A commercial lead at a county council sits down on a Monday to buy a case-management tool for adult social care. It is the repeatable, non-bespoke kind of thing, a service (a thing government does for or to people, from a passport to a benefit) that another council has very likely already bought. The honest answer to "how quickly can I have it" is not days, and not weeks, but, for anything needing a fresh competition, most of a year. There is no fast lane: there is a lane marketed as fast, a lane that is genuinely fast and quietly expensive, and a marketplace being built that is not finished. The thing that actually helps her on the Monday is none of those. It is being able to see what the estate already bought.
This is the fourth in a series about reuse in UK government. Parts 1 to 3 were about the code the public sector writes: publishing it, searching it, and mapping what the whole estate is built on, where reuse meant "do not rebuild the button." Part 4 carries it to the contracts government buys: "do not re-buy, in the dark, what has already been bought."
The receipts were always public. Every tender, every award, every amendment is published openly, under the Open Government Licence, across five separate portals: Contracts Finder, Find a Tender, Public Contracts Scotland, Sell2Wales and eTendersNI. The trouble is that "published" and "usable" are not the same thing, and the publishers say so themselves: the Open Contracting Partnership records the Cabinet Office admitting that before the Procurement Act there was no single picture of procurement, only "disparate and disconnected data sets" and "little insight" despite "lots of data." We have been here before: in 2010 the government published every item of central spend over five hundred pounds and waited for an "army of armchair auditors" who never turned up. Publishing, it turns out, is necessary but not sufficient; the missing thing was never transparency, but the ability to ask the receipts a question.
So over a weekend, on data that was already open, I built a small tool that does that. It is called uk-tenders-mcp, and it stacks around 677,000 contracting processes from those five portals onto one shared key, queryable over MCP, the Model Context Protocol (the open standard that lets an AI assistant query a data source directly). This is not the first time anyone has worked this data: Tussell sells analysis built on it, the Open Contracting Partnership has for years, and the government's own Central Digital Platform now consolidates the new regime's notices. The new thing is narrower, this particular cut: free, AI-queryable and deduplicated across the five. The endpoint is a public, read-only attributed mirror under the Open Government Licence v3.0, refreshed nightly with personal data redacted, but not the authority of record.
What you can actually do with it on Monday
The how is genuinely small, and I think that is the point. There is no login and nothing to install. You point your AI assistant at the live address, tenders.run.cns.me/mcp, and ask in plain English; the code, if you want to read it or run your own, is open at github.com/chrisns/uk-tenders-mcp. If you have an assistant that speaks MCP in front of you right now, you can try it before you finish reading this paragraph, which is rather the test of whether legible data is actually legible.
So picture the council lead doing exactly that. She types the thing she would otherwise ask a colleague she does not have:
Which councils have bought an adult social care case-management system in the last three years, who supplied it, and what did the awards top out at?
What comes back is not an essay and not a verdict; it is a shape. For example, a short table with one row per award: a handful of peer authorities down one side, the supplier each of them awarded to, the award ceiling as an order-of-magnitude benchmark, the date, and, on every row, a link to the official notice so she can verify any of it at source. That is the difference between walking into a meeting with a guess and walking in with a reference range. She has gone from a blank page to a shortlist of peers and a name to call, in the time it took the kettle to boil.
The same address answers the other questions she has standing at her desk. What is closing soon in her category, drawn from the live pipeline of several thousand open notices, so she can see what is genuinely on the market this week rather than last year. Total awards by buyer, by supplier, by category and over time, so she can see whether spend in her area sits with one or two suppliers or is spread thin. Who bought a given thing and who supplied it, and the same the other way round. None of that is a procurement. It is situational awareness before she starts: the peer authority that already bought it, and a name to call.
What it is not for matters just as much, and I would rather be honest about the edges than oversell the middle. The values are contract ceilings, not actual spend, so anything load-bearing gets checked on the official notice that each result links back to. It is a mirror of the record, not the record itself, so the notice is always the thing that counts. The names are not normalised, so the same supplier may appear under three spellings. And it cannot let her prefer a past performer when she comes to score a bid. It tells you where to look and who to ask, not what to buy.
The thing that makes all of this possible is the Part 3 move again. Each portal gives a buy its own publisher-scoped identifier, so one real purchase published twice looks like two, until one model collapses those duplicates into roughly 24,000 deduplicated clusters. We have not added a single new fact to the record. We have only made it answerable.
The lane that is marketed as fast
A framework is shared purchasing infrastructure that multiple buyers can call off from: a pre-agreed list of approved suppliers and prices. Calling off means buying from that list without re-running a full tender.
It is sold as the shortcut. But the documented timings tell a quieter story. The front end is the part that hurts: producing the requirements and the invitation to tender, the guidance notes say, "can take from a few days to a few months." The Government Commercial Agency's assisted further-competition service then runs to a thirty-working-day service level, around six weeks, excluding both ends; and before signing there is a mandatory eight-working-day standstill under the Procurement Act 2023, even within a pre-selected pool.
Stack those documented minima together with the business case, the approvals, the finance and legal sign-off, and supplier mobilisation, and the "shortcut" credibly reaches around eight months; eighteen months for a pilot worth tens of thousands of pounds is not unheard of. That is a practitioner aggregate built from published stage timings, not an official statistic: the Office for National Statistics holds no central data on average tender duration, so treat it as widely reported, not a government average.
