Build or Buy: The AI Decision Most Owners Get Backward
Off-the-shelf AI runs somewhere between 50 and 500 dollars per user per month. A custom build for a single, focused job runs from a few $1,000 to the mid tens of thousands, one time, plus maintenance every year after. Those two numbers are the whole decision, and most owners read them backward.
The reason is in how the two choices sound. Somebody proposes building something custom, and custom carries weight: tailored, proprietary, yours. Buying off the shelf sounds like settling. So the instinct, especially for an owner who takes the decision seriously, leans toward building. For most companies in the lower middle market that instinct points the wrong way. The serious choice is usually to buy, and knowing why is worth more than any tool I could name for you.
What the two paths actually cost
The subscription cost is ongoing. It grows with headcount and never fully goes away. That's the honest knock on buying: you rent forever, and the meter runs faster as you grow.
Custom flips the shape of the cost. Industry estimates for 2025 and 2026 put a focused, single-purpose build anywhere from a few thousand to the mid tens of thousands of dollars, with bigger multi-system automation running higher and a full platform climbing well into six figures. That's the one-time figure. The whole figure is larger, because custom systems carry maintenance, commonly put at several thousand to $20,000 a year, plus hosting and usage on top. That maintenance is mandatory. The underlying models change, the connections break, and a custom system nobody tends stops being an asset and starts being a problem that sits there costing you.
Run the arithmetic and the crossover is further out than the pitch implies. One comparison found that at 50 dollars per user per month, a 25,000 dollar custom build only barely beats the subscription over 3 years for a team of twenty, and that's before maintenance, project risk, and the months you spent waiting for it to get built.
Early in a lot of these conversations, I'll watch a capable owner, someone running a fifteen-million-dollar business, get ready to spend six figures building a tool that does a job you can subscribe to on a Monday afternoon. Good businesspeople. They heard "custom" and stopped doing the math.
The 80 percent rule
Cost is half the story. The other half is that the off-the-shelf tools are good, and they cover far more of your real needs than a custom build would improve on. A fair estimate is that ready-made AI tools handle around 80 percent of common small-business jobs with no custom development at all: drafting, summarizing, customer service triage, data analysis, marketing copy, meeting notes. Solved problems, and the companies selling those solutions have poured more into them than you could ever justify yourself. Build custom to do a job an existing tool already does well and you haven't bought an edge. You've paid a premium to reach a result you could have rented, and you've taken on the odds that your version comes out worse.
The question that decides it is how specific to you the work is. Importance has nothing to do with it. If thousands of other businesses need the same job done, a vendor already built the better version and spread the cost across all of them. If the job is genuinely peculiar to how your shop runs, buried in your data and your process, no vendor is serving it, and custom starts to earn its keep.
When building is the right call
The real reasons to build share a trait: the advantage compounds. A custom system built around your own accumulated data can get sharper as you feed it more, which a generic subscription won't do for you specifically. If a process is both central to how you make money and unlike anyone else's, owning it can beat renting a generic approximation. And at real scale the math changes, because a workflow run constantly across a large team can make the one-time build cheaper than the endless per-seat rent.
Every one of those cases is specific, central, and yours, which is a narrow set. Most of what a company wants from AI is none of the three, and most of it should be bought.
So treat the urge to build with suspicion, precisely because it flatters you. It feels like the committed choice, the move of someone who takes the work seriously. Taking it seriously usually means refusing to rebuild what already exists, spending the subscription money without embarrassment, and saving a custom build for the one or two places where what you do is truly unlike anything for sale. Build there. Buy the rest, and put the money you saved into the parts of the business a subscription will never reach.