A worksheet for making cloud and AI spend governable: what is attributable today, the allocation gaps, who receives the showback, and the unit economics to report.
Showback reports attributed cost back to the owner without moving any budget. Chargeback moves the budget too, so the owning team actually pays. Showback first is almost always the right order, because starting with chargeback before the numbers are trusted turns every month into a dispute about the allocation method rather than a conversation about the spend. Move to chargeback once recipients stop arguing with the figures.
Because a reduction target handed to someone who cannot see their own consumption is a request they cannot act on. The bill arrives as one number, so every conversation about it becomes an argument about blame rather than a decision about value. Attribution also tends to produce a real reduction on its own, before any engineering, because a team that can see what it is running turns off what it was not using.
The dimension that matches how decisions are actually made in your organisation: team, product, customer, or use case. If budget is held by product and you attribute by team, the reports go to people who cannot act on them. Pick the decision-making dimension rather than the one that is easiest to tag, because re-tagging an estate twelve months later costs more than getting the model right at the start.
Choose a method and state it: split evenly, split by consumption, or leave it in a central pool reported separately. All three are defensible and an unstated method is what makes recipients dispute the number. Then give the unattributable pool a named home, a target for shrinking it, and a date, or it becomes permanent and quietly absorbs anything inconvenient.
A unit of business outcome rather than a unit of consumption. An order, a claim, a report, an active customer, a processed document. Cost per gigabyte or per compute hour tells you how much you bought; cost per order tells you whether it was worth it. A platform cost that rises while units rise faster is working, and only the per-unit figure shows that.
Because pilots run at low volume with short context windows, and both of those change the per-unit economics. The same workflow at production volume with real document lengths and real retry behaviour is a materially different number. State the volume and context assumptions next to any AI unit cost, then state what the figure becomes at ten and a hundred times the volume.
Then the showback will produce awareness and no reduction, and it is better to know that before promising savings. Ask what an owner is empowered to resize, reschedule or decommission without a change board. If the answer is nothing, the first piece of work is the authority rather than the reporting, and that is a conversation with the platform owner rather than a tagging exercise.
Less than it looks. Tagging is the implementation of a decision about the attribution model, the shared-cost method, the unit of outcome, and who receives the number. Teams that start by tagging usually re-tag, because the model was never settled. Decide the four things in this worksheet, then tag once.
Cloud and AI spend is rarely out of control because the engineering is wasteful. It is out of control because the invoice has no owner, so every conversation about it is an argument about blame rather than a decision about value. This is the worksheet for fixing that first: what you can attribute today, what you cannot, who receives the number once you can, and what they are expected to do with it. Fill it in before anyone is asked to cut anything.
A worksheet for making cloud and AI spend governable: what is attributable today, the tagging and allocation gaps, who receives the showback, and the unit economics to report.
Cost attribution assigns each unit of cloud and AI spend to the team, product or use case that caused it, so the bill stops being one number nobody owns. Showback reports that attributed cost back to the owner without moving budget; chargeback moves the budget too. Attribution comes first, because a reduction target handed to someone who cannot see their own consumption is a request they cannot act on.
Start with the honest number, not the target. This is usually the first time anyone has written it down, and it is the figure the whole plan is measured against.
Cloud, data platform, AI and model APIs, SaaS and licences. Twelve months so seasonality is visible. One table.
Attributed means a person or team would recognise it as theirs, not that it carries any tag. Write the percentage. Most organisations are lower than they expect and the gap is the project.
Shared services, untagged legacy, cross-cutting networking and storage, and the AI spend that arrives under one organisation-wide key. Name the categories rather than leaving a residual.
If the honest answer is that it goes to one person who forwards it, that is the finding. A bill with no route to an owner cannot be governed by anybody.
Tagging is the implementation. The model is the decision, and getting it wrong means re-tagging everything twelve months later.
Team, product, customer, or use case. Pick the one that matches how decisions are actually made in your organisation, not the one that is easiest to tag. If budget is held by product, attributing by team produces reports nobody can act on.
Split evenly, split by consumption, or left in a central pool and reported separately. All three are defensible. An unstated method is what makes recipients dispute the number instead of acting on it.
Showback reports the cost to the owner. Chargeback moves the budget. Starting with chargeback before the numbers are trusted turns every month into a dispute about the allocation rather than a conversation about the spend.
A named home, a stated target for shrinking it, and a date. Without this, the unattributable pool becomes permanent and quietly absorbs anything inconvenient.
Cost per unit of consumption tells you how much you bought. Cost per unit of outcome tells you whether it was worth it. Only the second one survives a budget review.
