Can you replace a SaaS subscription with an app your own people build using AI? For a narrow set of apps, yes. The number that decides it is what the app costs to run after it’s built: sign-in, permissions, a record of what happened, and someone who answers when it breaks. Most rebuild cases skip that number, and it lands on IT.
It arrives as a hosting request
Someone in HR or finance has built a working version of a tool you pay for. They did it with an AI coding assistant, in a few days, and it does the three things their team actually uses. What they want from you is somewhere to put it.
Where does it live. Who operates it. What happens when the builder changes team. You’ve asked all of this before. What’s new is the volume, and the fact that the spend now has an owner.
of FinOps teams had AI spend in scope in 2026, up from 63% in 2025 and 31% in 2024
State of FinOps 2026, FinOps Foundation. Measures what sits in a FinOps team's scope, not whether anyone can attribute the spend.
of software spend Starbucks is building its own AI apps to cut into
Bloomberg, July 2026
Big companies are already doing this, so the question reaches you whether or not you went looking for it. Get the sum right before anyone cancels a contract.
Your current bill is bigger than the subscription
Most rebuild cases start with the subscription line from the contract and call that the saving. The number is real and it’s too small. Finance will find the rest of it, so find it first.
Three things sit on the side you are replacing:
- The subscription, including the increase you’ll be offered at renewal
- Admin time. Plenty of departmental tools need a part-time or full-time person to run cycles, fix integrations and manage access
- Implementation and services fees, which never show up in the renewal conversation but were in the original one
Our calculator counts none of the last two, which is why the saving it shows is the conservative version.
Owning the app means staffing it
Here’s the half that gets left out, and it’s why these cases get rejected. Someone has to keep the app alive after launch, and that person costs money every year you run it.
| What you stop paying | What you start paying |
|---|---|
| The subscription | The people who build it, once |
| Admin time on that tool | The people who keep it working, every year |
| Implementation and services fees | Somewhere to run it |
| Compute, storage and any AI calls it makes |
The swing factor is how many people, not the price of the platform. Change the builder assumption and the whole case moves. Change the hosting line and it barely registers.
Our own model puts it at about three weeks of one builder to rebuild an app, then about one full-time builder for every seven apps you run after that.1 Use those to start the argument, then replace them. Rebuild one app, measure what it actually took, and run the next five on your own number.
Count the people on both sides. A case that staffs the build and forgets the upkeep comes straight back from finance.
Somebody has to run it
A subscription includes things nobody itemises. The vendor signs people in, enforces who sees what, keeps a record, keeps the service up, and answers the phone. You’re paying for an operated service, and that part doesn’t arrive with the code your team writes.
An app on a builder’s laptop has none of it. Neither does one on a server somebody spun up for the weekend. That’s where rebuilds stall.
For every 20 AI builds, about 1 becomes a governed, running production app that delivers real value
Tray analysis, MIT NANDA + S&P Global
of organizations found at least one AI agent built without their security, IT or governance team knowing, in the past twelve months. 68% of the same group believed their visibility was strong
Cloud Security Alliance, Autonomous but Not Controlled, April 2026, n=418. Financed by Token Security, which sells agent identity security.
So price the running side before you price the build. Your team can build the app. The question is whether you can run twenty of them next year without hiring.
The vendor supplied all four. Rebuild the app and all four become yours, which is why this decision sits with IT and not with the department holding the renewal date.
Say yes once, not app by app
Granting a hosting exception every time doesn’t scale. What works is one path to production that every app takes, with those four things attached on the way out instead of bolted on when someone asks for an audit.
That’s what Tray Helix is. Your teams keep building in Claude Code, Codex or Cursor. One command puts the app on a managed runtime. Every deployed app sits behind sign-in, with your SSO on Enterprise Edition, and access is granted per project. The app never holds the credential: it names the system it needs, and Tray resolves the authentication server side, so there’s nothing hardcoded to leak or rotate. Projects, deployments, access changes and execution logs are recorded for each app, with a named owner on every one.
Helix runs on the Tray AI Orchestration Platform, the same runtime behind DocuSign, Cisco and GitHub, handling more than a trillion processes a year. Those are platform numbers, not Helix ones. They are the reason the operating side can be somebody’s day job rather than yours.
Where to start
Not with a portfolio review. Pick one subscription that renews in the next twelve months, is owned by one department, and is used only by your own employees. Rebuild that one, and let it set your numbers.
The scorecard for deciding which apps qualify, including the ones that look easy and aren’t, is in our free guide, Which SaaS tools should you rebuild with AI? If you’d rather see the money first, the calculator sizes it against your headcount and the categories you pay for.
Footnotes
-
Tray’s own cost model, as published in the FAQ on the SaaS savings calculator. It is an estimate for discussion, not a quote. Back
