Custom Software vs. SaaS for a Small Business: How to Decide in 2026

For twenty years the advice to a small business was simple: never build software, buy it. A subscription cost less than a developer's first month, the vendor handled the servers, and the product improved without you doing anything. That advice was right, and the reason it was right has quietly gone away. Building software got cheap. Not cheaper; cheap. So the question a small business faces now is not "can we afford custom" but "which of the two is actually less expensive for this process", and the answer is different for every process.

What SaaS is good at

A subscription product is the right answer whenever your process is the same as everyone else's. Payroll, accounting, email, file storage, video calls, a basic point of sale: the problem is standard, the product is mature, and the last thing you want is to be the only company running its own version. Buy it, connect it, forget it. The same goes for anything regulated where the vendor carries the compliance burden for you.

SaaS is also right when you do not yet know what you need. A cheap monthly tool is the fastest way to find out what your process really is, and the spreadsheet you build alongside it tells you what the tool is missing. Plenty of good custom systems start as a SaaS subscription plus a spreadsheet of workarounds.

Where SaaS starts to cost more than it saves

• Per-seat pricing at scale. A tool that was $40 a month for two people is $2,400 a month for forty, forever, and the price goes up every renewal. Over five years that is a six-figure sum for software you do not own.
• The workaround tax. The product does 80% of your process, and the other 20% lives in a spreadsheet, an export, or a person re-keying between two tools. That person is the integration, and their time is the real subscription cost.
• Three tools where one process lives. A CRM, a project tool and a billing tool that each hold a third of the job, none of which knows what the others know. Nobody has a full picture, so someone builds a fourth thing to reconcile them.
• The feature you cannot have. The one thing your business does differently is exactly the thing the vendor's roadmap will never prioritise, because you are one customer of ten thousand.
• Data you cannot get out. When you finally want to move, the export is a CSV with half the fields, and the history is gone.

Any of those is a signal. Two or three together mean the subscription is no longer the cheap option; it just still looks like it on the invoice.

What changed about custom

The old objection to custom software was never that it was worse. It was the cost and the wait: a quote in the tens of thousands, a project measured in months, and an agency that disappeared when the invoice cleared, leaving code nobody could maintain. Every part of that has shifted. Building with agentic AI tools under senior engineers, a working version of one process takes days. The price is fixed before it starts. And the code is written to be handed over, with its specs and history, to whichever team runs it next.

What has not changed is that custom software is only worth it for the process that is yours. Nobody should build their own payroll. Everybody should look hard at the process that makes their business different and ask why it is running inside somebody else's product.

A way to decide, process by process

Do not decide for the business; decide for each process. Take the one you are looking at and ask four questions.

• Is this process the same as every competitor's? If yes, buy. If it is the thing you win on, keep reading.
• How much of it does the product actually cover? Above 90% with no spreadsheet beside it: buy. Below that, count the hours the gap costs every week and multiply by fifty.
• What will the subscription cost over five years at your expected headcount? Compare it with a fixed-price build plus modest support. The build usually wins somewhere between year one and year two.
• Does it need to connect to the rest of your systems? Custom software connects to whatever you already run; a SaaS product connects to whatever its vendor decided to support.

Two examples

A twelve-person marketing agency uses a subscription project tool, a subscription CRM and a subscription invoicing app. Each works. None knows the others exist, so a project manager spends Friday afternoon reconciling hours to invoices in a spreadsheet, and the owner has no idea which clients are profitable. The fix is not a fourth subscription. It is a small internal app that reads all three and shows margin by client, built once, owned outright.

A regional distributor runs quoting inside a CRM that was never designed for it. Quotes take forty minutes, the price book is a spreadsheet, and a third go out with an error. A configure-price-quote platform would fix it for a five-figure licence and a six-month rollout. A quoting engine built around their real price book and stock takes five days, and the CRM stays exactly where it is.

The honest cost comparison

A mid-tier SaaS tool for a twenty-person team runs $500 to $2,000 a month depending on the category, rising with seats and tiers, and typically two or three such tools overlap on one process. A scoped custom build of that same process is a fixed price agreed up front, with support from $999 a month if you want it and nothing if you do not. The subscription is cheaper in month one. The build is cheaper by the end of year two, and it is the only one of the two you will own. The fuller version of these numbers is in how much AI consulting costs.

What to do next

Keep every subscription that covers a standard process well. Then list the processes that run partly in a tool and partly in a spreadsheet, and pick the one that costs the most hours or loses the most money. That is the one to build. If you would rather recognise it than describe it, the Rebuild Library has an entry for most of them. And if you want a second opinion on whether it is worth building at all, that is what the free AI assessment is for.