Businesses now have a genuinely realistic option that didn’t exist three years ago: replace the SaaS tools they rent with custom software they own.
If you work inside a company of almost any size, you know the daily texture of SaaS dependence. Every process starts with “can I do this with our CRM?” Every report starts with “can I get access to that data?” Every improvement idea dies with “it would be so much easier if we could just change how this works,” followed by the quiet acknowledgment that you can’t, because the tool belongs to someone else and your workflow lives inside their product decisions.
That resignation made sense when custom software was expensive. It’s stopped making sense, because AI has collapsed the cost of building, and the math that justified renting everything has flipped for a large class of tools.
The Numbers Behind SaaS Dependence Are Genuinely Crazy
The scale of what businesses hand over to SaaS vendors is worth sitting with for a minute. From Zylo’s 2026 SaaS Management Index and the industry data around it:
- The average organization now runs roughly 300 SaaS applications. Small companies average around 150, large enterprises over 600.
- Enterprise SaaS spend averages $55.7 million per year, with a median around $20 million, which works out to roughly $4,800 per employee per year in software rent.
- Around half of all purchased licenses sit idle, wasting an average of nearly $20 million a year per organization, and that waste keeps growing double digits year over year.
Read those numbers again as what they actually are: businesses paying tens of millions annually, in perpetuity, for tools that mostly don’t fit, that half the seats never touch, and that they’ll never own a line of. Renegotiating that stack is treated as a procurement problem. It’s actually a build-versus-rent problem, and the build side of the equation just changed.
What AI Actually Changed
The reason custom software lost to SaaS for fifteen years was simple: development was slow and expensive, and a subscription was faster than a project. AI-assisted development attacks exactly that constraint. The mechanical majority of programming (CRUD screens, API plumbing, form handling, report generation, the plumbing that made projects take months) is now dramatically faster, and I say that as a practitioner, not a spectator: I’ve written about how I get production-quality code out of Claude and ChatGPT, and AI acceleration is baked into how I build for clients every week.
To be precise about the claim: AI doesn’t remove the need for experienced engineering judgment. Architecture, data modeling, security and the discipline to keep systems simple still decide whether custom software becomes an asset or a liability. What AI removes is the labor cost that made all of that judgment too expensive to apply to internal tools. A system that would have been a $150K, six-month project in 2021 is now frequently a few weeks of focused work. At that price, the comparison against a $30K-per-year subscription you’ll pay forever stops being close.
What Owning Your Software Buys You
It does exactly what you want. Not 80% of what you want with workarounds for the rest. The tool encodes your process, instead of your process contorting to fit the tool. Every “it would be easier if we could just change…” becomes a change you simply make.
Your data is yours, all of it, right now. No export limits, no API rate tiers, no “that field isn’t available in your plan,” no begging a vendor for access to information about your own customers. Every integration question becomes a query against a database you control.
The rent stops. A subscription is a perpetual claim on your revenue that compounds with headcount. Custom software is a capital investment that costs maintenance, which for well-built internal tools is modest. Per-seat pricing for your own software is zero.
Nobody can take it away from you. No surprise 40% renewal increase, no acquisition that kills the product, no feature you depend on moving to the enterprise tier, no roadmap that ignores you. SaaS puts your operations at the mercy of someone else’s business model, and everyone who has been through a forced migration knows what that costs.
Where This Works, and Where It Doesn’t
Being honest about the boundaries, because “cancel all your SaaS” is not the argument.
Keep renting the things that are genuinely hard, genuinely generic and carry real liability: payment processing, payroll, email deliverability infrastructure, accounting. Nobody should rebuild Stripe.
The replacement targets are the middle of your stack, and once you look for them they’re everywhere: the CRM you use 20% of, the project tracker your team fights, the reporting tool that exists because your other tools won’t share data, the form and workflow products, the “platform” whose real job is moving rows between two systems you already own. These are ordinary databases with ordinary interfaces, priced like magic. They’re also exactly the tools where the fit problem hurts most, because they sit closest to how your business actually runs.
I’ve been living this argument for years, before AI made it easy. I run my own invoicing system, built custom, because every invoicing SaaS wanted a monthly fee for what is fundamentally a database table and a PDF. I’ve shown clients how self-hosted Mautic replaces HubSpot at a fraction of the cost, and how self-hosted consent management replaces another subscription line item. Open source plus a modest amount of custom glue was already beating SaaS on cost. AI just extended that same economics to fully bespoke systems.
How to Start
Don’t start with a moonshot. Start with an audit and a small win:
- Pull the SaaS spend report. Every line item, every renewal date, every seat count against actual usage. Half the value of this exercise is that nobody has ever looked.
- Find the resentment. The tool people complain about, work around, or maintain spreadsheets alongside is your first candidate. Spreadsheet shadow systems are the surest sign a tool doesn’t fit.
- Replace one mid-tier tool end to end. Something with real annual cost but bounded scope. Ship it, run it for a quarter, and let the renewal lapse.
- Bank the rent and repeat. The canceled subscription funds the next build, and each replacement compounds: your systems start talking to each other because they share your data instead of guarding it behind separate vendors’ APIs.
The businesses that internalize this early get a durable advantage: software shaped exactly to how they operate, data with no walls around it, and an operating budget that isn’t leaking $4,800 per employee per year to vendors whose product decisions they don’t control. The ones that don’t will keep asking permission to use their own information.
This build-over-rent shift is one of the most valuable conversations I have with clients right now, whether it starts as a digital growth engagement or a custom development project. The question worth asking about every renewal that crosses your desk this year is the one SaaS vendors hope you never ask: what would it cost us to just build this?