visibly breaking. This post is a field guide to doing it differently, using agentic AI to move the calculation to the moment of decision.
Start with what a category manager is actually trying to decide. The question is rarely "what is our CBAM exposure for last quarter". The question is "should I place this order with this supplier at this price, given that the landed cost has to include the CBAM liability I will incur when the goods cross the border". Those are different questions. The first is a reporting problem. The second is a procurement problem. The reporting problem can be solved by pulling shipment records into a template once a quarter. The procurement problem cannot, because by the time the shipment record exists, the purchase order has already been signed at a price that may or may not reflect the real economics.
CBAM liability has four moving parts. The first is embedded carbon intensity, measured in tonnes of CO2 per tonne of product. This varies by commodity, by production route, and by supplier. Primary aluminium from a smelter powered by hydroelectric generation has a very different profile from primary aluminium from a coal-powered smelter, and neither matches the profile of recycled aluminium. The second is quantity, which is the easy part if the order is clean and the hard part if the order is a mixed bill of materials with partial CBAM exposure. The third is the deemed carbon price already paid in the country of origin, which is a credit against the UK liability. The fourth is the certificate price the buyer will pay to settle the liability, which varies with the EU ETS reference and UK implementation parameters.
A spreadsheet can hold those four numbers. A spreadsheet cannot keep them current across a portfolio of suppliers, commodities, and destinations, and it cannot flag the case where the right answer changes mid-negotiation because one of the inputs moved. Category managers who have tried to run CBAM out of Excel describe the same failure mode: the numbers were defensible on the day the sheet was updated and drifted within days. The drift is not a clerical problem. It is a structural mismatch between the cadence of the calculation and the cadence of the inputs.
Agentic AI solves the cadence problem by running the calculation on demand against live sources. At WYRM Sentinel, the carbon agent queries per-commodity embedded carbon factors at the moment the buyer asks the question. The commodity agent pulls the current LME reference price. The FX agent pulls the current sterling cross. The compliance agent checks origin, sanctions, and deemed-carbon-price credits. The supplier agent confirms fundamentals on the counterparty. The shipping or air freight agent verifies the route is viable. A minimum of seven agents run per query, in parallel, and the fusion layer returns a single landed cost with CBAM exposure broken out as a line item. The whole thing resolves in seconds.
Take a concrete example. A UK buyer needs 15 kilograms of primary aluminium, origin Norway, delivered to a Midlands facility. A category manager running this through Sentinel sees the following resolved in under ten seconds. LME primary aluminium three-month, converted to sterling per kilogram, gives a base price. The carbon agent returns an embedded carbon intensity appropriate for Norwegian hydro-powered smelting, which is at the low end of the global range. Quantity is 15 kilograms, so embedded emissions are computed directly. Deemed carbon price paid at origin is credited per the current UK CBAM implementation. The certificate price is applied to the net liability. Freight is added from the shipping agent. The output is a single landed cost with every input cited and timestamped.
The scale of the CBAM adjustment at 15 kilograms is small in absolute terms, but the point of the example is not the magnitude. It is that the same logic runs identically at 15 tonnes and at 15,000 tonnes, with the same evidence trail. The category manager who places a 15-kilogram order and a 15-tonne order on the same day has the same audit record format for both, because the system does not care about scale. That consistency is what makes the approach defensible in a category review or a regulator query.
Multi-commodity orders are where the spreadsheet approach fails hardest and where the agentic approach earns its keep. A fabrication supplier might ship a single consignment that contains in-scope steel, in-scope aluminium, and out-of-scope fasteners. The CBAM liability attaches to the in-scope lines only, at different embedded carbon factors, and the out-of-scope content has to be separated cleanly. Doing that by hand across a year of orders is the kind of work that produces errors no one catches until the audit. Doing it per-order at the moment of purchase, with each line resolved by the agent responsible for that commodity class, produces a record that reconciles against itself by construction.
The Procurement Act 2023 is the other piece of context a UK category manager has to hold in mind. The Act raises the bar on auditability and defensibility for supplier decisions, particularly in the public sector but increasingly as a signal for mid-market enterprise as well. A buyer who can show the CBAM calculation that was visible on screen at the moment of decision, with the source URLs the calculation drew from, is in a materially stronger position than a buyer who can only show a reconciliation sheet produced a quarter later. The Act does not mandate a particular tool. It rewards a particular shape of evidence.
A practical question category managers ask is whether the embedded carbon factors used by the system are the right ones. The honest answer is that CBAM implementation allows both actual verified emissions data from the producer and default values where verified data is not available. A good agentic system does not hide this. The carbon agent returns the factor it used, the source, whether it was a verified supplier-specific value or a default, and how sensitive the final liability is to that choice. A category manager can then decide whether to push the supplier for verified data, whether to absorb the default-value margin, or whether to source differently. The point is to make the trade-off visible, not to claim a false precision.
For portfolios that span multiple in-scope commodities, the efficiency gain from running CBAM at decision time compounds. A team buying cement for construction projects, steel for fabrication, aluminium for packaging, and fertiliser for estate management would otherwise need parallel tracking workflows with different expertise in each. An agentic system collapses that into a single interface where the buyer asks the same kind of question and the relevant specialist agent answers. The category manager does not need to be a CBAM expert in six commodities. They need to be able to read the evidence trail and defend the decision.
A useful operating pattern for category teams is to run the agentic check before the final supplier selection rather than after. In teams we have worked with, this changes supplier behaviour fairly quickly. When the buyer can show a competing supplier the CBAM-adjusted landed cost, with the embedded carbon factor itemised, suppliers with better carbon profiles become competitive at higher nominal prices and suppliers with weaker profiles face visible pressure to provide verified data. The conversation stops being about headline price and starts being about total economic cost, which is what CBAM was designed to produce.
Data residency is a practical consideration for UK public sector and regulated-industry teams. WYRM Sentinel runs UK-first with UK data residency by default, and EU residency available on Enterprise plans. For CBAM workloads specifically, this matters because the underlying supplier and order records are commercially sensitive and often subject to contractual residency clauses. A system that forces CBAM calculation through a non-UK data path creates a compliance risk of its own, which is an ironic outcome for a compliance tool. The architectural choice to keep the evidence trail in the UK jurisdiction is part of the product, not a deployment option.
The short version of this field guide is that CBAM is not a reporting problem dressed as a compliance problem. It is a procurement problem dressed as a compliance problem, and it belongs at the moment of decision. Agentic AI is the architecture that puts it there, because running the calculation requires reasoning across commodity prices, foreign exchange, carbon factors, origin credits, certificate prices, and shipping in parallel, with an evidence trail the auditor will accept. A category manager who moves CBAM from the quarterly reconciliation to the purchase order signing moment removes a class of risk from the portfolio and earns back the time that spreadsheet reconciliation was consuming. That is the trade worth making in 2026.