Automatic Tender Response Software: Six Jobs, and What UK Law Won't Let You Automate
“Automatic tender response” is one phrase covering six jobs that have almost nothing to do with each other: find the notice, qualify out of it, extract the requirements, draft the answer, version what you sent, and check it before you submit. Very few products are strong at more than two.
And one of the six is no longer yours to buy. Since the Procurement Act 2023 came into force on 24 February 2025, the state runs a central register that holds your core supplier information and hands it to each buyer — free, whether or not you buy anything.
What automatic tender response software actually is
Every product in this category sells “automation” as one thing. It is six, and the honest question is not whether a tool automates tender responses but which of the six it is good at — because the failure you are actually suffering is usually one specific job, not all of them.
The split matters commercially. A team losing bids because it never sees the right notices has a find problem, and no amount of drafting quality fixes it. A team drowning in unwinnable bids has a qualify problem, and better drafting makes it worse by lowering the cost of saying yes. Work out which one is failing before you shortlist anything.
already does it
does it well
it is on you
Find
Every notice published under the Act goes to one place. Section 95 requires a Minister of the Crown to establish an online system for publishing notices under the Act [1], and Find a Tender is that system. So discovery is a solved problem in principle — the work software does is filtering, watching and alerting, not sourcing.
The Act does the first half: it made the corpus exist and be complete. It does not tell you which of several hundred notices a week is yours.
Qualify
Deciding not to bid is the highest-value decision in the process and the one no machine should make for you. A tool can surface the gates — the conditions of participation, the stated certifications, the deadline, the incumbent — but the judgement about whether you can win this one is commercial, and it is the judgement the rest of the process compounds.
Nothing in the Act helps here. Conditions of participation must be “a proportionate means” of ensuring you have the legal and technical ability to perform the contract [10], which constrains the buyer — it does not tell you your odds.
Extract
Pulling every requirement out of a pack, with the sentence each was read from, is the job current language models are genuinely good at, and it is boring enough that people do it badly by hand. It is also checkable: a requirement with its source sentence attached can be audited, and one without it cannot.
This is the strongest honest claim in the category, and notice how modest it is. Extraction is not drafting.
Draft
Shared, and the split is not where the marketing puts it. Assembling a first pass from answers you have already written and had assessed is reuse, and reuse is a solved software problem. Writing the parts that are specific to this buyer, this contract and this year is not, and those parts are where the marks are — because the published criteria under section 23 [4] are what you are scored against, and they change per procurement.
Chapter 12 sets out the things in a pack that a machine must not draft at all.
Version
Software, comfortably, and this is the least glamorous and most reliably useful of the six. Knowing exactly what you submitted, and when, is what makes the assessment summary you get back afterwards worth reading.
One correction, because the category states this wrongly: keeping bid versions is not a legal requirement on you. Section 98 binds contracting authorities, not suppliers [9]. Chapter 9 has the real reason.
Check
Genuinely three-way. A machine can verify the mechanical things — every question answered, every word limit respected, every mandatory document attached, every stated certification held. The Act supplies part of the checklist by fixing what an authority may ask for and what it may not [10]. And a human still has to read it, because the failure modes that lose bids are meaning failures, not format failures.
Register
The seventh row, and the one the category has not absorbed. Regulation 6 of the Procurement Regulations 2024 requires a contracting authority to obtain confirmation from the supplier that it “has registered on the central digital platform”, “has submitted its up-to-date core supplier information to that platform”, and “has given that up-to-date core supplier information to the contracting authority by means of a facility provided on that platform” [14].
Core supplier information is defined across regulations 9 to 12: basic information, economic and financial standing, connected persons, exclusion grounds [15]. That is most of what a selection questionnaire used to ask for, held centrally and re-used. Chapter 6 works through what it does and does not cover.
Three states, not a score: a filled mark means the job is done for you, a half mark means partly, and a faint dot means not at all. Every cell is argued in the panel beside it and cited to a numbered primary source below. Select a job to read the reasoning; all seven panels are in the served HTML whether or not JavaScript runs.
