Practitioner note · 01 September 2026
Four people, four questions, one desk. The ticket machine is the only part of this that is a joke.
A category strategy is a decision. A buying channel is that decision compiled into something a system can execute and a requester can find. The first gets written. The second usually does not.
The piece is mostly about indirect spend: a requester who needs something, a catalogue to find it in, a threshold, a marketplace. One term recurs throughout. Operable content means the version of a category strategy a system can act on without asking a person: the item, the price, the supplier, the routing, the threshold. It is not the material master. Materials management governs what a thing is and how many to hold. This is about whether the deal you struck is reachable by the person trying to buy.
The argument transfers even though none of the mechanics do. In direct, the operable content is the source list, the info record, the price condition and the scheduling agreement. A rate gets negotiated and never reaches the record the system reads when the order is created. Different objects, same missing job, and nobody owns that one either.
Procurement names an owner for every contract, supplier, category and risk. It has never named one for the thing the business actually buys from.
The tidy explanation is that nobody owns the buying channel. It falls between category management and P2P operations and appears in neither set of objectives. Clean argument. Survives a slide. Survives a steering committee. I have heard it in three organisations.
In my experience it is wrong. Category managers own the buying channel and they know it. It follows directly from the category decision and nobody else can make it. Ask one, and they will tell you it belongs in their strategy.
What they do not have is time. That is a harder problem, because a RACI does not create any. The 2026 benchmark has procurement workloads rising eight per cent while both headcount and operating budgets fall. Nobody is holding a spare Tuesday.
Sourcing, contracts, suppliers, master data and policy all have owners. The buying channel they add up to does not, so every question about it lands on the same desk, at a moment that desk did not pick.
Every one of those five is governed and audited. The order at the far end has an owner and a cycle-time target of its own. What none of them owns is the work of turning the five into something that last one can execute without asking a person. That work does have names. One government procurement lifecycle gives it a stage of its own, with the owner left blank. A major ERP splits it across two seeded roles, neither of them the category manager: one holds the catalogue and the priced agreements, the other the thresholds, supplier statuses and routing. Even the vendor cannot decide whether this is one job. What it has nowhere is an owner in the business, and work without an owner gets no budget line.
Two ends, both governed. One middle, belonging to nobody.
Never the category manager who wrote it. It is a requester who cannot find what they need, or one who finds it at a price that does not match the quote in their hand, weeks after the decision that caused it. By the time it reaches a report it has been renamed as a data issue, or an adoption issue, or a change issue. It is none of those. It is a translation nobody wrote.
Some credit first. Category managers are not underperforming at this. They are absorbing it. Every gap in the buying channel arrives at their desk as a question, an escalation or a chase, at a moment they did not choose, and they answer it because they are the only person who can. It is a queue with no ticket machine and no closing time.
Absorbing it is not a discipline problem. It is a role designed so that the cost of missing operable content lands on the one person least able to make time for it.
Nobody has measured what that costs. No time study, no work sampling, no diary study exists for this role anywhere. The only published figures are four US state government position descriptions, which put between fifteen and forty-five per cent of the role against supporting internal requesters. Four documents, one country, public sector. It is not a benchmark and I would not pretend otherwise. It is the only number there is, and every practitioner I have shown it to has called it low.
Two things carry the return and neither appears on any scorecard I have seen. Purchase order escalations, meaning orders that stall and get chased or corrected. And requester support, meaning hours spent answering what the buying channel should have answered. Count those two before you change anything.
You can hand somebody the title. You cannot hand them the Tuesday it would take to use it.
A category strategy is written for procurement people. Then it goes downstream, where four different readers need it in forms it was never written in.
The academic literature has the same blind spot. The one 2025 study of category management across seventeen firms sets out to bridge strategic intent and execution, then defines execution as supplier selection and contracting. Sign and store, not operationalise. That is the problem, stated by people who did not notice they were stating it.
People say sourcing needs a preferred supplier list, and stop there. A list is the smallest part of it.
