Full narration of the briefing. Click a section heading to jump the player to that moment.
AWS will arrive at your negotiation with a model of your business built from your own telemetry. They know your run rate, your growth curve, your idle instances, and which teams launched what last quarter. You cannot stop them knowing. You can arrive knowing it first, and better.
That is the whole of this episode: the four numbers that decide whether the commitment you sign is safe.
The cost and usage report is the only ground truth. Not the console summary, not the account team's deck, not last year's budget. Turn on the CUR, land it somewhere you control, and keep it. Every claim you make later should trace to a line in it.
And keep it somewhere that survives the relationship. If you ever buy through a reseller, your billing history sits inside their organization, and a member account that leaves loses access to its own cost data while the reseller keeps it. Export continuously to storage you own, from day one. Two more things belong in the same export.
Tag discipline, because an untagged workload cannot be defended, moved, or excluded from a commitment later, and it is also the mechanism migration funding is paid against. And a twelve month history, not three, because AWS will model your growth from a full year and a short baseline flatters whichever quarter you happened to pick.
Measure coverage before you measure ambition. What share of your steady state already runs under a Reserved Instance or a Savings Plan, and how well used is it? Published data across roughly three billion dollars of AWS compute puts the median effective savings rate at about fifteen percent, the top decile above forty, and the bottom quarter at zero. Zero.
A quarter of AWS customers get nothing from commitments, and the worst cases run negative, which is the arithmetic signature of paying for capacity nobody consumed. Rate discipline is worth more than the discount you are about to negotiate, and it is entirely within your control.
Optimise before you commit, because after is theirs. This is the rule that costs buyers the most money and it is completely counterintuitive. Every idle instance you kill, every gp2 volume you move to gp3, every rightsizing your FinOps team lands after signature reduces the spend that retires a commitment you already agreed. So your own efficiency programme becomes your shortfall.
Clean the estate first, measure the honest floor, then commit against it. Do it the other way round and you have paid AWS for savings you generated yourself.
Then separate the commodity bill from the AI bill. They behave differently now. General compute, storage, database and networking are the mature eighty percent of most estates, and AWS still competes for them. The accelerator lines do not behave that way.
AWS has said publicly it cannot meet demand in twenty twenty six, expects the same in twenty seven, and already has capacity reserved into twenty eight. It raised Capacity Block prices twice in six months, cumulatively about thirty eight percent, with effectively no notice, because that pricing is framed as market driven rather than as a fee change. So model them apart. Negotiate hard where AWS still needs the revenue, and budget conservatively where it does not.
A single blended forecast hides both facts and produces a commitment that is wrong in two directions at once. And check what is already contractually spoken for. Savings Plans and Reserved Instances you bought last year still bill; a Savings Plan cannot be resold and only a narrow class of Reserved Instances can. Those commitments are part of your floor whether you like them or not, so count them before you add another layer of promise on top.
The output is one page you can defend. Run rate by service from the CUR. Coverage and utilisation. The workloads already contracted to move, and the ones that are still a hope.
The AI burn, metered from a real pilot rather than extrapolated from enthusiasm. And a single sentence at the bottom: the number you would sign today and sleep. That is your floor. Everything above it is upside AWS can fund with credits, ramp, or a better rate, but it is not something you promise.
One last point. At Redress Compliance we build this baseline and run the negotiation on pure contingency: twenty five percent of what we save you. If we save you nothing, you pay nothing. Next episode: what a good AWS deal actually looks like, and why the published benchmark tables disagree with each other by a factor of three.
Redress Compliance works on contingency: our fee is 25 percent of what we save you. Nothing saved, nothing paid. Independent, buyer side only, never vendor funded. Want Redress to contact you? Reach out and we respond the same day.
Talk to a AWS negotiator