the scoring contradiction.
why most ai capability statements in uk bids quietly undermine the same bid's sustainability and social value scores — and what defensible looks like.
the next bid you read will probably score ai capability, sustainability and social value as separate technical criteria. fifteen to twenty per cent each is now standard on private-sector frameworks in the built environment. three sections, three leads, three narratives, scored independently.
i've started reading them as one document. the results are uncomfortable.
here is a paraphrased composite — drawn from real submissions, anonymised — that i suspect will be familiar to anyone who has been near a major pursuit in the last twelve months.
§4.2 — digital and ai capability "we have embedded microsoft copilot and azure openai across our delivery practice. our project managers use generative ai for risk register synthesis, programme commentary, and client reporting. we anticipate 30% productivity gains across the project lifecycle."
§5.1 — sustainability and net zero "we are committed to a science-based pathway to net zero by 2040. our scope 2 emissions are reported on a market-based basis using renewable energy certificates. we will report scope 3 emissions associated with this commission quarterly."
§6.3 — social value "we will deliver £x of measurable social value in the local authority area, including local employment, supply chain spend within 30 miles, and stem outreach in nearby schools."
read in sequence, three confident narratives. read together, a problem.
the ai capability section commits the bidder to inference workloads running, in practice, on a small number of hyperscaler data centres. local partnerships — the public-sector advisory body — has been publicly flagging this exact issue to uk local authorities since september 2025: cloud providers, ai platforms and outsourced it functions are now among the most significant scope 3 contributors to an organisation's emissions, and they are routinely the least visible.
the sustainability section commits the same bidder to scope 3 reporting that has to include those workloads — and to a renewable-energy story that, when examined under location-based accounting rather than market-based, often does not survive contact with the actual grid mix where the compute sits. the social value section commits to economic benefit in region — while the marginal pound of ai spend leaves the country entirely, lands in a hyperscaler's revenue line, and supports zero local employment.
none of those three sections is wrong on its own. together, they describe a bidder who has not cross-read their own submission.
that is the contradiction. and it is everywhere.
i want to be careful about what i am and am not saying.
i am not saying hyperscaler ai is wrong. for many workloads, on many bids, it is the right answer and the contradictions above are bounded, disclosable, and defensible. the problem is not the architecture. the problem is the silence about it.
the contradiction is not theoretical. the first uk ai growth zone, at culham, sits seven miles from the proposed abingdon reservoir — a juxtaposition that has prompted legitimate concern about trading public water resilience for private compute unless water is planned and metered as tightly as power. a recent application for a 314mw hyperscale data centre in thurrock concluded "minor adverse — not significant" for greenhouse-gas impacts in its environmental statement, in part because annual operational emissions are a small fraction of the uk's national total. that is a methodology choice that will start losing bids the moment a procurement team reads it carefully.
bid teams are writing ai sections as though the compute is free of carbon, free of jurisdiction, and free of social-value consequence — because the people writing the ai section are not the people writing the sustainability section, and nobody is reading across.
i am also not saying the answer is sovereign or local-first ai by default. i have built a local-first, governed ai system for project delivery, and i think that class of answer is under-represented in current bids — but a bid that swaps "copilot" for "sovereign ai" without doing the same cross-read is no more defensible. it is just contradicted in different sections.
what i am saying is this. a bid that scores well on ai capability, sustainability and social value as a coherent whole is currently a rarity. procurement teams have not yet started scoring for coherence — but they will, and the regulatory ground is already moving under the question. ppn 06/20 has been superseded by ppn 002 from october 2025 under the procurement act 2023, and the crown commercial service became the government commercial agency on 1 april 2026 — both signals that procurement is being rebuilt around outcomes, not just inputs. social value weightings have moved from the original 10% minimum to 15-25% on most current frameworks, with some authorities at 30%. when ai capability claims start being scored against the same evaluation matrix as those social value commitments, incoherence is going to be a price.
if you are bidding now, the cheapest move available to you is the one nobody is making: read your own submission across sections, and write a short, honest paragraph in each that acknowledges the others exist.
a defensible ai capability section, in 2026, looks roughly like this.
it states the architecture. it quantifies — or credibly bounds — the inference workload's carbon and water implications, and reconciles them against the sustainability section's commitments using the same accounting basis. it names the data jurisdiction and the transfer mechanism, and reconciles that against any social-value claim that depends on data or spend remaining in region. it is honest about which workloads warrant cloud inference and which warrant local, on-premise or sovereign alternatives, and why. it does not claim a 30% productivity gain it cannot evidence, and it does not promise governance it has not built.
it is, in other words, the section a chartered project manager would be willing to sign their name against — knowing that a competent technical interviewer will read all three sections together and ask the obvious question.
if you are leading a pursuit right now where the ai section has been written separately from the sustainability and social value sections, i would gently suggest that this week is a good week to read them together. most of the work of fixing the contradiction is in noticing it.
i am, separately, doing this work as an independent review service for bid teams and client-side evaluators in the private-sector built environment. if that is useful, my inbox is open. if it is not, the cross-read is still worth an hour of your time before submission.