The end of source-to-pay: procurement will architect the system and operate almost none of it

Procurement technology and the procurement function have evolved in step for thirty years, because capability lived inside tools and the tools sat in procurement. That constraint is disappearing, and it changes what the function is for.
1. I counted
At the start of October, when I gave the closing keynote at DPW Amsterdam, there were 955 procurement technology vendors in the world that I could find. I know because I had spent two years looking for them.
The estimates in circulation are much larger. Two thousand is common, three thousand is not unusual, and I have seen four thousand quoted. I counted because I had spent years on the buying side, building and buying procurement platforms at BT and at Maersk, in a market too noisy for me to make sense of. Finding every vendor and holding all of them against a single framework was, in hindsight, a fairly stupid way to spend two years, but it was the thing I needed and nobody had built it.
The best picture of this market before that belonged to Dr Elouise Epstein. Her Modern Procurement Platform Stack, better known as the spider map, first appeared in 2017, and anyone who has tried to explain to a CFO why procurement technology is complicated has probably reached for a version of it. Last year’s edition holds so many vendors that nobody is meant to read it, and that is its point.
Elouise describes four eras of procurement technology. This article picks up where they leave off, because I believe a fifth era arrived while the industry was still arguing about the fourth, and a sixth is beginning now.
2. Thirty years, two evolutions
Two evolutions have run side by side for about thirty years. One is the technology and the other is the procurement function itself. The industry discusses both constantly and almost always separately, yet placed next to each other they turn out to have the same shape.
In the first era both were dispersed. The technology consisted of point tools, each doing one thing simply and well, while buying was an activity nobody owned, scattered across finance, legal and whoever happened to be good at haggling (or negotiating, for the professionals reading).
Consolidation followed on both sides. Suites bolted the point tools together, which was orchestration before anyone had the word for it, and at about the same time businesses worked out that the money they spend drives both margin and risk exposure. They pulled buying into the middle, professionalised it and put it behind guardrails.
The third era brought specialisation. Cloud-native, best-of-breed products offered real depth in a single capability in place of a shallow version of everything, and category science arrived alongside them, with Porter, Kraljic and the purchasing chessboard. Somewhere in this era the number of tools overtook the number of people. Nobody noticed at the time, and nobody has undone it since.
Somewhere in the third era the number of tools overtook the number of people.
The fourth era began when AI arrived and the technology, for the first time, stopped sitting beside the people and started to wrap itself around them. The augmented category manager belongs here, and this is the point at which two evolutions that had run in parallel for thirty years became one.
In the fifth era the technology moves all the way inside, so that it stops being a tool the team uses and becomes part of how the team works. Once that is true it has no reason to stay in the middle, and in the sixth era it scatters back out across the business, still fused with the people doing the work and now connected. I call that sixth era federated.
None of these eras ended on a particular day. Most large organisations are running three of them at once, with a suite at the core, a handful of specialists around the edge and AI arriving in all of them, which is part of why the change is so hard to see from the inside.
3. Where capability is allowed to live
Underneath all six eras, technology has only ever answered one question, which is where capability is allowed to live.
For thirty years the answer never changed. Sourcing capability lived inside the sourcing tool, contract capability lived inside the contract tool, and the tools sat in procurement. Anyone who wanted the capability had to come to where the tool was, so the function organised itself around its tools every time they changed.
That constraint is going. Capability that lives in a model can live anywhere: in a workflow, in a browser, in a conversation, or in the hands of somebody who has never had a procurement job in their life. What sits inside a sourcing tool today, used by a handful of people in procurement, can sit inside the tools the rest of the business already lives in, whether that is Teams, Copilot or whichever AI assistant they happen to have open.
Most of us are treating the biggest change of our careers as a software purchase.
Nothing like this has been true at any point in those thirty years, and I think it is the biggest change any of us will see in our careers. Most of us are treating it as a software purchase.
4. Source-to-pay is one road
Source-to-pay is the first thing that has to go. It has defined how this market describes itself for twenty years, in every vendor category and every RFP template, and it covers only a fraction of what happens.
The difficulty is that source-to-pay describes a process and we have been using it to describe a market. That is like finding your way around a whole city with a map of one road, and then wondering why you keep getting lost.
Comparing 955 of anything requires one thing to hold them all against, and it has to reflect where the market is now, in the fifth era and heading into the sixth. The model I built has three bands. A foundation layer sits underneath everything and never switches off, the flow of work runs across the middle in the order the work moves, and a third band holds the intelligence and control that cut across all of it.
Those bands hold twelve capability areas, and beneath them sit 96 capabilities, which is the level at which we assess every vendor. The shape is revealing before it holds a single vendor, because depth follows need: intake and orchestration has twelve capabilities and data foundation has ten, so the structure alone shows where the market has crowded in.
One question falls out of the model that cuts through any vendor conversation: if this were taken away tomorrow, what would stop working? If nobody in a demo can answer it clearly, the buyer has learned something more useful than anything in the deck the vendor spent three weeks building.