Tender response automation under the Procurement Act 2023
Automation needs rigidity. A process that varies per buyer cannot be automated past the first step, and a process fixed by statute can be automated to the edge of judgement. The Act made several things rigid that used to be discretionary, and that is the whole reason this category works better in the UK public sector than in private-sector procurement.
Three of them matter. The award test is fixed: section 19 requires the authority to award to the supplier submitting the most advantageous tender [3], assessed against criteria it has to set out under section 23 [4]. The procedures are fixed: section 20(2) offers exactly two — a single-stage open procedure and a competitive flexible procedure [5]. The notices are fixed: there is a defined set of notice types and a defined order, and they all land in one place.
What is rigid is automatable. What remains discretionary — how the criteria are weighted, what “quality” means for this contract — is not, and no amount of model capability changes that, because the variation is in the buyer, not in the language.
Tender response software: which of these pages you actually want
We publish more than a dozen pages that all touch “tender response software”, and that is a problem we made, not one you should have to solve. So here is the map, with what each page owns, stated plainly enough that you can leave this one.
This page owns automation as a category: what the six jobs are, which the regime already does, and what cannot be automated at all. It deliberately does not rank products, compare two products head to head, or map the wider tool landscape — three jobs that belong to three other pages.
The category map
What the tool categories are and where the boundaries fall. Start here if you do not yet know what you are shopping for.
The ranked recommendation
Which is best, with the criteria stated. This page will not rank anything.
Head to head
X versus Y, feature by feature, when you have already shortlisted.
Eleven UK products, weighted by you
The competitor comparison, including where we lose.
The SME lane
What changes when the bid team is two people and one of them also delivers the work.
Pipeline and workflow
Managing a portfolio of bids rather than writing one.
Versioning, in depth
Chapter 9 here is a digest. This is the full treatment, and it outranks us on the term.
Tender response software
The term on its own, as a category name rather than as an automation argument.
The statute itself
What changed on 24 February 2025 and what it means for suppliers.
“Bid response software” is the same thing under a different word
Bid, tender, proposal and response are used interchangeably by UK buyers, and the software category does not split along them. If you searched for bid response software and landed here, nothing on this page is about a different product — the six jobs are the six jobs. The vocabulary chapter below sets out where the words genuinely do diverge, which is narrower than the market implies.
Automated tender response: the four steps, and what each one can't do
Almost every product in this category runs the same four-step pipeline. The steps are real and they work. What follows each one here is the limit — the thing the step cannot do, stated because the limit is what decides whether the tool helps you.
Ingestion — the notice and the pack come in
The notice is structured data; the pack rarely is. A tender notice under section 21 carries defined fields, so ingesting it is reliable. The attached documents are PDFs and spreadsheets written by whoever wrote them, and that is where ingestion actually gets hard.
The limit: a scanned or badly-tagged PDF degrades everything downstream, silently. A tool that does not tell you its extraction confidence is hiding the one thing you need to know at this step.
Library matching — your prior answers are found
Your last twenty responses contain most of the answer to the next one. Matching them to a new question is retrieval, and retrieval is mature.
The limit: it will happily return an answer that was true in 2024. Reuse without a freshness check is how a bid ends up asserting a certification that lapsed, or a headcount that halved, or a policy that was rewritten — and those are the assertions a buyer verifies.
Draft generation — a first pass is written
A model can produce a structurally complete, on-length, on-topic first draft against the published criteria faster than a person can open the document.
The limit: it optimises for plausibility, which is exactly the wrong target when the buyer is scoring specificity. A draft that reads well and names nothing loses to a worse-written answer with a real example in it. It will also invent commitments if you let it, and chapter 12 explains why that is a contractual problem rather than an editorial one.
Human review — the part that is not optional
Someone who can be held to the answer reads it, corrects it, and signs it.
The limit: review quality collapses when the draft looks finished. This is the least-discussed risk in the category and the best-evidenced one in ordinary editorial practice: people correct a rough draft and approve a polished one. If your process has no step that forces someone to check claims against evidence rather than read for tone, automation has moved your failure mode rather than removed it.