| Who reads it | What they need out of it | What happens when it never arrives |
|---|---|---|
| Sourcing | The route to market for each sub-category, and whether the next move is compete, renew, extend or hold. The demand picture, the tiering, the levers actually available, the should-cost position, and an event calendar tied to when contracts expire. The preferred supplier list is the output of all of that, not a substitute for it. | The contract expires and nobody updates the list. The same supplier keeps getting used, now gated, or worse, off contract entirely. |
| Order management | Which suppliers transact on which channel, and how: catalogue, punchout, price file, PO flip, marketplace, spot. The price basis, whether that is a contracted price, a rate card or a formula. Unit of measure, pack size, lead time. The thresholds, and where an order goes when it trips one. | The item is not there, or it is there at a price nobody has checked against the contract. |
| Accounts payable | Nothing at all, until something is wrong. Then all of it at once: the tolerance, what a valid match looks like, who resolves a mismatch, and the contract reference that proves the price was right in the first place. | Invoices stop, and somebody reconstructs the answer by hand. |
| The requester | One thing. To find it, at a price that matches, and to know it is allowed. | They buy it anyway, somewhere else. |
Two of those four consequences are measured and two are not. About half of purchase order lines never transact through a catalogue-enabled supplier. Invoice exceptions became the single biggest problem in accounts payable last year, and the share of invoices passing untouched has barely moved in four years. What nobody has published is the join. No study takes catalogue coverage or price accuracy as the cause and measures exceptions or off-channel buying as the effect. Both ends of this argument are counted. The arrow between them is not.
The middle column is not a summary of the strategy. It is the strategy, in the only form a system can act on, and it appears in nobody’s job description. Five things feed it, and each one bites in its own way when the compile never happens.
Master data deserves more than a line, because it is where the gap is easiest to see. Procurement classifies what it buys in a taxonomy built for analysis and reporting, and the most widely used one carries no attributes at any level, so it can bucket an item but it cannot describe one. Meanwhile the requester types sticky notes. In the standard format used to load a catalogue, the classification code is mandatory and the keyword field is optional, and the search runs on names, descriptions and keywords, not on the code. The code decides how the shelf is arranged. The words decide whether anything is found on it. Writing those words is in nobody’s job description, and most suppliers are never told the field exists.
So the strategy stays a document, and the buying channel it was meant to produce degrades quietly instead of failing loudly.
Ask how well the compliant route works and nobody knows, because the instruments do not exist.
One square is one requisition. The published number is the thin black stripe.
Rejection rates run three to five per cent and the number is close to meaningless. Nobody rejects a requisition, because rejection makes the requester start again. Approvers send it back, and most business approvals complete before procurement touches the order anyway, so the fix routes back through the requester and never shows up at all. Rejection turns up in a report. Sending it back turns up nowhere. Nobody publishes a return rate.
The second absence is structural. The CIPS category management cycle has no stage for catalogue or contract enablement, and the CIPS Global Standard splits the two by level as well as by segment: category management sits at managerial level with no systems content, while eCatalogues appear at tactical level, framed as using a system somebody else configured. The European public procurement competency framework makes the same cut, IT tools in one competency and category work in another.
One standard goes the other way, and it is not a professional body. The UK Government Commercial Function role profiles put catalogue and content management inside category management, and the wording escalates with grade: maintaining catalogues at the bottom, owning key aspects in the middle, setting the direction for the category at Grade 7. At the two grades above that, it disappears. The work is inside the role right up to the point the role becomes strategic.
Which sets up the part nobody has costed. The grades that carry this work are the junior ones, and more than half of supply chain leaders now expect agentic AI to cut entry-level hiring. Take those grades out and the work does not disappear. It loses the only place it was ever written down.
There is a temptation to answer all this with a better interface. Somebody tested that. Across two hundred and seventy-four users in four organisations, content quality, order processing and support quality predicted how much people bought through the buying channel. Usability did not. Screen speed and ease of navigation had no significant relationship with anything. One qualifier the same researcher published later, so you hear it from me first: usability does predict whether people accept a system. It just does not predict whether they buy on contract.
So handle the word simple carefully. It does not mean consumer-grade. It means the item is present, priced against the contract, and it arrives. Nobody is asking for a nicer screen. They are saying the thing they needed was not in it, which is an operable content problem, and it lands back on the person who owns the category.
A category strategy does not survive as a category. It survives as a list, a tier and a threshold, sitting in a system somebody keeps current.
The clearest working example is not procurement software. It is a marketplace. The large consumer-grade ones give an administrator three controls and only three. Preferred lifts an item in search. Restricted warns the buyer and can route for approval, and the purchase still completes. Blocked removes it. Only the third enforces anything. The other two are advice, offered at the moment nobody wants advice.
The mechanic is the part worth stealing. Those controls apply to individual items and individual sellers, up to ten thousand item identifiers inside one policy. Read that back as a procurement statement: somebody writes the list, keeps it current, and decides which items are preferred, which get a warning and which get stopped. That is the strategy itself, compiled.
True. Tail spend is commonly put at ten to twenty per cent of value across eighty per cent or more of the suppliers, and nobody is compiling a strategy for each one. They do not need to. Each category needs one decision about which route it takes, and that decision is the strategy.
Five routes, and the same tray underneath every one of them.
