Skip to content
Board advisory and board seats: full for 2H FY26 and 1H FY27. 1 of 3 positions open from 2H FY27 onward.
See details →
Writing

SaaS Sprawl Is a Missing Control Plane.

Cards bought two hundred tools, half of them duplicates, 40 percent of seats dark. The missing control plane: discovery, true-down, shadow-SaaS security.

By Michael YorkNovember 12, 2026 8 min read 1,760 words All postsTable of contents

Someone in finance runs a spend report, sorts by vendor, and the number that comes back is not a line item. It is a census. Two hundred SaaS subscriptions. A third to a half of them overlapping in function: three tools that take notes in meetings, two that make diagrams, four that store passwords. And close to 40 percent of the paid seats belong to people who have not signed in this quarter. The reflex, every time, is to reach for someone to blame, and procurement is standing closest.

That reflex is wrong, and it sends the fix in the wrong direction. Procurement did the job it was given. Each of those tools was bought by someone who needed it, approved by a manager who agreed, and paid for on a card that cleared. No policy was violated. The system worked exactly as designed, and the design is the problem. What the company is missing is not discipline at the point of purchase. It is a control plane.

I borrow the term on purpose. In cloud infrastructure, the control plane is the layer through which every resource gets created, entitled, observed, and destroyed: the place you go to answer what we have, who can touch it, what it is costing, and how to turn it off. SaaS never got one. It was bought a card swipe at a time, in an era where a single-sign-on button and a corporate card democratized purchasing to anyone with a problem and a budget code. Discovery, entitlement, metering, deprovisioning (the four things a control plane does) are each somebody else's spreadsheet, if they exist at all.

Where I sit shapes what I notice. I don't own my company's whole SaaS budget, and I won't pretend the whole thing is my P&L. I own a slice: the security tooling, the platform, the identity layer. But the half of this problem that actually bites (the shadow subscriptions, the third-party integrations, the seats that outlive the employees) is a security problem before it is a finance problem, and that half is squarely mine.

When my team built AgentOS, our governed internal agent platform, the agent was never the hard part. The plane was. The real engineering was scoped tool authority: every tool gets a narrow credential, and every action lands in the audit trail. SaaS is the same shape of problem, bought on a corporate card. So here are the three planes it needs, in the order they pay back.

Discovery is a reconciliation, not a survey

The instinct is to send a form around and ask every team to list their tools. You will get back a fraction of the truth, cheerfully. The apps people forget to mention are precisely the ones you most need to find, because "forgot about it" and "bought it on a personal card and expensed it" look identical on the org chart.

Discovery is a join across the four systems that already know the answer, run continuously instead of once a year. That join is the control plane's read path.

  • The identity provider. Every app wired to SSO shows up in your SAML and OIDC logs the moment someone signs in. This is your cleanest signal and your smallest set: the governed apps, the ones already doing it right.
  • The card and expense feed. Every subscription that never touched single sign-on is hiding here, categorized as "software" or, worse, "office supplies." Reconcile merchant names against the identity-provider list, and the gap between the two is your shadow estate.
  • The OAuth grant inventory. Pull the third-party app authorizations from your Google Workspace or Microsoft 365 tenant. Every "sign in with Google" and every "connect your calendar" is a live grant, and most of them point at tools nobody is paying for and nobody remembers approving.
  • The accounts-payable ledger. The annual invoices that never hit a card, the enterprise deals with a real contract and a real signature, live in AP and nowhere else.

Run those four together, dedupe, and you have the only artifact that matters: a real inventory. Not a static one. The value is in re-running the join, because the estate grows every week whether or not you are watching. A SaaS-management platform or a CASB will automate most of this; four exports and an afternoon get you 80 percent of the way. Either way, every downstream discipline, the cost work and the security work alike, is guessing until this exists.

License true-down, not renewal theater

Once you can see the estate, the money is in plain view, and not where most people go looking for it. The renewal is the wrong venue and the wrong verb. Renewal negotiation is an annual event where you argue about the unit price. True-down is a continuous practice where you fix the unit count, and the unit count is where the waste lives.