RFP, ITT, PQQ, SQ: automating across the vocabulary
Four acronyms, one pipeline, and a genuine divergence in only one place. Most of the difference is what the buyer calls the document; the part that actually changes your work is whether you are being selected or scored.
Automated RFP response, and why the UK case is different
“RFP” has no definition in the Procurement Act 2023 — the Act's vocabulary is tender notice, tender, open procedure and competitive flexible procedure. A UK public body may still title its pack an RFP, and many do, but the obligations attach to the procedure underneath the title, not to the word on the cover. So automating an RFP response in the UK public sector means automating a tender response, whatever the file is called. We work through that distinction properly on RFP vs tender and the RFP tender guide.
Where it does change is the private sector, where no procurement statute applies at all, the buyer can change the process mid-flight, and negotiation after submission is normal. Automation is more useful there for drafting and less useful for compliance, because there is no fixed compliance to check against.
RFP response automation across the selection and award stages
A selection questionnaire asks who you are. A tender asks what you will do. That is the divergence that matters, and it is the one the central digital platform has changed: the “who you are” half is now held centrally and re-used, which is chapter 6.
Depth on each document type lives on its own page — ITT response software for the invitation to tender specifically, and PQQ and SQ software for the selection stage, where the automated tender response case is now largely a question of what the platform already holds for you. RFI vs RFP covers the market-engagement stage that precedes both.
AI-powered auto-fill for bid content — and the auto-fill the state already built
Every product in this category sells the same story: stop retyping the same supplier facts into every selection questionnaire, let the software fill them in. It is a good story. It is also, for UK public procurement, describing a job the government has already done and does not charge for.
Regulation 6 of the Procurement Regulations 2024 requires a contracting authority to obtain confirmation from the supplier that it “has registered on the central digital platform”, that it “has submitted its up-to-date core supplier information to that platform”, and that it “has given that up-to-date core supplier information to the contracting authority by means of a facility provided on that platform” [14]. In an open or competitive flexible procedure that confirmation has to be obtained before the end of the tendering period.
What counts as core supplier information is not left vague. Regulations 9 to 12 define four sets: the supplier's basic information, its economic and financial standing information, its connected person information, and its exclusion grounds information [15]. That is most of what a selection questionnaire used to ask you for, every time, from scratch.
The Cabinet Office's own guidance puts the supplier-side effect plainly: once the information is entered, “it will be simple to review it and re-use it every time for each tender”, and suppliers “only need to register at the point that they want to bid for a contract” [16].
None of that makes the software category pointless. It relocates it. The auto-fill worth paying for is not the part the platform holds — it is the part it does not.
Left column: what regulations 9–12 define as core supplier information and regulation 6 routes to the buyer through the platform [14] [15]. Right column: what is left. No quantity is claimed and none is implied — the component's point is the asymmetry, and the asymmetry is the argument.
Tender automation UK: where the notices actually are
Automating discovery means knowing which registers exist and which of them your tooling actually reads. Most pages in this category list seven portals as if they were equivalent. They are not: two carry the statutory UK flow, one is a legacy archive, one is a different legal regime entirely, and the rest are buyer-side systems you meet one at a time.
We ingest Find a Tender and Contracts Finder. We say so plainly because a discovery tool that quietly covers two registers while implying seven is selling coverage it does not have — and because the below-threshold half of the market is the half most alert services skip. How our Find a Tender ingest works sets out what we take and how often, and you can search the corpus without an account.
The thresholds that decide whether any of this applies
Above these figures the full tendering regime applies. Below them, a lighter set of duties applies instead, and the notice obligations are different. They fell on 1 January 2026 [19].
Source: PPN 023, Annex A, summary of threshold amounts from 1 January 2026, cross-checked against the row it corresponds to [19]. Separately, section 87(4)'s below-threshold notification triggers — £12,000 for non-Welsh central government and £30,000 for everyone else — did not change on 1 January 2026 [13]; the amendment made by SI 2025/1200 was to section 85(3) [21]. The £25,000 figure that appears on a lot of pages, including the previous version of this one, was the old Public Contracts Regulations sub-central trigger and has not been current since 24 February 2025. We make no claim about whether these figures are inclusive or exclusive of VAT: PPN 023 does not say, and neither does the Schedule 1 column heading.