An order, a claim, a report, an active customer, a processed document, a resolved ticket. One per product area. This is the denominator.
Attributed cost divided by units produced, over the same period. Write the method next to it so the figure can be recomputed by someone else next quarter.
A platform cost that rises while units rise faster is working. The absolute bill going up is not, on its own, a problem. Say which you are managing.
Pilot figures mislead badly here. Pilots run at low volume with short context windows, and per-unit cost at production volume with real context is a different number. State the assumption, then state what it becomes at 10x and 100x.
A report with no recipient and no expected action is a dashboard. This section is what converts attribution into governance.
A person, with their manager's knowledge. Not a team and not a cost centre code.
Monthly in an existing operational meeting beats a quarterly review nobody attends. Attach it to a meeting that already happens.
A percentage move, an absolute move, or a unit-cost move. Agree it now, while nobody knows which way their own line will go.
The uncomfortable question. If they cannot resize, schedule, or decommission anything without a change board, the showback produces awareness and no action, and you should know that before promising savings.
Reduction work gets credited to whoever is in the room when the bill falls. Write the baseline down first so the result can be checked rather than claimed.
Spend, attributed share, and cost per unit, for a stated period. Without this, no later claim can be verified and the second year's request will not be believed.
Often a real reduction before any engineering, because a team that can see its own consumption turns off what it was not using. State the expectation so it can be tested.
Rightsizing, scheduling, storage tiering, commitment coverage. Kept separate from the attribution effect so the two are not conflated in the reporting.
Ninety days after the showback goes live. Name the person. A measurement nobody owns does not happen, and this is the one that funds the next phase.
A cloud provider had let 14 business units each build their own data warehouse over eight years. Previous consolidation attempts had stalled on business unit resistance rather than on technology, which is an attribution and ownership problem wearing a technical costume.
14 business unit data warehouses, built independently over eight years
Each unit kept ownership of its own data and published through a shared catalogue, rather than being centralised
These separate a firm that will make the spend governable from one that will run a one-off reduction exercise and leave.
What share of our spend do you expect to be attributable at the end, and what will the remainder be made of?
Will you report cost per unit of outcome, or only cost per unit of consumption?
Are you proposing showback or chargeback, and what has to be true before chargeback is safe?
What reduction do you expect from attribution alone, before any engineering work?
Who owns the tagging standard after you leave, and what stops it decaying?
What will the baseline be, and will it be recorded before any optimisation starts?
Showback reports attributed cost back to the owner without moving any budget. Chargeback moves the budget too, so the owning team actually pays. Showback first is almost always the right order, because starting with chargeback before the numbers are trusted turns every month into a dispute about the allocation method rather than a conversation about the spend. Move to chargeback once recipients stop arguing with the figures.
Because a reduction target handed to someone who cannot see their own consumption is a request they cannot act on. The bill arrives as one number, so every conversation about it becomes an argument about blame rather than a decision about value. Attribution also tends to produce a real reduction on its own, before any engineering, because a team that can see what it is running turns off what it was not using.
The dimension that matches how decisions are actually made in your organisation: team, product, customer, or use case. If budget is held by product and you attribute by team, the reports go to people who cannot act on them. Pick the decision-making dimension rather than the one that is easiest to tag, because re-tagging an estate twelve months later costs more than getting the model right at the start.
Choose a method and state it: split evenly, split by consumption, or leave it in a central pool reported separately. All three are defensible and an unstated method is what makes recipients dispute the number. Then give the unattributable pool a named home, a target for shrinking it, and a date, or it becomes permanent and quietly absorbs anything inconvenient.
A unit of business outcome rather than a unit of consumption. An order, a claim, a report, an active customer, a processed document. Cost per gigabyte or per compute hour tells you how much you bought; cost per order tells you whether it was worth it. A platform cost that rises while units rise faster is working, and only the per-unit figure shows that.
Because pilots run at low volume with short context windows, and both of those change the per-unit economics. The same workflow at production volume with real document lengths and real retry behaviour is a materially different number. State the volume and context assumptions next to any AI unit cost, then state what the figure becomes at ten and a hundred times the volume.
Then the showback will produce awareness and no reduction, and it is better to know that before promising savings. Ask what an owner is empowered to resize, reschedule or decommission without a change board. If the answer is nothing, the first piece of work is the authority rather than the reporting, and that is a conversation with the platform owner rather than a tagging exercise.
Less than it looks. Tagging is the implementation of a decision about the attribution model, the shared-cost method, the unit of outcome, and who receives the number. Teams that start by tagging usually re-tag, because the model was never settled. Decide the four things in this worksheet, then tag once.