The mechanic is the FinOps discipline cloud teams already run, pointed at seats instead of instances. Meter usage: last sign-in per seat, per app, refreshed monthly. Reclaim anything dark for ninety days, not by asking, but by defaulting to removal and letting the rare real user raise a hand. De-duplicate the overlap: three note-takers and two diagramming tools is not resilience, it is a standard nobody set. Pick one, migrate, cancel the other two. Right-size the tiers, because half the "enterprise" plans in the estate were bought for one feature a lower tier now includes. The benchmarks SaaS-management vendors publish land in the same range year after year, unused seats near a third and redundant tools in the double digits, so none of this is a number you have to take on faith.

Do the true-down before the renewal, not during it. Walking into a renewal already right-sized changes the negotiation entirely: you are buying what you use, from a position of having proven you will cut what you don't. Walk in with 40 percent dead seats and you are negotiating a discount on waste, while the vendor knows your renewal date better than you do. The seat count is the lever. The price is the distraction.

Shadow SaaS risk lives in identity, not spend

Here is where the finance framing quietly fails, and where I get most interested. The risk in the shadow estate is almost never the twelve-dollar-a-month subscription. Nobody's data leaves the building because a designer expensed a font tool. The risk is the OAuth grant that came bundled with it.

When an employee clicks "connect your Google account" to sign up for some workflow app, they are not just creating a login. They are minting a non-human identity: a token that can read their mail, their files, their calendar, on their behalf, from a third party you never assessed, indefinitely, until someone revokes it. That token does not appear on a spend report. It appears in your data-egress path. The Cloud Security Alliance has put non-human identities at roughly 45 to 1 against humans, and this grant sprawl is a large part of why. Each of those grants is a standing door into your tenant that a vendor breach turns into your breach.

This is why the discovery join has to include the OAuth inventory rather than only the invoice list, and why SaaS management should not sit purely in finance. We serve more than 1,500 financial institutions whose examiners become our examiners by proxy, and a third party holding a live token into a mailbox is a vendor-risk finding whether or not a dollar ever changed hands for it. So score the grants by scope, not by price. A free tool with read/write access to your entire Drive outranks a five-figure subscription that never leaves its own sandbox. A grant nobody can name is a grant you close.

The plane is defined by its off-ramp

The test of whether SaaS is actually on a control plane is not how it gets in. It is how it gets out. A subscription you cannot turn off in one action is not governed, no matter how clean the onboarding looked: a seat that survives the employee, a tool with no owner, a login that HR's offboarding never reached.

The primitive for this already exists and is boring, which is how you know it is load-bearing. SCIM, the provisioning standard that lets your identity provider create and, more importantly, destroy accounts in a downstream app automatically, is the write path of the SaaS control plane. An app behind SSO with SCIM deprovisioning switched on turns an employee's departure into a single event that ripples out and shuts every door. An app bought on a card, with a local password, does not. It keeps its door open, and its seat on the invoice, for years, and it is exactly the orphaned account an exit-diligence team or an examiner finds in an afternoon.

So make enrollment the gate. The price of admission for any tool that touches company data is that it rides SSO for entry and SCIM for exit. Tools that can do neither get sandboxed, or replaced by one that can. That single rule shrinks the estate more than any spend review, because it makes the ungovernable tools visibly ungovernable, and it turns offboarding from a checklist somebody forgets into a property of the system.

Build the plane, not another spreadsheet

SaaS sprawl will not be fixed by a stricter buyer or an annual cleanup. It gets fixed by building the plane it should have passed through all along. Four moves, in order.

  • Reconcile, don't survey. Join your identity provider, card feed, OAuth grants, and AP ledger into one living inventory, and re-run it.
  • True down before you renew. Meter seats, reclaim the dark ones by default, collapse the duplicates to one standard.
  • Score the grants, not the invoices. Treat every OAuth token as a non-human identity and a data-egress path, and revoke on scope.
  • Gate on the off-ramp. SSO to get in, SCIM to get out, or you don't hold the data.

Do those and the two-hundred-tool census stops being an embarrassment you rediscover once a year and becomes a number you watch, the way you watch cloud spend or open vulnerabilities. Same discipline, new surface.

The practical fight is where you draw the enrollment line: how hard a rule you make SSO-and-SCIM before a tool is allowed near real data, and what you do with the long tail of useful little apps that will never support either. Exceptions are manageable right up until the exception becomes the estate.

SaaS ManagementLicense OptimizationShadow ITProcurement