Cloud cost optimization with AI. Size the commitment first, then bank the savings beneath it.

Your commitment sets the discount on the whole cloud bill, so size it from actual consumption before you chase efficiency. Then let AI rank the savings beneath it in dollars, month by month.

How does AI help with cloud cost optimization?

AI turns cost data into a ranked list of decisions. It reads the daily cost exports, sizes each savings workstream in dollars, and prices the same workload on other clouds. That replaces an endless backlog of recommendations with a queue your owners can work through in order.

Most cloud teams already have the picture. They lack the step from picture to decision: which workstreams return money this quarter, what the next commitment should be, and what a workload would cost elsewhere. The order of those decisions matters more than the tool you use to make them.

What the AI layer reads

In 2026 every large cloud publishes detailed billing data in a common format, the FinOps Open Cost and Usage Specification (FOCUS). That makes a single model across providers practical.

• AWS. Cost and Usage Report data delivered through AWS Data Exports, including a FOCUS export.
• Azure. Scheduled exports from Azure Cost Management, which also offer a FOCUS dataset.
• Google Cloud. The Cloud Billing export to BigQuery, with a FOCUS view on top.
• OCI. Oracle's cost reports, which Oracle publishes in FOCUS format.

What it should produce

The native tools, such as AWS Compute Optimizer, Azure Advisor and Google Cloud Recommender, list resource level suggestions. They are useful, but they rarely show which suggestions add up to real money. Their commitment purchase recommendations also cannot see the savings you plan or the EDP, MACC or credit contract you already signed.

The output we want from an AI pass is short. It gives a dollar value per workstream with a named owner, a list of workloads eligible for scheduling drawn from tags and usage patterns, and a consumption trend line set against the commit trajectory.

Why should you size the cloud commitment before you rightsize?

Because the commitment sets the price of everything beneath it. Rightsizing saves the cost of one workload, while the commit sets the discount on the entire bill for years. That commit is an AWS EDP, an Azure MACC, Google Cloud committed use discounts or OCI Universal Credits.

A commit sized 20 percent too high turns every efficiency win beneath it into unused commitment at true up. Your engineers cut spend, consumption falls, and the saving reappears as a shortfall charge. So size and negotiate the commit first, then bank the workstream savings underneath it.

Treat the commit as a flagship contract

A cloud commit is a major contract that happens to be priced in consumption, and it deserves the same discipline as your largest software renewal. Discount tiers, term length, carry forward of unused commitment, product eligibility and renegotiation triggers are all open to negotiation. Each can be benchmarked against comparable deals.

The vehicle level detail sits in our AWS EDP negotiation guide, the Azure MACC negotiation guide and the Google Cloud CUD optimization guide.

How should you size a cloud commitment?

Start from trailing twelve month consumption and work forward with three scenarios. Minimum is the floor you are confident of, medium is the trend, and growth is the plan. Price the cost of unused commitment and the cost of overage in each, then choose knowing what each one risks.

The vendor's sizing model works the other way. It starts from your growth story and builds up. In our review file, four out of five commits were sized above trailing consumption growth, and the forecast behind the sizing almost always traced back to a model the vendor built.

A worked example with three scenarios

Say your trailing twelve months of net cloud consumption come to $10 million. The trend points to 5 percent growth, the business plan to 20 percent. Engineering also plans an optimization program worth 6.5 percent of spend, about $650,000 a year. Assume each step up in commit buys one extra discount point, worth roughly $100,000.

The growth commit is 20 percent above trailing consumption. If only the trend arrives, it leaves $2.15 million unused to win $200,000 of discount, a net loss of $1.95 million. Even when the plan arrives in full, the savings program leaves $650,000 of the commit unspent.

The medium commit shows the mechanism most plainly. Its entire unused balance equals the savings your engineers delivered. The minimum commit gives up at most about $200,000 of discount and carries no shortfall in either outcome.

Why a coverage target is the wrong scorecard

The standard advice measures FinOps maturity by coverage, the share of spend under commitment, and pushes it toward 100 percent. We disagree, because coverage rewards exactly the overcommitment that costs buyers money. Every point bought above the consumption floor is a prepaid bet that a growth forecast comes true.

