Start with off-the-shelf tools. Build custom only when a tool cannot do the job, and you can name the job it cannot do.
Should I build custom AI or use off-the-shelf tools?
How the options compare:
- The AI already inside your software. Email, spreadsheets, accounting, the CRM. Cost: usually nothing extra, or a small add-on. Fits: most of the first year. I switch it on with your team in a 90-minute session for $450.
- Off-the-shelf AI tools. Chat assistants, meeting notes, writing tools, industry tools. Cost: a monthly subscription per person. Fits: common jobs that thousands of businesses share.
- Connected tools. Off-the-shelf tools tied together with an automation platform. Cost: a build plus a monthly platform bill. Fits: work that moves between two or three systems.
- Custom AI. Software built for your process and your data. Cost: a fixed-price build plus running costs. Fits: work no tool does, or work that is the product you sell.
When is off-the-shelf enough?
When the job is common. Writing a first draft, summarizing a meeting, answering a routine customer email, cleaning up a spreadsheet. Somebody has already built a tool for that, and it gets better every month without you paying for the improvements.
The usual problem with off-the-shelf tools is that nobody uses them. If that is where you are, we bought software nobody uses is the page for you. Building something custom will not fix a habit problem.
When does custom AI pay for itself?
Three cases.
- The work is your product. At Appraisal Dream, where I hold the fractional Chief AI Officer seat, I built their AI software, because the software is what they sell. More than 50 appraisers are on the waitlist for it. No tool on the shelf did that job.
- Your rules are unusual. Regulated work, odd documents, or a process with steps no vendor planned for.
- The tool bill outgrew the build. When per-person subscriptions across a whole company cost more each year than building and running the thing yourself.
What does custom cost compared to a subscription?
A subscription is a small bill forever. A build is a bigger bill once, plus a smaller running bill. What an AI automation project costs breaks down both. Do the math over three years, and count the time somebody on staff spends looking after the custom one.
What should I not do?
Do not build custom because a vendor demo looked expensive. Do not build custom to avoid learning a tool. And do not let anyone build it on accounts in their name. You own the code, the accounts, and the data, or you do not own it.
The AI build page says what I build and what I tell people to buy instead. If you have a choice in front of you right now, describe it on the start form.
Answered by Jared Lukes, Minneapolis. Twenty-five years in brand, UX, and technology, currently holding fractional AI and technology seats at three companies. About Jared.