This is one of the first questions I try to answer straight — including the times the right answer costs me a project.
The check I run before proposing anything custom
Before I ever price a custom build, I look hard at what the client already pays for. This has caught real, specific examples: a client whose booking software already had native lapsed-customer alerts built in, another whose design software had a free image-capture extension sitting unused, a project management tool with a smart-status feature nobody had turned on. In every one of those cases, proposing a custom build would have meant charging for something the client could already get for a few dollars a month — or nothing.
When buying is genuinely the right answer
If a feature exists, works, and covers your actual need — even if it's not perfectly tailored — buy it. The math almost never favors building something custom to replace a $20/month feature that already does 90% of what you need. This is true even when the tier you'd need to pay for feels annoying. Annoying is still cheaper than custom, most of the time.
When building is genuinely the right answer
Custom makes sense when the problem is specific to your operation in a way no vendor building for a broad customer base would ever solve — coordination across your specific locations, a workflow unique to how your team actually works, or a process that touches multiple tools in a way no single platform's AI feature will ever bridge. If you've genuinely checked and nothing off-the-shelf fits, that's the signal to build.
The market rate, for reference
Once you're in build territory, a real custom multi-agent system from a solo consultant or small boutique typically runs somewhere between $5,000 and $25,000, plus $1,000-$3,000 a month ongoing — meaningfully less than what a larger firm or in-house team would cost for equivalent work, because a senior solo consultant delivers the same implementation work as a much bigger team at a fraction of the price, for businesses under roughly $10M in revenue.
The bottom line
I'd rather tell a client "you already have this" and lose a build than sell something they didn't need. That's not altruism — it's the only way this kind of relationship holds up past the first project.
Before you commit to building anything custom, it's worth a real look at what your current tools already do. That check is built into every AI Opportunity Assessment.