In our file the median customer carried 12 to 18 percent unused or misapplied commitment at true up, while reporting excellent coverage to the board. Put unused commit at true up on the dashboard instead and drive it toward zero. Buy the floor, negotiate flexibility on the rest, and let coverage land wherever honest scenarios put it.

What the account team will say, and what to say back

• "Your growth plan supports the next tier, and it comes with two more points." Show the scenario table. Ask for the higher tier discount on a floor level commit with an annual ramp, or for the tier to apply once actual consumption crosses it.
• "Customers always grow into their commitment." Then carry forward of any shortfall into the next term costs the vendor nothing. Ask for it in writing.
• "Your coverage is low compared with similar customers." Coverage tells you how much you prepaid. Reply that you manage to unused commit at true up, because that is the figure your CFO pays for.
• "The best discount needs a longer term." Ask for the price of the shorter term as well, then ask for a reduction right at each anniversary on the longer one.

Which cloud cost workstreams return money in year one?

Four workstreams fund themselves inside the quarter, and their ranking rarely changes. Typical first year yield as a share of cloud spend runs roughly 4 percent from commit resizing, 3 percent from right sizing, 2 percent from scheduling and 1.5 percent from storage. The size varies by customer; the order seldom does.

The three operational workstreams sit below the commit. Cross cloud pricing sits beside them as a negotiating tool for the commit itself.

• Database and compute right sizing, roughly 3 percent. Instances provisioned for a launch day three years ago, databases on performance tiers their query load never touches, and dev environments running at production scale. Per instance evidence makes each change hard for the approving owner to dispute.
• Compute scheduling, roughly 2 percent. Nonproduction environments run nights and weekends for no one. Scheduling recovers 60 to 70 percent of run time on eligible workloads, and AI extracts the eligibility list from tagging and usage patterns.
• Storage cleanup and tiering, roughly 1.5 percent. Orphaned volumes, forgotten snapshots and hot tier data no one read this year. This workstream meets the least resistance, because no one defends a snapshot from 2023.
• Cross cloud workload pricing. The same workload priced on AWS, Azure, Google Cloud and OCI, with egress and licensing effects included. Occasionally it justifies a move. More often it gives you bargaining power at the commit table.

The three operational workstreams together come to roughly 6.5 percent of spend. On the $10 million example above, that is about $300,000 from right sizing, $200,000 from scheduling and $150,000 from storage, plus about $400,000 from resizing the commit.

Fix tagging first

Workstream sizing is only as good as the ownership tags on the resources. A two week tagging sprint often achieves more than a quarter of analysis, because untagged spend has no owner to approve a change. Our cloud tagging strategy guide covers the tag set we ask for.

The commit side of the same picture is covered in the AWS Savings Plans guide and the OCI cost optimization guide.

Tools that do the scheduling and tiering

You rarely need new software to execute. Instance Scheduler on AWS, Start/Stop VMs v2 on Azure and instance schedules on Compute Engine all stop and start machines on a timetable. S3 Intelligent-Tiering shifts objects between access tiers on its own, and lifecycle rules on each cloud archive or delete data once owners agree on retention periods.

Price the same workload on the other clouds

Price your largest workloads on at least one competing cloud before the renewal opens. Buyers who brought a credible comparison to a commit renewal negotiated measurably better discount tiers than buyers who brought last year's contract.

The comparison must be honest about egress. One that ignores data transfer and licensing effects will be dismantled by the vendor in a single email. Our egress negotiation guide shows what to include.

What contract terms should a cloud commitment include?

Ask for terms that let the commit shrink or roll over when consumption falls short. The discount rate gets the attention, but carry forward and reduction rights decide what a bad forecast costs you.

• Carry forward of unused commitment. A shortfall in one year rolls into the next year's commitment and is not invoiced.
• True down rights. A right to reduce the commit at an anniversary if consumption trends below the floor.
• An annual ramp. Lower early years and higher later ones, so the commit rises as migrations go live.
• Broad product eligibility. Which services and Marketplace purchases retire the commit. Microsoft, for example, counts eligible Microsoft Marketplace offers toward a MACC.
• Shortfall treatment. On a MACC, any shortfall at the end date is charged as an Azure prepayment credit, and consumption paid from that credit does not count toward the MACC. Know the equivalent on every vehicle you sign.
• Renegotiation triggers. An acquisition, divestiture or cancelled migration should reopen the commit.
• Credit term and rollover on OCI. Universal Credits that expire unused are where OCI buyers overpay, so negotiate the term, the rollover and the rate card together.