Automated compliance checking: what a machine can actually verify
“Compliance automation” is sold as one capability. Split it and you find one half that is genuinely mechanical and one half that is a legal position about your own company — and the second half is not something you want a model asserting on your behalf.
Automated document assembly for tenders — the part that is genuinely solved
Assembling the response pack itself — the right documents, in the right order, in the right format, with the right file names and every mandatory attachment present — is a solved problem and an unglamorous one. It is also, in our experience of the failure reports buyers publish, a meaningful share of the bids that are rejected before anyone reads them. If you automate one thing, automate this, because it is the one place where the machine's advantage is total and the judgement content is zero.
Auto-versioning for bid responses and compliance
Keeping every version of every answer, with who changed what and when. Useful — and routinely sold to you on a legal basis that does not exist.
The claim you will meet is that versioning is a compliance requirement. It is not. Section 98 imposes record-keeping duties on contracting authorities: an authority must keep records sufficient to explain a material decision, and must keep them until three years after the contract is entered into [9]. There is no corresponding statutory retention duty on suppliers anywhere in the Act. Any page telling you that you must version your bids to stay compliant is selling you someone else's obligation.
The real reason is better than the invented one, and it is a clock. Section 106 gives you 30 days from the day you first knew, or ought to have known, of the circumstances giving rise to a claim [8]. Inside those 30 days you may want to read the assessment summary the authority owes you [6] against what you actually submitted — and you can only do that if you still have what you actually submitted, in the state you submitted it.
That is the whole case: not compliance, but the ability to use a short statutory window that closes whether or not you are ready. It is also why version history matters more the closer a decision was.
Bid versioning covers the mechanics properly — branching, approvals, content libraries and the difference between versioning a document and versioning an answer. It is the better page on this subject and it outranks us on the term, so this chapter is a digest by design rather than by omission. Version control for bid content takes the content-library angle.
Sector by sector: where automation fits differently
The six jobs are constant. What changes by sector is which of them is your bottleneck, and in one sector the legal regime changes underneath you entirely.
NHS and healthcare
The correction first, because a lot of pages carry the error: NHS England has not been dissolved. Its abolition is being legislated and had not taken effect when this page was written; the Bill transfers functions to integrated care boards and the Department of Health and Social Care, not to individual trusts [30].
The fact that changes your work is older and firmer: healthcare services procurement runs under the Provider Selection Regime, in force since 1 January 2024, which sits outside the Procurement Act [29]. Automation tuned to the Act's notice types and deadlines does not map onto it cleanly. NHS tender software has the detail.
Construction and FM
The works threshold is £5,193,000 [19], so a great deal of construction work sits below the full regime and moves through framework call-offs and below-threshold notices instead. The bottleneck here is rarely drafting — it is evidence assembly, because the same accreditations, method statements and site records are demanded in slightly different shapes each time. Construction bid management goes further.
Local government
Sub-central authorities work to the £207,720 threshold and publish a high volume of lower-value work, which makes qualify the binding constraint rather than draft: there is more here than you can bid for, and the skill is declining well. Below section 87(4)'s £30,000 trigger the notice duties change again [13].
Technology and digital
The framework routes dominate, and they invert the model: a G-Cloud listing is a catalogue entry, not a tender, so “automating the response” means keeping service definitions accurate rather than writing prose. G-Cloud 15 was awarded on 6 August 2026 [32]. If you also sell AI systems into the public sector, ISO 42001 in tenders is increasingly where the questions are going.
What we can and cannot tell you about time saved
This is the section every other page in this category leaves out, and it is the reason to trust the rest of this one.
Every published benchmark for AI bid automation that we have been able to find is a vendor measuring its own product. We are a vendor in this category. Citing those numbers as evidence for our own effect would be circular, and we will not do it.
It is worse than a single conflict of interest, because the figures circulate. Trace the apparently-independent statistics in this market and they collapse into each other: a comparison page cites a win-rate report published by another vendor, which cites a survey run by a third, which measured its own users. The number acquires authority purely by being repeated across sites that all sell into the same category. A figure with three citations and one origin is not corroborated; it is laundered.
