First year recoverable spend ran 5 to 12 percent with a median near 8, and it split in thirds rather than sitting in the one place every cost program looks first
Shelfware is the famous number and roughly a third of the answer. Commitment resizing and invoice recovery carried the rest, and only one of the three pays out inside sixty days.
Prepared by Redress Compliance · August 18, 2026 · Cost optimization advisory. 30 to 35 cost reviews led, 2024 to 2025.
Executive summary
First year recoverable spend ran 5 to 12 percent of the software and cloud budget, with a median near 8. That is the number a business case should carry, and it is well below what most cost programs are sold on.
The recovery split about evenly three ways. Roughly a third from license right sizing, a third from commitment resizing, and a third from renewals and invoice recovery, across the reviews led.
Shelfware averaged 21 percent of entitled seats before any reclamation. It is the most visible third and the one every program starts with, which is why the other two thirds stay unworked.
Cloud commitments were sized to vendor growth forecasts rather than consumption in four out of five estates. That is the third with the largest single ticket and the longest wait, because it pays out at the commitment renewal.
Where does the recoverable money actually sit?
In three roughly equal thirds, not in one pile. License right sizing, commitment resizing, and renewal and invoice recovery each carried about a third of the first year total across the reviews led.
That split is the useful finding, because the three thirds pay out on completely different clocks. A program sequenced by size will start in the wrong place.
| Workstream | What it needs | Typical year one yield | Speed to cash |
|---|---|---|---|
| Shelfware reclamation | Identity logs, SAM exports, entitlements | 1 to 3 percent of software spend | 30 to 60 days |
| Tier right sizing | Per user feature usage and plan pricing | 1 to 3 percent | 60 to 90 days |
| Commitment resizing | Daily cost exports and consumption history | 2 to 4 percent of cloud and AI spend | At the commitment renewal |
| Invoice and uplift recovery | Invoices, contracts, price lists | 0.5 to 2 percent, recurring | Immediate and monthly |
The size of the shelfware third is worked separately in our shelfware report, and the tooling that produces the evidence is compared in our SAM tools guide.
Why the sequence matters more than the total
Invoice recovery pays immediately and repeats monthly. Commitment resizing pays once, later, and largest. A program funded on quarterly results has to start with the first and hold its nerve for the second.
Why do cost programs stall before the money arrives?
They stall on labor, not on ideas. Everyone knows shelfware exists. Finding it, proving it, and chasing it across a thousand line items is weeks of work that nobody has spare.
Three gates account for most of the failures. Data assembly takes a quarter, and analysis lands at a grain too coarse to act on.
The third gate is timing. Recommendations arrive after the renewal that mattered has already closed, which turns the program into an annual record of spend rather than a way to reduce it.
A program that produces a deck, a steering committee and no money has not failed analytically. It has run out of the one input nobody budgeted, which is attention. Automation is worth having because it is the only thing that survives the third month.
What the tooling changes and what it does not
It removes the labor, not the judgment. Deciding who loses an E5 seat is still a conversation with an owner, and it still needs somebody willing to have it.
The enterprise cost optimization brief
The four workstreams, the data each one needs, the yields measured against them, and the sequence that banks the first result inside a quarter.
Get the brief →What 30 to 35 cost reviews showed
In the 30 to 35 cost reviews Morten Andersen led in 2024 and 2025, first year recoverable spend ranged from 5 to 12 percent of the software and cloud budget, with the median near 8. The split into thirds was consistent enough to plan against.
Shelfware averaged 21 percent of entitled seats on the major platforms before any reclamation had been attempted. In the reviews, 15 to 30 percent of seats showed no meaningful activity in 90 days.
Cloud commitments were sized to vendor growth forecasts rather than to consumption in four estates out of five. The forecast was usually written with help from the party selling the commitment, which is the part nobody says out loud.
Tier mix was the quietest third. Microsoft publishes the arithmetic on its plan pricing pages, so every seat carrying an edition it does not need is a public number, not an estimate.
The programs that survived into a second year all automated the monitoring. The manual ones decayed as attention moved elsewhere, which is the failure mode that returns the savings to the vendor a renewal later.
