There's a meme in operations circles that goes something like: “Don't pave the cowpath.” The point being that bad processes don't get better by being automated. It's true, narrowly. It's also been weaponized by every SaaS vendor in the world to justify making you change the way you run your business so their product fits. The right rule is harder to fit on a coffee mug: build software around your good process, fix your bad process before you do, and don't confuse the two.
The decision tree
When we sit down with a new client, we work through four questions in order. The answers determine whether they should buy, build, or do nothing yet.
1 · Is the process load-bearing or incidental?
Load-bearing means: this is how you make money. If a process is one of three or four steps that turn an inbound lead into delivered work and paid invoices, it is load-bearing. Forcing a load-bearing process into a generic SaaS shape is how you bleed margin for ten years without noticing. Incidental processes — expense reports, holiday calendars, the company directory — should run on whatever cheap SaaS works.
2 · Is the process the same as other people in your industry?
If you sell the same widget the same way as 10,000 other companies, a vertical SaaS for your industry probably exists and probably fits. Use it. If your process is shaped by something unusual — a custom intake, a multi-discipline approval, a hand-built quoting model, a niche delivery method — then no off-the-shelf tool was designed with you in mind. They will all almost-fit.
3 · How often does the process change?
Stable processes can be encoded. Processes still in flux should be left in spreadsheets and Slack until they settle. The worst time to commission custom software is six months into a new operation. The right time is when you've been doing the same thing the same way for at least a year and you're still annoyed by the tooling.
4 · What's the cost of working around the tool?
This one is uncomfortable. Add up the workarounds: spreadsheets that shadow the CRM, the second tool you bought because the first wouldn't do this, the manual export-and-reformat that someone does every Friday. Multiply by the hourly cost of the people doing it. Annualize. We have clients for whom that number is over $200,000 a year. The custom build is suddenly cheap.
Where buying is the right call
The buy-don't-build cases are real. Accounting. Payroll. Email marketing. Calendaring. Customer support ticketing for a generic SaaS-shaped business. These are markets where the SaaS has been refined for two decades and the wins from customizing them are small. Use a great tool, integrate via API, move on.
Where building wins
Where SMBs win with custom software is almost always at the seams. The integration between intake and quoting. The handoff between sales and operations. The custom shape of a deal in a niche industry. The way an AI agent should read your incoming work and route it. These are places where a 10-page SaaS feature list will give you nine pages of stuff you don't need and miss the one thing you do.
SaaS is a great fit when your process matches the industry average. Custom is a great fit when it doesn't.
The hybrid that usually wins
Most of the operations we build end up hybrid. A custom CRM-and-operations spine in CatalystOS, integrated via API with QuickBooks for accounting, Gmail for inbox, a payment processor for billing, and whatever vertical tools the team genuinely loves. We don't replace the things that already work. We replace the things that almost-work, and we wire the rest together.
The architecture is simple and stays simple: one system of record (CatalystOS) plus a small set of integrated specialists. No tool sprawl. No shadow spreadsheets. The team uses one screen for the core work and external tools only where it makes sense.
The honest answer
Build software around your process is not a slogan for everything custom; it's an argument for being honest about which processes are yours and which are generic. The generic ones, buy. The ones that define how you operate, build. Don't pave the cowpath; but don't replace the cowpath with a six-lane highway in the wrong direction either.