Two specific things we removed from this page rather than carry forward, and we would rather name them than quietly delete them:
What we can state without qualification is what the software does, not what it saves: requirements are extracted from the notice with the sentence each came from, prior answers are matched from your own library, and mechanical gaps are flagged before submission. How much time that returns depends almost entirely on how much reusable content you already have — which is why a single headline number would be misleading even if we had one, and why the vendors publishing one are not measuring the thing you care about.
We are too early to publish win rates from our own users, and we will not invent them. If an independent, named measurement of AI-assisted tender response time appears — a procurement body or an academic study rather than a vendor survey — we will cite it here and say whose it is.
What automation cannot do — the un-automatable inventory
Not “does badly”. Cannot, or must not. Seven of them, and the last one is the argument this page exists to make.
Seven things a machine cannot do for you, with the reason and — where there is one — the statutory consequence of pretending otherwise. The last row is the one this page exists to make, and no competing page in the SERP audit made it.
Is using AI to write a tender allowed?
Yes, and the government has said so in writing. It is also worth quoting the whole sentence rather than the first half of it, which is what the rest of this category does.
PPN 017, published 17 February 2025, addresses transparency of AI use in procurement. Its paragraph 9 says:
“There are potential benefits to suppliers using AI to develop their bids, enabling them to bid for a greater number of public contracts. It is important to note that suppliers' use of AI is not prohibited during the commercial process but steps should be taken to understand the risks associated with the use of AI tools in this context, as would be the case if a bid writer has been used by the supplier.” [22]
The second bolded clause is the half everyone drops, and it is the more useful half. The government's own framing puts AI-assisted bidding in the same category as hiring a bid writer: an ordinary supplier practice that carries risks the buyer may reasonably ask about. Not a loophole, not an endorsement — a normal thing, treated normally.
Three things worth being precise about, and three worth not saying:
What not to say: that AI bidding is “government-endorsed”. It is not endorsed; it is not prohibited, which is a different statement. Also worth noting: PPN 017 applies to central government departments, executive agencies and non-departmental public bodies. A council or an NHS body is not automatically in scope.
Assessment summaries: the feedback loop nobody automates
The Procurement Act created a structured, mandatory, per-bidder feedback artefact and handed it to every supplier that submitted an assessed tender. Almost nothing in this software category does anything with it.
Section 50(3) is the provision: before publishing a contract award notice in respect of a contract awarded under section 19, a contracting authority must provide an assessment summary to each supplier that submitted an assessed tender [6]. An assessment summary means information about the authority's assessment of your tender and, if different, of the most advantageous tender submitted.
The qualifier matters and is routinely dropped: the duty attaches to contracts awarded under section 19 — that is, following a competitive tendering procedure. It is not a universal right attaching to every award, and a page that states it flatly is overstating it.
Where it applies, though, look at what you are given: an assessment of your own bid, and an assessment of the winning one, from the buyer, in writing, before the award notice goes out. That is the highest-value input to your next bid for the same buyer, and it arrives on a schedule.
The automation opportunity here is not drafting. It is retention and comparison: keeping the submitted version, parsing the summary against it, and carrying what you learn into the next response for that buyer. That is a corpus and versioning job rather than a language job, which may be exactly why the category has ignored it.
Questions people actually ask about tender response automation
AI-generated drafts are designed as starting points, not final submissions. The automation handles structural compliance, formatting, and content matching — bid managers add commercial judgement, win themes, and buyer-specific positioning before submission.
Documents are encrypted at rest and in transit, and your content library is not used to train models or shared with other organisations. We hold no security certifications yet — no ISO 27001, no Cyber Essentials — and we say so rather than implying otherwise, because buyers use those as pass/fail gates.
Yes. Above-threshold UK tenders (£135,018+ for central government goods and services; £207,720+ for other public bodies, in force from 1 January 2026) published on Find a Tender are ingested automatically and processed by the same automation pipeline.
First draft generation typically completes in under 30 minutes for standard ITT and PQQ documents. Complex multi-lot tenders with 50+ questions may take longer depending on document size.