More detail on each clause sits in our guides to AWS EDP flexibility provisions and MACC sizing.

AI platform commitments need shorter terms

Model usage agreements with the AI platforms now reach commit scale and behave the same way. They are consumption based, justified by a forecast, and oversized by default, usually on numbers taken from a pilot.

One difference changes the negotiation. Model efficiency improves fast, and a commitment sized on today's model usage ages quickly, so a shorter term and stronger flexibility clauses are worth more than an extra discount point. Ask for model substitution rights and true down rights. Our cloud AI commitment guide covers the clauses.

How do you run cloud cost optimization as a monthly rhythm?

Run it monthly from fresh exports, never as an annual review. Findings expire because cloud environments change daily, so a rightsizing list from last quarter describes infrastructure that no longer exists.

Week two is where analysis becomes a decision with a name on it.

Week four matters because FinOps sees consumption and procurement sees the contract, and only the joint session puts both on one page. The habit matters more than the tooling, since an analysis worth a percentage of the bill in week one is worth little by the following quarter.

How to check your commitment position today

• AWS. The Savings Plans and Reserved Instance reports in Cost Explorer, which show coverage and how much of each commitment you use, plus your EDP addendum for the annual commit figure.
• Azure. In Cost Management + Billing, the MACC view under Credits + Commitments (Enterprise Agreement) or Benefits (Microsoft Customer Agreement).
• Google Cloud. The committed use discount analysis reports in Cloud Billing, which show how much of each commitment you actually use.
• OCI. Cost Analysis and your subscription details, to see credits consumed against credits expiring.

Common mistakes and what they cost

• Rightsizing before resizing the commit. The savings land as unused commitment, so the bill does not move.
• Signing on the vendor's forecast. It starts from your growth story and works up, so you prepay for growth that has not happened yet.
• Buying Google Cloud CUDs project by project. Commitments overlap and some sit idle while others run out.
• Comparing clouds without egress. The vendor dismisses the whole comparison, and your bargaining power goes with it.
• Letting Savings Plans or reservations lapse unnoticed. Usage falls back to on demand rates the day they expire, and the jump shows up a month later on the invoice.

The wider cost program sits with our cost optimization practice, and the vehicle by vehicle detail with the AWS practice and the Google Cloud practice.

What have we seen in cloud commitment reviews in 2024 and 2025?

The commit decided the outcome more than any efficiency program beneath it. That held across the 20 to 25 cloud commitment reviews we ran in 2024 and 2025, and three patterns recurred.

• Oversized by default. Four in five commits were sized above trailing consumption growth, usually on a growth story rather than on what the customer had actually consumed.
• Comparison beat history. Buyers who arrived at renewal with a cross cloud workload comparison won better discount tiers than those who arrived with last year's contract.
• Workstreams paid for themselves. Database right sizing, storage cleanup and compute scheduling funded themselves inside the quarter in most environments we reviewed.

The customers in the best shape ran optimization monthly from fresh exports. The ones that reviewed annually described, accurately, an environment that had already changed.

Size the commit to consumption instead of ambition, negotiate it like your largest contract, then bank the savings beneath it and keep the loop running.

What to do next

1. Export six months of daily costs. Pull the trailing six months from every cloud account, map commit trajectories against the consumption trend, and put the date of each next true up in the calendar.
2. Build three scenarios for the next commit. Price minimum, medium and growth with the cost of unused commitment and the cost of overage in each, and choose knowing what an oversized commit would cost.
3. Change the dashboard. Put unused commit at true up on it in place of coverage, and drive that number toward zero.
4. Bank the workstreams this quarter. Run a two week tagging sprint, then take storage, scheduling and right sizing in that order of ease.
5. Price your top workloads on the other clouds. Include egress and licensing, and do it before the commit conversation starts, because the comparison helps even if you never move.
6. Get an outside view on sizing. Our cost optimization team runs the sizing and the negotiation with you.

Frequently asked questions