- Your quote benchmarked against 500,000+ real closed deals, adjusted for size, region, and industry
- Commitments resized against trailing consumption rather than the growth forecast
- Uplift caps and invoice variances flagged with the exact quote, the page, and the replacement text
Why are cloud and AI commitments the largest third?
Because they are sized once, for years, against a number the vendor helped write. Committed spend agreements, AWS enterprise discount programs, Azure commitments, Google Cloud committed use discounts, and the newer AI platform commits all share that shape.
The fix is to resize against trailing consumption with a minimum, a medium and a growth scenario, then price the unused commitment against the overage risk it is supposed to protect against.
The discipline that keeps it banked
Uplift caps, a benchmark at 120 days out, and a continuous invoice watch. The practices the FinOps Foundation documents work here, provided somebody owns the monthly check rather than the annual one.
What does a 90 day sequence actually look like?
Baseline in the first 30 days, bank the quick wins by 60, and have the structural fixes queued by 90. The sequence follows data availability rather than the org chart.
- Days 0 to 30, baseline. Pull identity logs, entitlements, cost exports, invoices and contracts into one view, which is the step that usually takes a quarter and does not need to.
- Days 30 to 60, quick wins. Reclaim inactive seats and recover invoice variances, both of which pay without a vendor conversation.
- Days 60 to 90, queue the structure. Tier reassignment and commitment resizing need owners and a renewal date, so they are staged rather than closed.
The structural third lands at a renewal date, which is why it belongs inside a standing renewal program rather than inside a one off project.
Never fund it from headcount
Fund the program from the first corrected renewal. A program funded from headcount competes with the teams whose cooperation it needs, which is the fastest way to lose both.
What the reviews measured, 2024 to 2025
Two cuts of the engagement file set the expectation a business case should carry.
Against a range of 5 to 12 percent of the software and cloud budget, across the 30 to 35 reviews led.
Averaged across the major platforms before any reclamation had been attempted, which is the most visible third of the total.
Eight percent is a smaller headline than most cost programs are sold on. It is also a number that survives contact with a finance review, which is the only kind of number worth putting in a business case.
Your first five moves
- Set the expectation at 5 to 12 percent with a median near 8, so the program is judged against what the engagement file actually shows.
- Start with invoice and uplift recovery, because it pays immediately, repeats monthly, and needs no vendor conversation to begin.
- Pull 90 day activity per seat before any negotiation, since 15 to 30 percent of seats on the major platforms showed none at all.
- Resize every cloud and AI commitment against trailing consumption, not against the growth forecast the vendor helped write.
- Automate the monitoring before attention moves on. The cost optimization practice runs the baseline first and the negotiation second.
Frequently asked questions
How much spend is actually recoverable in year one?
Between 5 and 12 percent of the software and cloud budget, with a median near 8, across the 30 to 35 cost reviews led in 2024 and 2025.
Where does the recoverable money sit?
In three roughly equal thirds: license right sizing, commitment resizing, and renewal and invoice recovery. Sequencing by size sends a program to the slowest third first.
How much shelfware does a typical estate carry?
Shelfware averaged 21 percent of entitled seats on the major platforms before any reclamation. In the reviews, 15 to 30 percent of seats showed no meaningful activity in 90 days.
Which workstream pays out fastest?
Invoice and uplift recovery. It recovers at full value, repeats every month, leaves a paper trail, and needs no vendor conversation to start.
Which workstream is worth the most?
Commitment resizing, at 2 to 4 percent of cloud and AI spend, but it pays at the commitment renewal rather than inside the quarter.
Why do cost programs stall?
On labor rather than ideas. Data assembly takes a quarter, analysis lands too coarse to act on, and recommendations arrive after the renewal that mattered has closed.
How are cloud commitments usually mis sized?
They are sized to a vendor growth forecast rather than to consumption, which was the case in four estates out of five. The correction is to resize against trailing usage with three scenarios.
How long does a first pass take?
About 90 days. Baseline in 30, quick wins banked by 60, and the structural fixes queued with owners and renewal dates by 90.
How do you stop savings leaking back?
Uplift caps, a benchmark at 120 days before renewal, and a continuous invoice watch. Programs that automated the monitoring survived into a second year and manual ones decayed.
How should the program be funded?
From the first corrected renewal, never from headcount. A program funded from headcount competes with the teams whose cooperation it needs to work at all.