Not unless you are asked. PPN 017 creates no free-standing duty on suppliers to volunteer that AI was used; it gives contracting authorities optional example disclosure questions that can be added to an invitation to tender. If a question is asked, answer it truthfully — and note that PPN 017 puts supplier AI use in the same category as using a bid writer, which is to say a normal practice with risks a buyer may reasonably ask about. PPN 017 applies to central government departments, executive agencies and non-departmental public bodies.
It can, and this is the most under-used opportunity in the category. For contracts awarded under section 19 of the Procurement Act 2023, the authority must provide an assessment summary to every supplier that submitted an assessed tender before it publishes the contract award notice. Reading that summary against what you actually submitted is only possible if you kept the submitted version — which is the real argument for versioning, and it matters because the challenge window under section 106 is only 30 days from when you knew or ought to have known.
Where to go next
Each of these owns a job this page deliberately does not do.
The category map
What the tool categories are, and which one your problem lives in.
Best RFP software
The ranked view, with the criteria shown.
Software comparison
Head to head, once you have a shortlist.
AI tools for bids and tenders
Eleven UK products against eleven criteria — including where we lose.
Tender software for SMEs
The small-team lane, where qualifying out matters more than drafting.
Bid management and pipeline
Running a portfolio rather than a document.
Public-sector tender software
The public-sector framing specifically.
Bid versioning
Chapter 9's subject, properly.
The Procurement Act 2023
The statute under all of the above.
Find a Tender, ingested
Notice monitoring and alerting — the find job on its own.
ITT response software
The invitation to tender specifically, with compliance as a gate.
PQQ and SQ software
The selection stage, and what the platform now holds for you.
NHS tender software
Where the Provider Selection Regime changes the rules.
Construction bid management
Evidence assembly rather than drafting.
G-Cloud 15
Where a response is a catalogue entry, not a tender.
ISO 42001 in tenders
Increasingly asked of anyone selling AI into the public sector.
Social value in procurement
The commitments chapter 12 says never to let a model invent.
How to write a tender response
Writing to published criteria, in the criteria's own order.
Sources
Every bracketed number in the body resolves here. Legal claims cite the provision, not a summary of it; guidance claims cite the guidance and its publication or update date. All accessed 17 August 2026.
- Procurement Act 2023 (c. 54), legislation.gov.uk. Sections cited individually below.
- The Procurement Act 2023 (Commencement No. 3 and Transitional and Saving Provisions) (Amendment) Regulations 2024 (SI 2024/959) — the Act in force 24 February 2025.
- Procurement Act 2023, s.19 — Award of public contracts following a competitive tendering procedure. The most advantageous tender test.
- Procurement Act 2023, s.23 — Award criteria.
- Procurement Act 2023, s.20 — Competitive tendering procedures. s.20(2) defines the open procedure and the competitive flexible procedure.
- Procurement Act 2023, s.50 — Contract award notices and assessment summaries. s.50(3): the duty arises “in respect of a contract awarded under section 19”; s.50(4)–(5) define an assessment summary and an assessed tender.
- Procurement Act 2023, s.51 — Standstill periods on the award of contracts. Eight working days.
- Procurement Act 2023, s.106 — Time limits on claims. 30 days from when the supplier first knew or ought to have known.
- Procurement Act 2023, s.98 — Record-keeping. The duty is on the contracting authority; records kept until three years after the contract is entered into. No corresponding supplier-side retention duty appears in the Act.
- Procurement Act 2023, s.22 — Conditions of participation, including the “proportionate means” test and the limits on requiring audited accounts and pre-award insurance.
- Procurement Act 2023, s.57 — Excluding suppliers from a competitive award, with Schedule 6 (mandatory exclusion grounds) and Schedule 7 (discretionary exclusion grounds).
- Cabinet Office, Guidance: Debarment, on the debarment list established under the Act and how authorities must consider it.
- Procurement Act 2023, s.87 — Regulated below-threshold contracts: duty to publish notices. s.87(4): £12,000 for a central government authority other than a Welsh one, £30,000 otherwise.