Every one of the 955 vendors sits against that model. DPG Labs has also captured more than 40,000 changes to what vendors say about themselves, which shows when a story shifts. I wanted one objective place to see all of it at once, and I think our world needs a standard, so I built one, and no vendor paid to be in it.
5. What they claim and what they can prove
Across all 955 vendors, nearly every one claims nearly all of it. Parity at the level of the claim is now close to total, which means the question at the heart of every RFP has stopped working. We ask what a vendor can do, and every answer is yes.
We ask what a vendor can do, and every answer is yes.
The differences appear underneath the claims. At the time of the keynote, 642 of the 955 had AI somewhere in the stack. Of those, 538 had added AI to a product they already had, and 104 had rebuilt around it.
From the outside the two groups are indistinguishable. They use the same words, make the same claims and often give the same demo. The gap is invisible from a website and invisible from a pitch, and it is probably the most useful single thing a buyer could know.
None of this is an accusation. Adding AI to an existing product is very often the right commercial decision, because rebuilding is expensive and slow, and it breaks things. Neither route is wrong, but a buyer should know which one they are buying, and today most cannot.
The same pattern showed up a few weeks later in what vendors post about: 33% of them had talked about agentic AI, and about 50 had built their product that way.
Six years ago I would have believed what I was told, and carelessness had nothing to do with it. I could not judge the capability, so I judged the pitch.
6. The CPO who stopped buying
One conversation has occupied me more than any of the data. Some time ago I had lunch with a CPO I have known for years, at a household-name multinational with more than a hundred thousand employees, and he told me he had stopped buying procurement software. He had not slowed down or consolidated his stack; he had stopped altogether.
Two engineers work for him, reporting to the CPO and not to the CIO. They build exactly what the function needs and deploy it in days, with no pilot, no proof of concept and no centre of excellence with a steering committee, because that is simply how his organisation now runs.
His is a US business, so data sovereignty never came up over lunch, whereas in a European company it would come up in the first five minutes. I would not treat his approach as a template. I would treat it as a signal of how fast the floor is moving.
Every industry has pioneers and settlers, and the pioneers do the hard yards that prove what is possible. Most procurement functions cannot do what he has done, and will not be able to for five or six years, because they will not get that engineering talent, they are carrying technical debt, and they do not yet have a clear enough vision. I mean that as a description of where most functions are, with no criticism intended.
The pattern from here is familiar. Pioneers prove an idea, the supply market turns it into a product, and it arrives as something everyone else can buy. A wave of vendors that are AI native, composable and built to connect is the likely result.
7. Buying well in the meantime
Most functions will therefore keep buying technology for some years, in a market where every vendor says yes to everything. The data suggests three habits worth forming now.
The first is to ask how a product came by its AI, and to ask for the evidence: what the vendor rebuilt, what it added, and when. Neither answer rules a vendor out, and a vendor who cannot or will not answer has told the buyer something as well.
The second is to judge a product on whether its capability can leave its own screen. If the sixth era is arriving, a capability that works only inside the vendor’s interface still depends on the old constraint, and “composable” and “built to connect” become buying criteria as much as architectural ones.
The third is to write the guardrails down before anyone needs them. Buying spread across the business works only inside rules that procurement has designed: which suppliers and terms it has already approved, which decisions anyone can take alone, and which need a second pair of eyes. Rules that live in a category manager’s head cannot travel to where the buying happens.
8. Architect the system, operate almost none of it
Within five years, and possibly sooner, I expect procurement to architect the system and operate almost none of it.
Stakeholders will do the buying themselves, with AI, at the moment the need arises and inside the tools they already use. They will run a sourcing event, onboard a supplier or agree a contract without opening a procurement system, filling in a procurement form or waiting two weeks for a category manager to become free, and they will do it inside guardrails that procurement has architected.
Architecting is real work, and harder than operating. Somebody has to decide where each capability lives, how the pieces connect, what the business can do alone, and how the data returns so that one team can still see the whole.
Control no longer requires centralisation.
What makes this possible is a change in the trade the function made a generation ago. Procurement centralised to gain control, giving up proximity to the business in return for governance, consistency and leverage. It was the right trade for thirty years, and the function no longer has to make it, because control no longer requires centralisation.
To many people in procurement that will sound like being made redundant, and for about a fortnight it sounded that way to me.
What changed my mind was an apparently stupid question, which is why procurement exists at all. It exists because the company buys things, and a company buys things only because it sells something; a company that sells nothing has no purpose and buys nothing. Every negotiation, every contract and every saving any of us has booked sits downstream of selling, and always has, although we rarely describe it that way.
The closest anyone can get to the value is therefore the place where the selling happens. The most commercially astute, most technically aware person in an organisation does not, in my view, belong in a central procurement function. That person belongs out in the business, and I would call the move a promotion.
For thirty years the profession has asked how to get a seat at the table. Every conversation in which somebody is about to spend money is a table, any large organisation has hundreds of them, and we have been queueing outside one door.
Based on my closing keynote at DPW Amsterdam, 1 October 2026. The figures are as they stood on the day; DPG Labs now tracks 992 procurement technology vendors.


