Extending device life against an endpoint support deadline

Separate the arithmetic from the constraint. The 40% figure is a rate effect: holding the estate size constant, embodied emissions per year fall from one-third to one-fifth of a device footprint, which is indeed a 40% reduction. That part is sound. What it does not show is total emissions over the period, which depend on failure rates, replacement of consumables, and any increase in use-phase energy from older hardware.

Identify the binding constraint. The regulator's expectation of supported software on customer-data endpoints is not a trade-off the bank can negotiate. A blanket five-year policy therefore fails as written, regardless of its carbon merit - and the failure occurs at month 40, in the middle of a refresh cycle, which is the worst possible time to discover it.

Notice what the constraint actually covers. Support attaches to the build, not to the chassis. That opens options the proposal did not consider: re-imaging with an extended-support build, virtual desktop or thin-client models for low-risk roles, and role-based segmentation.

Weigh the financials honestly. EUR 4.2 million deferred capex is real, but so are extended-support licence fees, a higher failure and repair rate, and residual values that collapse after year three. A deferred cost is not an avoided cost.

Use the pilot evidence properly. Failures were entirely consumable-related, which points to a maintenance and parts strategy rather than a hardware-quality problem - and it supports second-life devices for low-intensity roles, not for the whole estate.

Conclusion: reject the blanket policy, keep the objective, and replace it with segmentation plus a decision gate at month 36.

Segment on two dimensions, not one. The relevant axes are data sensitivity/regulatory exposure, and workload intensity. A device's appropriate life follows from what it holds and what it does, not from its purchase date.

• Customer-data roles: stay within the supported build. Refresh on the support calendar (three to four years), with mandatory certified data destruction and a take-back route so the embodied carbon is recovered through reuse rather than recycling.
• Low-sensitivity task roles and contact centre: extend to five years or use refurbished units, validated by the pilot, with a consumables replacement allowance built into the budget.
• Compute-intensive roles: refresh on performance need; a device that slows a fee-earning user costs more than the embodied carbon saves.

State the criteria explicitly so the policy survives staff turnover: supportability status, data classification, total cost of ownership including support and failure costs, measured failure rate by model, residual value trajectory, and availability of a take-back channel.

Add non-negotiable rules. No device leaves service without a data destruction certificate; every buy-back contract carries downstream reporting obligations so the bank knows the device was reused, not dumped.

Measure what the policy is meant to achieve: average device life, percentage of estate on refurbished or second-life units, units recovered versus scrapped, kilogrammes of e-waste, and embodied emissions avoided against a documented baseline.

Run it as a pilot with a gate. Apply to two segments for six months, review failure and support-ticket data at month three, and scale only if the data holds. A policy that cannot be reversed is a policy that will be reversed for you.

The error is a boundary error. The bank's reported footprint for purchased devices comes from what it actually buys - devices purchased multiplied by a per-device footprint - which sits in scope 3, purchased goods and services. Extending device life reduces the number of purchases, and that reduction is genuine and belongs in the inventory, provided the baseline year and the calculation method are documented. What does not belong in the inventory is a separate 'avoided emissions' figure derived from devices the bank did not buy; that is a comparative claim about an alternative scenario, and it is reported outside the inventory, clearly labelled, if at all.

Watch for double counting. If the buy-back partner refurbishes and resells the devices, the downstream benefit belongs to whoever uses the device next. The bank cannot claim both the avoided purchase and the onward reuse benefit from the same unit.

Get defensible per-device data. Use a supplier product carbon footprint where available, and where not, state the estimate, its source and its uncertainty. A 300 kgCO2e figure with no provenance will not survive assurance.

Keep evidence at device level: purchase records, model footprints, buy-back contracts, and certificates showing destination and treatment. A headline percentage without an audit trail is the most common way a credible programme becomes a credibility problem.

Report both sides of the ledger. Fewer purchases is a reduction; higher support licences and more frequent repairs are costs, and energy use from older hardware is an emission. Present the net position, and disclose the method so a reader can reproduce it.