- The Procurement Regulations 2024 (SI 2024/692), reg. 6 — Sharing core supplier information through central digital platform. reg. 6(3) requires the confirmation before the end of the tendering period in an open or competitive flexible procedure; reg. 6(5) sets out the three steps quoted.
- The Procurement Regulations 2024 (SI 2024/692), Part 2 — regs. 9 to 12 define the supplier's basic information, economic and financial standing information, connected person information and exclusion grounds information.
- Government Commercial Function, Central Digital Platform — factsheet (HTML), updated 23 June 2026. Source of both quoted phrases.
- The Procurement Act 2023 (Commencement No. 4) Regulations 2025 (SI 2025/1316). Reg. 2(1) commences s.69 (other than Welsh-regulated procurement) and s.71 on 1 January 2026; reg. 2(2) commences Welsh-regulated s.69 and s.70 on 1 April 2026.
- Procurement Act 2023, s.71 — Assessment of contract performance.
- Cabinet Office, PPN 023: 2026 Threshold Amounts, published 27 November 2025. Annex A, summary of threshold amounts from 1 January 2026. PPN 023 makes no statement about VAT.
- Procurement Act 2023, Schedule 1, as it stood at 24 February 2025 — the previous threshold figures, for the comparison only.
- The Procurement Act 2023 (Threshold Amounts) (Amendment) Regulations 2025 (SI 2025/1200). The below-threshold amendment is to s.85(3), not to s.87(4).
- Cabinet Office, PPN 017: Improving transparency of AI use in procurement, published 17 February 2025. Paragraph 9 is quoted in full; Annex B carries the example disclosure questions.
- Cabinet Office, PPN 002: Taking account of social value in the award of contracts.
- Cabinet Office, PPN 026: The social value model, published 5 August 2026, applying to procurements commenced on or after 1 January 2027.
- Cabinet Office, PPN 006: Taking account of Carbon Reduction Plans in the procurement of major government contracts.
- Cabinet Office, PPN 019: Requirements to publish on Contracts Finder — new below-threshold opportunities go to Find a Tender.
- Find a Tender: notice types.
- House of Commons Library, Public procurement statistics (CBP-9317). Cited for the consolidated Whole of Government Accounts figure only; this page does not reproduce the SME-share percentages the briefing re-reports from third parties.
- The Health Care Services (Provider Selection Regime) Regulations 2023 (SI 2023/1348) — healthcare services procurement outside the Procurement Act.
- Health Bill 2026-27, bills.parliament.uk. The vehicle for abolishing NHS England. Its parliamentary stage could not be re-verified from this seat on 17 August 2026 (the site blocks automated access), so this page asserts only that the Bill had not taken effect and that functions transfer to integrated care boards and the Department of Health and Social Care, not to individual trusts.
- Government Commercial Agency, G-Cloud suppliers' guide.
- Government Commercial Agency, G-Cloud 15 (RM1557.15) — framework award 6 August 2026. The agency's own domain is the source for its renaming from Crown Commercial Service.
- The Public Contracts (Scotland) Regulations 2015 (SSI 2015/446), reg. 67 — Contract award criteria. The most economically advantageous tender test, still in force for devolved Scottish procurement.
- Procurement Act 2023, s.126 — Extent, with s.2 — Contracting authorities, which excludes devolved Scottish authorities.
- Cabinet Office, Departmental Small Business Procurement Targets. Referenced as the current, primary, forward-looking SME figures; no baselines are published, so they cannot be converted into an implied current share.
- Procurement Act 2023, s.54 — Time limits. Minimum tendering periods.
- Procurement Act 2023, s.16 — Preliminary market engagement, with s.17 — Preliminary market engagement notices.
Corrections policy: where a figure on this page turns out to be wrong we append the correction and say what was believed before, rather than editing the mistake away. Four figures carried by the previous version of this page were deleted in this rebuild rather than corrected, because none of them had a source that survived being looked at — they are named in “What we can and cannot tell you about time saved” above. Two claims on that version were factually wrong and are corrected in place: the below-threshold trigger (£25,000 was the old Public Contracts Regulations figure; s.87(4) sets £12,000 and £30,000) and the status of NHS England, which has not been dissolved.