The gate sits at admission. Inside a route, execution carries no further friction. Which route a category takes is decided by its risk profile, and that is where most policies go wrong: they set friction by transaction value alone, when a small purchase in a regulated category can carry more exposure than a large one that is not. The one published case of a switch to risk-based routing is Canadian federal defence procurement, where average approval time fell from 164 days to 51.
The competitive-quote route is the one the market is building hardest, and it is worth having. It curtails the tail and pulls spend back toward suppliers you already qualified. It also carries a prerequisite nobody advertises: the agent assumes something upstream has already established that this requisition is off contract and off catalogue. It does not check. Point it at a thin catalogue and it will run a fast, elegant, fully automated competitive event for something you negotiated last year, and book the cycle time as a win.
Twenty-five kinds of AI capability now ship across these seven stages. The one stage with none is the one the other six all read from.
I went looking for every class of AI capability on the market this year and placed each one on the seven stages between a category strategy being agreed and an order raised against it. Twenty-five of them, and they are not evenly spread.
Agents apply policy at intake and route a request to a channel. They draft contracts, generate a category strategy from spend data, bundle sourcing events, award on a weighted score, and in one case negotiate price and payment terms across three exchanges with nobody watching. Every suite shipped a studio inside eighteen months for humans to build agents in. Not one shipped an agent that authors the operable content those agents run on. The closest any of them gets is turning a negotiation somebody already ran into a priced agreement, behind a human approval gate, in a category somebody already sourced by hand. We have automated the building of the machines and left the fuel to somebody’s spreadsheet.
The benchmark data agrees. The 2026 study naming where procurement AI deployment actually concentrates lists contract management, market intelligence and spend analytics. Operable content appears nowhere on it.
Every one of them named. Six stages carry something. One does not.
One near miss, because it is the honest test. In service management, not procurement, a product already generates a catalogue item from a plain sentence, question set and all. But that item is a request form, with no price, no supplier and no threshold in it, which is exactly where operable content begins. Meanwhile the catalogue product attached to the largest business network in the world still sells content collection, validation, enrichment and classification as a human support subscription.
The obvious argument is a warning: agents remove your ability to keep getting away with the old operating model, because the human hesitation it relied on was doing more work than anyone admitted. True, and not the useful part. It also assumes the agents work. Gartner puts roughly one hundred and thirty vendors of the thousands claiming agentic capability as actually having it, and expects more than four in ten agentic projects to be cancelled by the end of 2027.
The useful part is a mechanism. The translations never get written because writing them is slow, unrewarded work that needs someone who knows the category. A conversational agent can interrogate that person in their own language and turn the answers into operable configuration inside the platform’s guardrails. It does not take ownership away, it removes the reason the owner never gets to it, and because the output is configuration and not a document it is measurable from day one.
Between six and eighteen hours a week. That is the size of the prize, and it is the only number anyone has published.
Ownership without capacity is not ownership. It is blame with a name on it. Assert that category managers own the buying channel, fund nothing, and you have invented a new way to hold someone accountable for a job you never gave them time to do.
And a fair second objection: a shared guided buying channel does not decompose neatly by category. Thresholds, policy defaults and the search experience are common ground.
Where I land: you touch it, you own it, and the organisation funds the touching. Ownership asserted, capacity provided once, downstream requirements settled when the strategy is written so they flow through by design. Done that way it is not an extra job. It is the job that gives the rest of theirs back.
Fifteen to forty-five per cent of the role against supporting requesters. On a forty-hour week that is six to eighteen hours, and nearly all of it goes on answering questions the buying channel should have answered.
Compile the strategy once, into the forms the teams downstream can use, and you remove the cause instead of chasing the symptom. The questions stop arriving because the answer now sits where people were going to look anyway. The prize is somewhere inside six to eighteen hours a week per category manager. Where inside that range, nobody can tell you, because nobody has measured it going out or coming back. That is a reason to go and measure it.
The field has not mapped this either. The most complete review of purchasing strategy traces five levels, from firm strategy down to how a category manager will work with each individual supplier, and stops there. There is no rung for the buying channel, the catalogue or the transaction.
Compiling a category strategy into a buying channel has no name, which is part of why it has no budget. Source-to-Pay describes what procurement does. Procure-to-Pay describes what the business does. The part in between, from a category strategy being agreed to an order raised against it, belongs to neither.
I have started calling it S2O, Strategy-to-Order. Same grammar as the terms your audience already uses, so it needs no explaining in a meeting. Work with a name can appear in a target operating model. Work without one stays a by-product.
On findability: no published benchmark compares search success in enterprise catalogues against marketplaces. Where this piece says requesters cannot find things, that is my own observation and not a measured claim.
I am collecting practitioner answers, and I read every one. Nothing is published without asking you first.