Cost at the unit
Which cost is the cost
The estate publishes four different weekly costs to serve and they span $12.59m. more
Four costs, one estate
Each prices a different population with a different rate method · spread $12.59m a week
Four costs, one estate. more
It prices Every document the shared services estate carries, once, at the tower that carries it.
Why this one — It counts every document once, at the tower that carries it, so it has no double count to unwind and no selection to defend. It is the only one of the four whose population is the same object the operations pages, the queue and the film all count.
Open the tower estateIt prices Every unit of work the functional organisation counts, which is not the same object as a document.
Why it is not the one to quote — It double counts. A single purchase order is a unit in sourcing, a unit in payables and a unit in the ledger close, so the same document is priced up to three times. The population it prices is 1.63 times the document population, which is the size of the overlap, not new work.
Open the function mapIt prices What was actually booked to a cost center. It is the only one of the four that a finance system would recognise.
Why it is not the one to quote — It is incomplete by construction. It carries what was coded to these cost centers, and the work the towers do is not all coded there: shared platform, allocated management and anything absorbed inside a business unit never reaches the ledger lines. It runs 21.3 percent below the tower rollup for that reason.
Open the rate cardIt prices Only the outcome chains that could be assembled end to end from evidence. It is the most rigorously costed of the four and the least representative.
Why it is not the one to quote — It prices the work that assembles cleanly. Work that could not be assembled into a chain is not in it at all, which is exactly the expensive, broken work you would most want priced. Scaling it up multiplies that selection bias by fifty.
Open cost per outcomeAnnualized figures are the weekly figure multiplied by fifty two. more
Published volume against the daily ledger
900,433 published · 921,379 on the ledger year average · gap +20,946 (+2.3%)
The towers publish a weekly transaction count. The daily ledger carries 52.1 weeks of arrivals. Those two numbers had never been set against each other in one place, which is what let the deep week and the published week be quoted interchangeably. They are not the same number and they were never meant to be: the published week is the ledger year average, not the deep week. Read that way, 10 of 12 towers agree within one percent. 2 towers carry +22,020 of the +20,946 total gap between them; everything else nets to −1,074, which is 0.12 percent of the estate and is rounding, not a finding.
| Tower | Published a week | Ledger deep week | Ledger year average | Gap to published | Gap |
|---|---|---|---|---|---|
| R2RRecord-to-Report Tower | 87,512 | 81,990 | 103,427 | +15,915 | +18.2% |
| O2COrder-to-Cash Tower | 107,549 | 99,410 | 113,654 | +6,105 | +5.7% |
| H2RHire-to-Retire Tower | 126,636 | 113,910 | 126,336 | −300 | −0.2% |
| ITSIT Service Tower | 85,277 | 83,505 | 85,025 | −252 | −0.3% |
| A2LAwareness-to-Lead Tower | 51,156 | 47,158 | 50,943 | −213 | −0.4% |
| C2LConcept-to-Launch Tower | 51,438 | 46,269 | 51,243 | −195 | −0.4% |
| S2PSource-to-Pay Tower | 142,093 | 125,988 | 142,260 | +167 | +0.1% |
| AIOAI Operations Tower | 68,456 | 61,550 | 68,327 | −129 | −0.2% |
| D&AData and Analytics Tower | 33,397 | 30,619 | 33,275 | −122 | −0.4% |
| LGLLegal Operations Tower | 40,630 | 37,260 | 40,587 | −43 | −0.1% |
| P2PPlan-to-Perform Tower | 25,194 | 23,163 | 25,211 | +17 | +0.1% |
| M2DMake-to-Deliver Tower | 81,095 | 71,526 | 81,091 | −4 | −0.0% |
| All towers | 900,433 | 822,348 | 921,379 | +20,946 | +2.3% |
The deep week column is the seven days the operations pages sample. more
Banked against forward
What the change has already taken out, and what is left to take
The two are different in kind, and the difference matters more than either figure. $18.85m a week is banked: the towers already run at $9,075,562 against a stated before of $27,926,042. $184k a week is forward: it does not exist yet, it needs $4.00m of build to obtain, and it pays that back in 21.7 weeks. Forward is 0.98 percent of banked. The honest reading of that ratio is not that the remaining work is worthless. It is that this estate is close to its modeled ceiling, and the next dollar is far harder to find than the last one was.
If inference costs more than modeled
Inference is 21.0 percent of the unit cost · break even at 10.91 times the modeled price
Every cost on this platform rests on one modeled input nobody here has metered: the price of inference. It is $1.90m a week, $98.92m annualized, and 21.0 percent of what a document costs to carry. If that price is wrong the case does not collapse, it degrades, and it degrades slowly. The case only fails when inference costs 10.91 times the modeled price, at which point the estate costs exactly what it did before. That is the one sensitivity worth arguing about, so it is published rather than buried.
The before is held constant across every row. more
The spend ledger, quarter by quarter
Latest complete quarter 2026-Q2 · 2026-Q3 is open and is not used
The weekly spend figure above divides the latest complete quarter by thirteen. The most recent quarter on the ledger is still open and runs well below its predecessors for that reason alone. Quoting it would understate the ledger by about a quarter, so it is shown here and excluded from the comparison.
What this page settles and what it does not
Read before quoting anything above
- Which of the four costs to quote, and why the other three are not it.
- Whether the published weekly volume is the deep week or the year average. more
- How much of the volume gap is attributable to named towers, and how much is residual.
- Whether the saving is banked or forward, kept apart rather than added together.
- How wrong the inference price can be before the case fails.
- None of these four is measured at source. more
- The before is asserted. No system here observed the estate at $27,926,042 a week.
- The two towers outside tolerance need a data owner to say which of the two counts is right.
- The spend ledger and the tower rollup will not converge until shared platform and allocated management are coded to the towers.
Actions
What is waiting on a person
Each count below is a place where the estate publishes a number it cannot yet defend as a measurement. more
Decompose a sub-function that cannot be priced forward at all
These carry no decomposable marginal case, so no forward saving can be quoted for them either way. more
Move a cost view from read out of this platform to measured at source
All four costs are read out of this platform or modeled inside it. more
Retire a published cost that is not the one to quote
Three costs to serve remain in circulation beside the recommended one, spanning $12.6m a week. more
Settle a tower whose published volume does not match the ledger
These towers carry 22,020 transactions a week of the total gap on their own. more
Meter the inference price the whole case rests on
Inference is 21.0 percent of the unit cost and is modeled, not billed. more
Operations
What this desk is allowed to start
A surface that only reports is not operable. more
Trigger and bound
This page can set the four published costs to serve beside each other with the population and unit cost each implies, name one of them as the number to quote and record why the other three are not, reconcile published volume against the daily ledger tower by tower, keep banked saving apart from forward saving, and show how far the modeled inference price can move before the case fails. more
Live observability
What the record shows right now
Every tower by whether its published weekly volume reconciles to the daily ledger year average, an exhaustive partition of all 12 towers … more
Current distribution
12 towers
Is policy and strategy coming to fruition
Whether the written intent is holding here
One cost is now named: $9.1m a week, $471.9m annualized, $10.08 a document, with the other three still published and each carrying a written reason it is not the one.
Holding on the record
The position this page exists to test is that an estate which publishes four different costs to serve has not measured its cost to serve at all, and that naming one of them is worth more than showing all of them. more