Ask a delivery lead what their organization spends on Azure DevOps licenses, and you will usually get a number. Ask how that number was reached, and you will usually get an estimate: headcount times a seat price, minus a rough guess for the people who probably do not need a paid seat.
The estimate is usually wrong, and usually wrong in the same direction. Not because anyone is careless, but because an estimate cannot reconcile individual entitlements, actual activity, and what the role of each person requires. So you pay for more than you think.
This is not a story about a vendor overcharging. Azure DevOps bills exactly what it is told to bill. The gap opens up because the thing being billed (a paid seat attached to an identity and charged every month whether or not anyone uses it) drifts out of alignment with who is actually doing the work. Nobody decides to waste the money. It accumulates quietly, one stale assignment at a time.
Here is what that drift actually looks like when you go seat by seat.
1. Inactive Basic seats
The most common pattern is the simplest: a paid Basic seat held by a current, valid user who has simply gone quiet, with no meaningful activity in months. The access level was right for the role they once had; the role ended, changed, or paused, and the seat did not.
An estimate cannot see this, because an estimate works from the org chart (who should have a seat), not from activity. A per-seat audit works from the other end: who has a paid seat and has generated no signal for a defined window. Those two lists are never the same, and the difference is money leaving every month.
2. Over-provisioned Test Plans
Basic + Test Plans is, at Microsoft list pricing, the most expensive of the standard access levels, and the one most likely to be handed out defensively. A team adopts Test Plans, someone provisions the whole team to be safe, and then two or three people actually author or execute test cases. The rest carry the premium tier and use only the Basic half of it.
This is the highest-value line to inspect, because the per-seat premium is the largest: a handful of over-provisioned Test Plans seats can cost more than a dozen misassigned Basic seats.
3. Orphaned and departed identities
People leave. Contractors roll off. An acquisition brings in a directory that never gets fully reconciled. Each of those events can leave an identity that still holds a paid seat but no longer maps to a working person, sometimes a duplicate of someone who now signs in under a different account.
These seats are invisible to headcount math because they are not headcount. They are residue. They keep billing until someone reconciles the license roster against reality, which is exactly the kind of tedious, per-identity work that never reaches the top of any list.
4. The wrong access level
Not all waste is a seat that should be removed. A large share is a seat that should be downgraded. Azure DevOps offers a Stakeholder access level that is free and sufficient for people who need to view work, file items, and follow progress but do not write code or run pipelines, such as executives, business analysts and occasional reviewers.
When those users sit on a paid Basic seat, the fix is not deprovisioning; it is right-sizing. This is the pattern estimates are worst at, because it does not reduce the number of users, only the number of paid users. You have to look at what each person actually does to know a free tier would cover them.
5. Zero-footprint seats
Finally, the seats with no footprint at all: assigned, paid, and attached to an identity that appears in no project, no repository, and no pipeline. Sometimes it is a provisioning script that ran too broadly. Sometimes it is a project that was archived while its members kept their seats. Either way, the seat produces nothing and costs the same as one that does.
Patterns one, three, and five look alike, but the distinction is exactly what makes each one fixable: an inactive seat is a current user who has gone quiet, a departed seat is an identity whose person is gone, and a zero-footprint seat is one that never had a project or repository footprint to begin with. Inactivity calls for review, a departed identity for removal, a zero-footprint seat for reconciliation.
Put the five side by side, and the pattern underneath them is the same: an estimate sees an aggregate, and an audit names the seat:
| Waste pattern | What an estimate misses | What an audit names | |
|---|---|---|---|
| Inactive Basic | the total inactive percentage | the specific users, and how long each has been inactive | |
| Over-provisioned Test Plans | the average license mix | the users whose activity does not justify the premium tier | |
| Departed identities | headcount assumptions | accounts still consuming a paid seat after the person left | |
| Wrong access level | aggregate seat counts | the exact Basic seats a free Stakeholder seat would cover | |
| Zero-footprint seats | the total licensed-user count | paid users with no project, repository, or pipeline footprint |
Why the math compounds
None of these is dramatic on its own. A single misassigned seat is a rounding error. The reason license waste is worth naming is that it is recurring: every wasted seat is charged every month, so a small monthly figure is really that figure times twelve, standing, until someone acts on it. A quiet 15% of a mid-size license bill left uncorrected for a year is a real number, and it is a number you are paying for capacity nobody is using.
Put rough figures on it. At Microsoft list pricing around the time of writing, Basic at roughly $6 per user per month, Basic + Test Plans at roughly $52, twenty unnecessary Basic seats run about $1,440 a year, and downgrading five over-provisioned Test Plans seats to Basic saves roughly $2,760. A small, unglamorous reconciliation problem is already north of $4,000 a year. Treat those numbers as illustrative: Azure DevOps pricing changes, so check current rates, and a real audit computes the figure against your own organization, not a benchmark.
The compounding is also why estimates fail as a management tool. An estimate gives you a total. A total tells you roughly how much you spend; it does not tell you which seat to change, so it cannot be acted on. You cannot deprovision an average.
Name it, do not estimate it
The alternative to estimating is auditing: go identity by identity, join the access level of each seat to its actual activity, and produce a list where every paid seat is either justified or named as waste, with the specific action attached: remove, downgrade or reconcile. That output is deterministic. Point the same instrument at the same organization twice and you get the same list, because it is reading facts, not forming an opinion.
That is what License Utilization, part of Helix Delivery Intelligence, is designed for. It reads your Azure DevOps organization entitlements and activity and reports paid seats by access level and last access, so the list underneath the number can be acted on. It is calculated by fixed rules, not by an AI model. How Helix handles credentials and data is set out in the Privacy Policy.
Most organizations already have a number for their Azure DevOps license spend. Very few have the list underneath it. The list is where the money is.
STRATAGEM LABS