The constraints matter as much as the clock. A framework sets its terms and prices in advance, so because the price is fixed you cannot haggle; the ready-made terms only work if they already suit. The supplier pool is closed, the term is capped at four years for most agreements, and you fit your requirement to the framework, not the reverse. None of this makes frameworks useless: a genuinely small, low-risk call-off from a fixed-price agreement, taking the listed terms as they are, can complete in days, because the notice and selection overhead really is removed. It is the call-off that can be fast, not the fresh competition; and the convenience quietly costs competition.
The lane that is genuinely fast
The fastest route of all is not to compete, and to award to the supplier already in the building. It is quick. It is rarely a good deal. The National Audit Office found that 37 per cent of contracts in 2021 to 2022, and around 39 per cent by value, were awarded without competition. That is the NAO's own category, bundling direct awards, single-bid tenders and frameworks used without further competition; it describes how things were bought, not in itself a verdict of waste. But the diagnosis of why is blunt: authorities are "forced into extensions through failing to plan early enough," and Tussell finds framework awards growing while open competition shrinks. The early data under the Act points the same way: direct awards rose from 15 per cent of procedures in March 2025 to 24 per cent by May. Sadly, the fast lane is becoming the default lane, the one where the public most often does not get the best price.
Knowing who actually delivered
Here is where the open record starts to earn its keep, and for all procurement, not only digital. Under the Procurement Act 2023, public contracts over five million pounds must have at least three KPIs set, published, and assessed every twelve months, rated from Good down to Inadequate, and persistently poor performance can become a discretionary ground for excluding a supplier. So the published record may now tell you something it never could before: not only who won, but, in places, who delivered.
I should not overclaim this. The debarment list in statute is currently effectively empty. The KPIs bite only above five million pounds and only on new contracts, the rating is self-assessed, and the Act bars conditions that require a prior award by a particular buyer, so past performance enters mainly as a negative signal, never as a preference you can bake into the score. What the open record gives you is not a shortcut in the award. It is the name of someone to phone: if a neighbouring authority bought the same thing and rated it well, you have not found a procurement preference, you have found a colleague who knows whether it was any good.
The lane we are building
For digital commodities specifically, there is a proper destination on its way, and I should be open about this: it is work I am leading. In June 2025 DSIT announced that a one-stop shop for tech could save taxpayers 1.2 billion pounds a year: the National Digital Exchange. One front door instead of a thicket of frameworks. Pre-approved deals at nationally negotiated prices, so a team does not re-run the same competition the estate has already run a hundred times. A way to match that team to the suppliers that fit what it actually needs in a matter of hours rather than months. It is led from within DSIT's Government Digital Service as part of the Digital Commercial Centre of Excellence, itself a product of the blueprint for modern digital government, and we are building it in the open, the same way the rest of this argument says government should buy. The 1.2 billion pounds a year is the prize, and it is the right ambition to be aiming at.
For the digital case-management tool our county council lead wants, this is the answer taking shape. It is in beta, the code and the catalogue out where you can see them: not yet open to everyone, but genuinely on its way, and closer every month. That is the honest meaning of the word "yet", the cleanest reason I know to keep it. The fast lane is real, we are pouring the road, and it is nearly here. The one thing it will not do is everything: the National Digital Exchange is digital by design, so it is no help for her facilities contract or her agency-staff spend. That is the load-bearing limit, and it is exactly why seeing what the estate already bought matters for all of her procurement, and not just the part of it that happens to be digital.
What I am adding next
Two things, briefly. The first is the framework thicket. The convenient story is that you go to the Government Commercial Agency and the problem is solved, but the GCA is only roughly a quarter of the routes: about a fifth of the 125 billion pounds a year of common goods and services on the NAO's reading, around 26 per cent on its own self-reported share (different denominators, not to be stitched into one). The majority runs through public buying organisations like ESPO, YPO and Scape, and commercial operators that run or sit on public frameworks, such as Bloom, Constellia, Alexander Mann Solutions and Pagabo. The overlap is the point, not any wrongdoing: the NAO estimates between 8,000 and 21,000 frameworks exist, and Tussell counted 1,856 used to award in 2025. No buyer has a map of it.
The second is government procurement cards: a pre-authorised spend route, not a procurement. In March 2025 the Cabinet Office froze most of the roughly 20,000 cards after card spend "quadrupled," surpassing 600 million pounds, and banned card use where a proper procurement route exists: about as plain an admission as you could want that it bypasses procurement. Those transactions over five hundred pounds are published monthly: another record waiting to be made legible.
None of this required me to open anything; the doctrine was settled years ago. The Service Manual's twelfth standard says make new source code open because public services are built with public money, and that chain (which gave us the GOV.UK Design System, worth around 22 million pounds a year) has no honest reason to stop at the line between what we write and what we buy.
A word of caution, because the armchair-auditor lesson still applies: a tool that should change behaviour may simply sit unloved, as the 2010 spend data did. Legibility makes a thing possible; it does not make anyone use it. But for the commercial lead at her desk on the Monday, the calculation has shifted. She still has no fast lane, but she can now see what the estate already bought, and find the name of someone who can tell her whether it worked. That is not the whole answer, but it is the end of deciding blind.
Nobody in public service should have to start from a blank page alone. And nobody should have to re-buy, in the dark, what the public has already paid for.
(Views in this article are my own.)